{ "passed": true, "started": "20261026T232618Z", "scope": "Actual HTTPS browser, typed synthetic turns, real local/API/TTS. No physical microphone or acoustic-stop assertion.", "source_sha256": "da1dd8cec48714dabeb3d056fec603eed2f49370e04ea481f1fd03f5dc158ddb", "server": { "ready": true, "phase": "Ready: Small RL + Nemotron 3.5 + TYGER", "source_sha256": "da1dd8cec48714dabeb3d056fec603eed2f49370e04ea481f1fd03f5dc158ddb", "speech_pause_default_ms": 700.0, "sentence_stream": true, "decision_interval_ms": 320, "identity_models": { "fast": { "frames": 4, "mix_layers": true, "label": "V4 layer mix · 320 ms · seed 2023" }, "long": { "frames": 25, "mix_layers": false, "label": "V5 optimized · 2000 ms · utterance" }, "long3": { "frames": 38, "mix_layers": false, "label": "V5 optimized · 3040 ms · utterance" } }, "asr_model_selection": true, "tts": "Humaneness Voice Small 1.1 Alpha RL r128 v1", "llm_default": { "model": "qwen", "provider": "cerebras/fp16" } }, "turns": [ { "input": "Heute war ein wirklich anstrengender Tag. Erzähl mir etwas Aufmunterndes.", "enabled": true, "events": [ { "type": "latency_start", "reply_id": 1, "input": { "kind": "typed" }, "at": 1848.1999969482422 }, { "type": "reply", "text": "Das klingt anstrengend.", "reply_id": 1, "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das klingt anstrengend.", "start_s": 0, "end_s": 1.02 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 0, "sentence_text": "Das klingt anstrengend.", "sentence_role": "reaction", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "selected", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "sentence_ready_ms": 333.74, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "contentment.soft" }, "sources": [], "lookups": [], "at": 2181.699996948242 }, { "type": "latency_update", "reply_id": 1, "metrics": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16 }, "at": 2738.7999992370605 }, { "type": "reply", "text": "Das klingt anstrengend. Oh, das tut mir leid, dass du so müde bist!", "reply_id": 1, "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das klingt anstrengend.", "start_s": 0, "end_s": 2.56 }, { "emotion": "interest", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "interest, warm", "text": "Oh, das tut mir leid, dass du so müde bist!", "start_s": 2.56, "end_s": 5.96 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 1, "sentence_text": "Oh, das tut mir leid, dass du so müde bist!", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 } ], "sentence_ready_ms": 2374.86, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "interest.soft" }, "sources": [], "lookups": [], "at": 4227.599998474121 }, { "type": "latency_update", "reply_id": 1, "metrics": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 } ], "answer_first_audio_ms": 3506.07 }, "at": 5365.199996948242 }, { "type": "reply", "text": "Das klingt anstrengend. Oh, das tut mir leid, dass du so müde bist!\nStell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.", "reply_id": 1, "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das klingt anstrengend.", "start_s": 0, "end_s": 2.56 }, { "emotion": "interest", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "interest, warm", "text": "Oh, das tut mir leid, dass du so müde bist!", "start_s": 2.56, "end_s": 6.27 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.72, "direction": "playful, rapid", "text": "Stell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.", "start_s": 6.27, "end_s": 11.37 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 2, "sentence_text": "Stell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 } ], "answer_first_audio_ms": 3506.07, "sentence_ready_ms": 5018.82, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "delivery.playful.intense" }, "sources": [], "lookups": [], "at": 6866.599998474121 }, { "type": "latency_update", "reply_id": 1, "metrics": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true }, "at": 8053.799999237061 }, { "type": "reply", "text": "Das klingt anstrengend. Oh, das tut mir leid, dass du so müde bist!\nStell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.\nEr hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.", "reply_id": 1, "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das klingt anstrengend.", "start_s": 0, "end_s": 2.56 }, { "emotion": "interest", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "interest, warm", "text": "Oh, das tut mir leid, dass du so müde bist!", "start_s": 2.56, "end_s": 6.27 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.72, "direction": "playful, rapid", "text": "Stell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.", "start_s": 6.27, "end_s": 11.9 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "amusement, soft", "text": "Er hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.", "start_s": 11.9, "end_s": 17.34 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 3, "sentence_text": "Er hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true, "sentence_ready_ms": 8637.05, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "amusement.soft" }, "sources": [], "lookups": [], "at": 10498.099998474121 }, { "type": "latency_update", "reply_id": 1, "metrics": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true }, "at": 11698.899997711182 }, { "type": "reply", "text": "Das klingt anstrengend. Oh, das tut mir leid, dass du so müde bist!\nStell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.\nEr hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.\nUnd weißt du was?", "reply_id": 1, "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das klingt anstrengend.", "start_s": 0, "end_s": 2.56 }, { "emotion": "interest", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "interest, warm", "text": "Oh, das tut mir leid, dass du so müde bist!", "start_s": 2.56, "end_s": 6.27 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.72, "direction": "playful, rapid", "text": "Stell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.", "start_s": 6.27, "end_s": 11.9 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "amusement, soft", "text": "Er hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.", "start_s": 11.9, "end_s": 18.3 }, { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "contentment, warm", "text": "Und weißt du was?", "start_s": 18.3, "end_s": 19.66 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 4, "sentence_text": "Und weißt du was?", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3745, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3924, "performance": { "device_seconds": 2.9482006225585935, "raw_audio_seconds": 6.4, "aggregate_rtf": 0.4606563472747802, "audio_seconds_per_second": 2.170815632772564, "batches": [ { "batch_size": 1, "device_seconds": 2.9482006225585935, "audio_seconds": [ 6.4 ], "model_wall_seconds": 3.5646182149648666, "first_audio_seconds": 0.7783311157226562, "graph_capture_seconds": 0, "live_callback_seconds": 0.6068725809454918 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 }, { "sentence": 3, "total_ms": 3938.22, "audio_s": 6.4 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true, "sentence_ready_ms": 12576.35, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "contentment.soft" }, "sources": [], "lookups": [], "at": 14430.79999923706 }, { "type": "latency_update", "reply_id": 1, "metrics": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3745, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3924, "performance": { "device_seconds": 2.9482006225585935, "raw_audio_seconds": 6.4, "aggregate_rtf": 0.4606563472747802, "audio_seconds_per_second": 2.170815632772564, "batches": [ { "batch_size": 1, "device_seconds": 2.9482006225585935, "audio_seconds": [ 6.4 ], "model_wall_seconds": 3.5646182149648666, "first_audio_seconds": 0.7783311157226562, "graph_capture_seconds": 0, "live_callback_seconds": 0.6068725809454918 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 }, { "sentence": 3, "total_ms": 3938.22, "audio_s": 6.4 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true }, "at": 15564.29999923706 }, { "type": "reply", "text": "Das klingt anstrengend. Oh, das tut mir leid, dass du so müde bist!\nStell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.\nEr hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.\nUnd weißt du was? Er hat darin das beste Roboter-Omelett der Welt gebraten!", "reply_id": 1, "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das klingt anstrengend.", "start_s": 0, "end_s": 2.56 }, { "emotion": "interest", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "interest, warm", "text": "Oh, das tut mir leid, dass du so müde bist!", "start_s": 2.56, "end_s": 6.27 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.72, "direction": "playful, rapid", "text": "Stell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.", "start_s": 6.27, "end_s": 11.9 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "amusement, soft", "text": "Er hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.", "start_s": 11.9, "end_s": 18.3 }, { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "contentment, warm", "text": "Und weißt du was?", "start_s": 18.3, "end_s": 20.25 }, { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "contentment, warm", "text": "Er hat darin das beste Roboter-Omelett der Welt gebraten!", "start_s": 20.25, "end_s": 23.31 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 5, "sentence_text": "Er hat darin das beste Roboter-Omelett der Welt gebraten!", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3745, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3924, "performance": { "device_seconds": 2.9482006225585935, "raw_audio_seconds": 6.4, "aggregate_rtf": 0.4606563472747802, "audio_seconds_per_second": 2.170815632772564, "batches": [ { "batch_size": 1, "device_seconds": 2.9482006225585935, "audio_seconds": [ 6.4 ], "model_wall_seconds": 3.5646182149648666, "first_audio_seconds": 0.7783311157226562, "graph_capture_seconds": 0, "live_callback_seconds": 0.6068725809454918 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 1644, "alignment_ms": 52, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1826, "performance": { "device_seconds": 1.3384681282043458, "raw_audio_seconds": 2.88, "aggregate_rtf": 0.46474587784873117, "audio_seconds_per_second": 2.1517135442468347, "batches": [ { "batch_size": 1, "device_seconds": 1.3384681282043458, "audio_seconds": [ 2.88 ], "model_wall_seconds": 1.4674943834543228, "first_audio_seconds": 0.7459358520507813, "graph_capture_seconds": 0, "live_callback_seconds": 0.1282988339662552 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 }, { "sentence": 3, "total_ms": 3938.22, "audio_s": 6.4 }, { "sentence": 4, "total_ms": 1834.77, "audio_s": 1.95 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true, "sentence_ready_ms": 14412.35, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "contentment.soft" }, "sources": [], "lookups": [], "at": 16260.099998474121 }, { "type": "latency_update", "reply_id": 1, "metrics": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3745, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3924, "performance": { "device_seconds": 2.9482006225585935, "raw_audio_seconds": 6.4, "aggregate_rtf": 0.4606563472747802, "audio_seconds_per_second": 2.170815632772564, "batches": [ { "batch_size": 1, "device_seconds": 2.9482006225585935, "audio_seconds": [ 6.4 ], "model_wall_seconds": 3.5646182149648666, "first_audio_seconds": 0.7783311157226562, "graph_capture_seconds": 0, "live_callback_seconds": 0.6068725809454918 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 1644, "alignment_ms": 52, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1826, "performance": { "device_seconds": 1.3384681282043458, "raw_audio_seconds": 2.88, "aggregate_rtf": 0.46474587784873117, "audio_seconds_per_second": 2.1517135442468347, "batches": [ { "batch_size": 1, "device_seconds": 1.3384681282043458, "audio_seconds": [ 2.88 ], "model_wall_seconds": 1.4674943834543228, "first_audio_seconds": 0.7459358520507813, "graph_capture_seconds": 0, "live_callback_seconds": 0.1282988339662552 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 }, { "sentence": 3, "total_ms": 3938.22, "audio_s": 6.4 }, { "sentence": 4, "total_ms": 1834.77, "audio_s": 1.95 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true }, "at": 17481.09999847412 }, { "type": "reply", "text": "Das klingt anstrengend. Oh, das tut mir leid, dass du so müde bist!\nStell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.\nEr hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.\nUnd weißt du was? Er hat darin das beste Roboter-Omelett der Welt gebraten!\nAber bevor wir weiterträumen: Triffst du mich gleich, oder hast du lieber ein paar Minuten zum Durchatmen?", "reply_id": 1, "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das klingt anstrengend.", "start_s": 0, "end_s": 2.56 }, { "emotion": "interest", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "interest, warm", "text": "Oh, das tut mir leid, dass du so müde bist!", "start_s": 2.56, "end_s": 6.27 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.72, "direction": "playful, rapid", "text": "Stell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.", "start_s": 6.27, "end_s": 11.9 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "amusement, soft", "text": "Er hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.", "start_s": 11.9, "end_s": 18.3 }, { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "contentment, warm", "text": "Und weißt du was?", "start_s": 18.3, "end_s": 20.25 }, { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "contentment, warm", "text": "Er hat darin das beste Roboter-Omelett der Welt gebraten!", "start_s": 20.25, "end_s": 23.88 }, { "emotion": "affection", "secondary": "contentment", "blend": 0.25, "intensity": 0.72, "direction": "affection, natural", "text": "Aber bevor wir weiterträumen: Triffst du mich gleich, oder hast du lieber ein paar Minuten zum Durchatmen?", "start_s": 23.88, "end_s": 29.66 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 6, "sentence_text": "Aber bevor wir weiterträumen: Triffst du mich gleich, oder hast du lieber ein paar Minuten zum Durchatmen?", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3745, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3924, "performance": { "device_seconds": 2.9482006225585935, "raw_audio_seconds": 6.4, "aggregate_rtf": 0.4606563472747802, "audio_seconds_per_second": 2.170815632772564, "batches": [ { "batch_size": 1, "device_seconds": 2.9482006225585935, "audio_seconds": [ 6.4 ], "model_wall_seconds": 3.5646182149648666, "first_audio_seconds": 0.7783311157226562, "graph_capture_seconds": 0, "live_callback_seconds": 0.6068725809454918 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 1644, "alignment_ms": 52, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1826, "performance": { "device_seconds": 1.3384681282043458, "raw_audio_seconds": 2.88, "aggregate_rtf": 0.46474587784873117, "audio_seconds_per_second": 2.1517135442468347, "batches": [ { "batch_size": 1, "device_seconds": 1.3384681282043458, "audio_seconds": [ 2.88 ], "model_wall_seconds": 1.4674943834543228, "first_audio_seconds": 0.7459358520507813, "graph_capture_seconds": 0, "live_callback_seconds": 0.1282988339662552 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2280, "alignment_ms": 38, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2445, "performance": { "device_seconds": 1.844489884853363, "raw_audio_seconds": 3.76, "aggregate_rtf": 0.49055582043972423, "audio_seconds_per_second": 2.03850399553637, "batches": [ { "batch_size": 1, "device_seconds": 1.844489884853363, "audio_seconds": [ 3.76 ], "model_wall_seconds": 2.1057667583227158, "first_audio_seconds": 0.787409912109375, "graph_capture_seconds": 0, "live_callback_seconds": 0.2597941681742668 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 }, { "sentence": 3, "total_ms": 3938.22, "audio_s": 6.4 }, { "sentence": 4, "total_ms": 1834.77, "audio_s": 1.95 }, { "sentence": 5, "total_ms": 2458.68, "audio_s": 3.63 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true, "sentence_ready_ms": 16872.18, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "affection.soft" }, "sources": [], "lookups": [], "at": 18726.89999771118 }, { "type": "latency_update", "reply_id": 1, "metrics": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3745, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3924, "performance": { "device_seconds": 2.9482006225585935, "raw_audio_seconds": 6.4, "aggregate_rtf": 0.4606563472747802, "audio_seconds_per_second": 2.170815632772564, "batches": [ { "batch_size": 1, "device_seconds": 2.9482006225585935, "audio_seconds": [ 6.4 ], "model_wall_seconds": 3.5646182149648666, "first_audio_seconds": 0.7783311157226562, "graph_capture_seconds": 0, "live_callback_seconds": 0.6068725809454918 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 1644, "alignment_ms": 52, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1826, "performance": { "device_seconds": 1.3384681282043458, "raw_audio_seconds": 2.88, "aggregate_rtf": 0.46474587784873117, "audio_seconds_per_second": 2.1517135442468347, "batches": [ { "batch_size": 1, "device_seconds": 1.3384681282043458, "audio_seconds": [ 2.88 ], "model_wall_seconds": 1.4674943834543228, "first_audio_seconds": 0.7459358520507813, "graph_capture_seconds": 0, "live_callback_seconds": 0.1282988339662552 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2280, "alignment_ms": 38, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2445, "performance": { "device_seconds": 1.844489884853363, "raw_audio_seconds": 3.76, "aggregate_rtf": 0.49055582043972423, "audio_seconds_per_second": 2.03850399553637, "batches": [ { "batch_size": 1, "device_seconds": 1.844489884853363, "audio_seconds": [ 3.76 ], "model_wall_seconds": 2.1057667583227158, "first_audio_seconds": 0.787409912109375, "graph_capture_seconds": 0, "live_callback_seconds": 0.2597941681742668 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 }, { "sentence": 3, "total_ms": 3938.22, "audio_s": 6.4 }, { "sentence": 4, "total_ms": 1834.77, "audio_s": 1.95 }, { "sentence": 5, "total_ms": 2458.68, "audio_s": 3.63 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true }, "at": 19931.89999771118 }, { "type": "pipeline_metrics", "reply_id": 1, "metrics": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3745, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3924, "performance": { "device_seconds": 2.9482006225585935, "raw_audio_seconds": 6.4, "aggregate_rtf": 0.4606563472747802, "audio_seconds_per_second": 2.170815632772564, "batches": [ { "batch_size": 1, "device_seconds": 2.9482006225585935, "audio_seconds": [ 6.4 ], "model_wall_seconds": 3.5646182149648666, "first_audio_seconds": 0.7783311157226562, "graph_capture_seconds": 0, "live_callback_seconds": 0.6068725809454918 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 1644, "alignment_ms": 52, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1826, "performance": { "device_seconds": 1.3384681282043458, "raw_audio_seconds": 2.88, "aggregate_rtf": 0.46474587784873117, "audio_seconds_per_second": 2.1517135442468347, "batches": [ { "batch_size": 1, "device_seconds": 1.3384681282043458, "audio_seconds": [ 2.88 ], "model_wall_seconds": 1.4674943834543228, "first_audio_seconds": 0.7459358520507813, "graph_capture_seconds": 0, "live_callback_seconds": 0.1282988339662552 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2280, "alignment_ms": 38, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2445, "performance": { "device_seconds": 1.844489884853363, "raw_audio_seconds": 3.76, "aggregate_rtf": 0.49055582043972423, "audio_seconds_per_second": 2.03850399553637, "batches": [ { "batch_size": 1, "device_seconds": 1.844489884853363, "audio_seconds": [ 3.76 ], "model_wall_seconds": 2.1057667583227158, "first_audio_seconds": 0.787409912109375, "graph_capture_seconds": 0, "live_callback_seconds": 0.2597941681742668 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 4037, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4221, "performance": { "device_seconds": 3.1560501308441165, "raw_audio_seconds": 6.72, "aggregate_rtf": 0.4696503170898983, "audio_seconds_per_second": 2.129243744998014, "batches": [ { "batch_size": 1, "device_seconds": 3.1560501308441165, "audio_seconds": [ 6.72 ], "model_wall_seconds": 3.8533879593014717, "first_audio_seconds": 0.784553955078125, "graph_capture_seconds": 0, "live_callback_seconds": 0.6907781064510345 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 }, { "sentence": 3, "total_ms": 3938.22, "audio_s": 6.4 }, { "sentence": 4, "total_ms": 1834.77, "audio_s": 1.95 }, { "sentence": 5, "total_ms": 2458.68, "audio_s": 3.63 }, { "sentence": 6, "total_ms": 4237.13, "audio_s": 6.72 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true, "total_ms": 21109.56, "audio_s": 30.6, "sentences": 7, "generation_rtf": 0.6899 }, "at": 22965 }, { "type": "turn_done", "barged": false, "at": 34528.59999847412 }, { "type": "sync", "arrival_order_error_ms": null, "method": "capture timestamps + actual scheduled PCM sample positions", "hardware_accuracy": "not independently measured; inspect device latency availability", "at": 34529.59999847412 } ], "audio": [ { "at": 2747.199996948242, "reply_id": 1, "samples": 3590, "previous_end": 0, "now": 0.011609977324263039, "next": 0.31570578231292523, "underflows": 0 }, { "at": 2994.7999992370605, "reply_id": 1, "samples": 9600, "previous_end": 0.31570578231292523, "now": 0.011609977324263039, "next": 0.7157057823129253, "underflows": 0 }, { "at": 3227.3999977111816, "reply_id": 1, "samples": 9600, "previous_end": 0.7157057823129253, "now": 0.011609977324263039, "next": 1.1157057823129253, "underflows": 0 }, { "at": 3466.5, "reply_id": 1, "samples": 9600, "previous_end": 1.1157057823129253, "now": 0.011609977324263039, "next": 1.5157057823129252, "underflows": 0 }, { "at": 3700.699996948242, "reply_id": 1, "samples": 9600, "previous_end": 1.5157057823129252, "now": 0.058049886621315196, "next": 1.915705782312925, "underflows": 0 }, { "at": 3971.8999977111816, "reply_id": 1, "samples": 9600, "previous_end": 1.915705782312925, "now": 0.3250793650793651, "next": 2.315705782312925, "underflows": 0 }, { "at": 4204.699996948242, "reply_id": 1, "samples": 9600, "previous_end": 2.315705782312925, "now": 0.5572789115646258, "next": 2.715705782312925, "underflows": 0 }, { "at": 4227.099998474121, "reply_id": 1, "samples": 250, "previous_end": 2.715705782312925, "now": 0.5804988662131519, "next": 2.7261224489795914, "underflows": 0 }, { "at": 5365.599998474121, "reply_id": 1, "samples": 14390, "previous_end": 2.7261224489795914, "now": 1.7182766439909296, "next": 3.325705782312925, "underflows": 0 }, { "at": 6204.199996948242, "reply_id": 1, "samples": 38400, "previous_end": 3.325705782312925, "now": 2.5658049886621317, "next": 4.925705782312924, "underflows": 0 }, { "at": 6831, "reply_id": 1, "samples": 32640, "previous_end": 4.925705782312924, "now": 3.1927437641723357, "next": 6.285705782312925, "underflows": 0 }, { "at": 6866.199996948242, "reply_id": 1, "samples": 3600, "previous_end": 6.285705782312925, "now": 3.2275736961451247, "next": 6.435705782312925, "underflows": 0 }, { "at": 6866.5, "reply_id": 1, "samples": 10, "previous_end": 6.435705782312925, "now": 3.2275736961451247, "next": 6.436122448979591, "underflows": 0 }, { "at": 8057.299999237061, "reply_id": 1, "samples": 14390, "previous_end": 6.436122448979591, "now": 4.4117913832199545, "next": 7.035705782312925, "underflows": 0 }, { "at": 8887.099998474121, "reply_id": 1, "samples": 38400, "previous_end": 7.035705782312925, "now": 5.2477097505668935, "next": 8.635705782312925, "underflows": 0 }, { "at": 9790.5, "reply_id": 1, "samples": 38400, "previous_end": 8.635705782312925, "now": 6.141678004535147, "next": 10.235705782312925, "underflows": 0 }, { "at": 10430.199996948242, "reply_id": 1, "samples": 21120, "previous_end": 10.235705782312925, "now": 6.791836734693877, "next": 11.115705782312926, "underflows": 0 }, { "at": 10497.399997711182, "reply_id": 1, "samples": 22800, "previous_end": 11.115705782312926, "now": 6.849886621315193, "next": 12.065705782312925, "underflows": 0 }, { "at": 10497.899997711182, "reply_id": 1, "samples": 10, "previous_end": 12.065705782312925, "now": 6.849886621315193, "next": 12.066122448979591, "underflows": 0 }, { "at": 11699.29999923706, "reply_id": 1, "samples": 14390, "previous_end": 12.066122448979591, "now": 8.057324263038549, "next": 12.665705782312925, "underflows": 0 }, { "at": 12510.599998474121, "reply_id": 1, "samples": 38400, "previous_end": 12.665705782312925, "now": 8.870022675736962, "next": 14.265705782312924, "underflows": 0 }, { "at": 13413.79999923706, "reply_id": 1, "samples": 38400, "previous_end": 14.265705782312924, "now": 9.775600907029478, "next": 15.865705782312924, "underflows": 0 }, { "at": 14333.29999923706, "reply_id": 1, "samples": 38400, "previous_end": 15.865705782312924, "now": 10.692789115646258, "next": 17.465705782312924, "underflows": 0 }, { "at": 14429.899997711182, "reply_id": 1, "samples": 24000, "previous_end": 17.465705782312924, "now": 10.785668934240363, "next": 18.465705782312924, "underflows": 0 }, { "at": 14430.599998474121, "reply_id": 1, "samples": 10, "previous_end": 18.465705782312924, "now": 10.785668934240363, "next": 18.46612244897959, "underflows": 0 }, { "at": 15564.699996948242, "reply_id": 1, "samples": 14390, "previous_end": 18.46612244897959, "now": 11.92344671201814, "next": 19.065705782312925, "underflows": 0 }, { "at": 16230.29999923706, "reply_id": 1, "samples": 32400, "previous_end": 19.065705782312925, "now": 12.585215419501134, "next": 20.415705782312926, "underflows": 0 }, { "at": 16259.199996948242, "reply_id": 1, "samples": 10, "previous_end": 20.415705782312926, "now": 12.620045351473923, "next": 20.416122448979593, "underflows": 0 }, { "at": 17629.89999771118, "reply_id": 1, "samples": 14390, "previous_end": 20.416122448979593, "now": 13.990022675736961, "next": 21.015705782312928, "underflows": 0 }, { "at": 18262.699996948242, "reply_id": 1, "samples": 38400, "previous_end": 21.015705782312928, "now": 14.616961451247166, "next": 22.61570578231293, "underflows": 0 }, { "at": 18676.59999847412, "reply_id": 1, "samples": 13440, "previous_end": 22.61570578231293, "now": 15.034920634920635, "next": 23.175705782312928, "underflows": 0 }, { "at": 18726, "reply_id": 1, "samples": 20880, "previous_end": 23.175705782312928, "now": 15.081360544217688, "next": 24.04570578231293, "underflows": 0 }, { "at": 18726.699996948242, "reply_id": 1, "samples": 10, "previous_end": 24.04570578231293, "now": 15.081360544217688, "next": 24.046122448979595, "underflows": 0 }, { "at": 19935.59999847412, "reply_id": 1, "samples": 14390, "previous_end": 24.046122448979595, "now": 16.288798185941044, "next": 24.64570578231293, "underflows": 0 }, { "at": 20733.79999923706, "reply_id": 1, "samples": 38400, "previous_end": 24.64570578231293, "now": 17.08988662131519, "next": 26.245705782312932, "underflows": 0 }, { "at": 21582.699996948242, "reply_id": 1, "samples": 38400, "previous_end": 26.245705782312932, "now": 17.937414965986395, "next": 27.845705782312933, "underflows": 0 }, { "at": 22497.39999771118, "reply_id": 1, "samples": 38400, "previous_end": 27.845705782312933, "now": 18.854603174603174, "next": 29.445705782312935, "underflows": 0 }, { "at": 22886.79999923706, "reply_id": 1, "samples": 7680, "previous_end": 29.445705782312935, "now": 19.249342403628116, "next": 29.765705782312935, "underflows": 0 }, { "at": 22964.5, "reply_id": 1, "samples": 24000, "previous_end": 29.765705782312935, "now": 19.319002267573698, "next": 30.765705782312935, "underflows": 0 }, { "at": 22964.79999923706, "reply_id": 1, "samples": 10, "previous_end": 30.765705782312935, "now": 19.319002267573698, "next": 30.7661224489796, "underflows": 0 } ], "metrics": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 313.93, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 3, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 433.34, "llm_first_sentence_ms": 462.24, "finish_reason": "stop", "usage": { "prompt_tokens": 1426, "completion_tokens": 141, "total_tokens": 1567, "cost": 0.0016056117, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00162183, "upstream_inference_prompt_cost": 0.00141174, "upstream_inference_completions_cost": 0.00021009 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "tts_first_audio_ms": 559.3, "first_audio_ms": 893.16, "tts_server_stages": [ { "generation_ms": 1898, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2031, "performance": { "device_seconds": 1.6986648178100585, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.6635409444570541, "audio_seconds_per_second": 1.5070660045225321, "batches": [ { "batch_size": 1, "device_seconds": 1.6986648178100585, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.7031137943267822, "first_audio_seconds": 0.22162399291992188, "graph_capture_seconds": 0, "live_callback_seconds": 0.0016628429293632507 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2459, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2635, "performance": { "device_seconds": 1.9940713052749635, "raw_audio_seconds": 4.56, "aggregate_rtf": 0.43729633887608854, "audio_seconds_per_second": 2.2867788067243757, "batches": [ { "batch_size": 1, "device_seconds": 1.9940713052749635, "audio_seconds": [ 4.56 ], "model_wall_seconds": 2.2873197197914124, "first_audio_seconds": 0.7693322143554687, "graph_capture_seconds": 0, "live_callback_seconds": 0.28742706775665283 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3436, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3608, "performance": { "device_seconds": 2.6846351861953734, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.47264703982312917, "audio_seconds_per_second": 2.1157437067081037, "batches": [ { "batch_size": 1, "device_seconds": 2.6846351861953734, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.257918931543827, "first_audio_seconds": 0.791426025390625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5697517246007919 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3745, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3924, "performance": { "device_seconds": 2.9482006225585935, "raw_audio_seconds": 6.4, "aggregate_rtf": 0.4606563472747802, "audio_seconds_per_second": 2.170815632772564, "batches": [ { "batch_size": 1, "device_seconds": 2.9482006225585935, "audio_seconds": [ 6.4 ], "model_wall_seconds": 3.5646182149648666, "first_audio_seconds": 0.7783311157226562, "graph_capture_seconds": 0, "live_callback_seconds": 0.6068725809454918 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 1644, "alignment_ms": 52, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1826, "performance": { "device_seconds": 1.3384681282043458, "raw_audio_seconds": 2.88, "aggregate_rtf": 0.46474587784873117, "audio_seconds_per_second": 2.1517135442468347, "batches": [ { "batch_size": 1, "device_seconds": 1.3384681282043458, "audio_seconds": [ 2.88 ], "model_wall_seconds": 1.4674943834543228, "first_audio_seconds": 0.7459358520507813, "graph_capture_seconds": 0, "live_callback_seconds": 0.1282988339662552 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2280, "alignment_ms": 38, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2445, "performance": { "device_seconds": 1.844489884853363, "raw_audio_seconds": 3.76, "aggregate_rtf": 0.49055582043972423, "audio_seconds_per_second": 2.03850399553637, "batches": [ { "batch_size": 1, "device_seconds": 1.844489884853363, "audio_seconds": [ 3.76 ], "model_wall_seconds": 2.1057667583227158, "first_audio_seconds": 0.787409912109375, "graph_capture_seconds": 0, "live_callback_seconds": 0.2597941681742668 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 4037, "alignment_ms": 56, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4221, "performance": { "device_seconds": 3.1560501308441165, "raw_audio_seconds": 6.72, "aggregate_rtf": 0.4696503170898983, "audio_seconds_per_second": 2.129243744998014, "batches": [ { "batch_size": 1, "device_seconds": 3.1560501308441165, "audio_seconds": [ 6.72 ], "model_wall_seconds": 3.8533879593014717, "first_audio_seconds": 0.784553955078125, "graph_capture_seconds": 0, "live_callback_seconds": 0.6907781064510345 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.56, "quick_reply_tts_ms": 2039.62, "sentence_tts": [ { "sentence": 0, "total_ms": 2039.62, "audio_s": 2.56 }, { "sentence": 1, "total_ms": 2643.32, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 3617.67, "audio_s": 5.63 }, { "sentence": 3, "total_ms": 3938.22, "audio_s": 6.4 }, { "sentence": 4, "total_ms": 1834.77, "audio_s": 1.95 }, { "sentence": 5, "total_ms": 2458.68, "audio_s": 3.63 }, { "sentence": 6, "total_ms": 4237.13, "audio_s": 6.72 } ], "answer_first_audio_ms": 3506.07, "generation_calls": 1, "llm_total_ms": 5008.32, "response_complete": true, "total_ms": 21109.56, "audio_s": 30.6, "sentences": 7, "generation_rtf": 0.6899 }, "text": "Das klingt anstrengend. Oh, das tut mir leid, dass du so müde bist!\nStell dir vor, ein kleiner Roboter wie ich hat heute beschlossen, eine Wolke zu streicheln.\nEr hat so lange gepusht, bis die Wolke sich in eine riesige, weiche Pfanne verwandelt hat.\nUnd weißt du was? Er hat darin das beste Roboter-Omelett der Welt gebraten!\nAber bevor wir weiterträumen: Triffst du mich gleich, oder hast du lieber ein paar Minuten zum Durchatmen?", "first_pcm_ms": 966.4, "new_underflows": 0, "scheduled_playback_ms": 1120.91, "max_gap_ms": 0.0, "passed": true }, { "input": "Was ist der Sinn des Lebens? Erzähl mir deine Gedanken dazu.", "enabled": true, "events": [ { "type": "latency_start", "reply_id": 2, "input": { "kind": "typed" }, "at": 34833.59999847412 }, { "type": "reply", "text": "Das ist eine große Frage, oder?", "reply_id": 2, "performance": { "sentences": [ { "emotion": "interest", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "interest, soft", "text": "Das ist eine große Frage, oder?", "start_s": 0, "end_s": 2.04 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 0, "sentence_text": "Das ist eine große Frage, oder?", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.76, "llm_first_sentence_ms": 251.47, "sentence_ready_ms": 252.87, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "interest.soft" }, "sources": [], "lookups": [], "at": 35095.19999694824 }, { "type": "latency_update", "reply_id": 2, "metrics": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.76, "llm_first_sentence_ms": 251.47, "finish_reason": "stop", "usage": { "prompt_tokens": 1577, "completion_tokens": 67, "total_tokens": 1644, "cost": 0.0016444494, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00166106, "upstream_inference_prompt_cost": 0.00156123, "upstream_inference_completions_cost": 9.983e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 308.78, "response_complete": true, "answer_first_audio_ms": 1403.16, "tts_first_audio_ms": 1150.19, "first_audio_ms": 1403.17 }, "at": 36223.79999923706 }, { "type": "reply", "text": "Das ist eine große Frage, oder?\nIch denke, der Sinn liegt vielleicht darin, kleine Momente wahrzunehmen, wie dieses Gespräch.", "reply_id": 2, "performance": { "sentences": [ { "emotion": "interest", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "interest, soft", "text": "Das ist eine große Frage, oder?", "start_s": 0, "end_s": 2.75 }, { "emotion": "contemplation", "secondary": "contentment", "blend": 0.25, "intensity": 0.72, "direction": "contemplation, calm", "text": "Ich denke, der Sinn liegt vielleicht darin, kleine Momente wahrzunehmen, wie dieses Gespräch.", "start_s": 2.75, "end_s": 7.17 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 1, "sentence_text": "Ich denke, der Sinn liegt vielleicht darin, kleine Momente wahrzunehmen, wie dieses Gespräch.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.76, "llm_first_sentence_ms": 251.47, "finish_reason": "stop", "usage": { "prompt_tokens": 1577, "completion_tokens": 67, "total_tokens": 1644, "cost": 0.0016444494, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00166106, "upstream_inference_prompt_cost": 0.00156123, "upstream_inference_completions_cost": 9.983e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 308.78, "response_complete": true, "answer_first_audio_ms": 1403.16, "tts_first_audio_ms": 1150.19, "first_audio_ms": 1403.17, "tts_server_stages": [ { "generation_ms": 1698, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1870, "performance": { "device_seconds": 1.3051309804916382, "raw_audio_seconds": 2.8, "aggregate_rtf": 0.4661182073184423, "audio_seconds_per_second": 2.1453785419646154, "batches": [ { "batch_size": 1, "device_seconds": 1.3051309804916382, "audio_seconds": [ 2.8 ], "model_wall_seconds": 1.5115912109613419, "first_audio_seconds": 0.7318548583984374, "graph_capture_seconds": 0, "live_callback_seconds": 0.20131730288267136 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 1879.16, "audio_s": 2.75 } ], "sentence_ready_ms": 2132.67, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "contemplation.soft" }, "sources": [], "lookups": [], "at": 36960.89999771118 }, { "type": "latency_update", "reply_id": 2, "metrics": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.76, "llm_first_sentence_ms": 251.47, "finish_reason": "stop", "usage": { "prompt_tokens": 1577, "completion_tokens": 67, "total_tokens": 1644, "cost": 0.0016444494, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00166106, "upstream_inference_prompt_cost": 0.00156123, "upstream_inference_completions_cost": 9.983e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 308.78, "response_complete": true, "answer_first_audio_ms": 1403.16, "tts_first_audio_ms": 1150.19, "first_audio_ms": 1403.17, "tts_server_stages": [ { "generation_ms": 1698, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1870, "performance": { "device_seconds": 1.3051309804916382, "raw_audio_seconds": 2.8, "aggregate_rtf": 0.4661182073184423, "audio_seconds_per_second": 2.1453785419646154, "batches": [ { "batch_size": 1, "device_seconds": 1.3051309804916382, "audio_seconds": [ 2.8 ], "model_wall_seconds": 1.5115912109613419, "first_audio_seconds": 0.7318548583984374, "graph_capture_seconds": 0, "live_callback_seconds": 0.20131730288267136 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 1879.16, "audio_s": 2.75 } ] }, "at": 38155.79999923706 }, { "type": "reply", "text": "Das ist eine große Frage, oder?\nIch denke, der Sinn liegt vielleicht darin, kleine Momente wahrzunehmen, wie dieses Gespräch.\nOder darin, gemeinsam zu lachen, wenn ein Roboter versucht, Wolken zu streicheln.", "reply_id": 2, "performance": { "sentences": [ { "emotion": "interest", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "interest, soft", "text": "Das ist eine große Frage, oder?", "start_s": 0, "end_s": 2.75 }, { "emotion": "contemplation", "secondary": "contentment", "blend": 0.25, "intensity": 0.72, "direction": "contemplation, calm", "text": "Ich denke, der Sinn liegt vielleicht darin, kleine Momente wahrzunehmen, wie dieses Gespräch.", "start_s": 2.75, "end_s": 8.43 }, { "emotion": "affection", "secondary": "amusement", "blend": 0.25, "intensity": 0.72, "direction": "warm, playful", "text": "Oder darin, gemeinsam zu lachen, wenn ein Roboter versucht, Wolken zu streicheln.", "start_s": 8.43, "end_s": 12.51 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 2, "sentence_text": "Oder darin, gemeinsam zu lachen, wenn ein Roboter versucht, Wolken zu streicheln.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.76, "llm_first_sentence_ms": 251.47, "finish_reason": "stop", "usage": { "prompt_tokens": 1577, "completion_tokens": 67, "total_tokens": 1644, "cost": 0.0016444494, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00166106, "upstream_inference_prompt_cost": 0.00156123, "upstream_inference_completions_cost": 9.983e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 308.78, "response_complete": true, "answer_first_audio_ms": 1403.16, "tts_first_audio_ms": 1150.19, "first_audio_ms": 1403.17, "tts_server_stages": [ { "generation_ms": 1698, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1870, "performance": { "device_seconds": 1.3051309804916382, "raw_audio_seconds": 2.8, "aggregate_rtf": 0.4661182073184423, "audio_seconds_per_second": 2.1453785419646154, "batches": [ { "batch_size": 1, "device_seconds": 1.3051309804916382, "audio_seconds": [ 2.8 ], "model_wall_seconds": 1.5115912109613419, "first_audio_seconds": 0.7318548583984374, "graph_capture_seconds": 0, "live_callback_seconds": 0.20131730288267136 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3318, "alignment_ms": 71, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3518, "performance": { "device_seconds": 2.6611100687980653, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4685052938024763, "audio_seconds_per_second": 2.1344476001195494, "batches": [ { "batch_size": 1, "device_seconds": 2.6611100687980653, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1393518820405006, "first_audio_seconds": 0.7850762329101563, "graph_capture_seconds": 0, "live_callback_seconds": 0.47048160433769226 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 1879.16, "audio_s": 2.75 }, { "sentence": 1, "total_ms": 3531.39, "audio_s": 5.68 } ], "sentence_ready_ms": 5665.07, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "delivery.warm.soft" }, "sources": [], "lookups": [], "at": 40494.59999847412 }, { "type": "latency_update", "reply_id": 2, "metrics": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.76, "llm_first_sentence_ms": 251.47, "finish_reason": "stop", "usage": { "prompt_tokens": 1577, "completion_tokens": 67, "total_tokens": 1644, "cost": 0.0016444494, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00166106, "upstream_inference_prompt_cost": 0.00156123, "upstream_inference_completions_cost": 9.983e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 308.78, "response_complete": true, "answer_first_audio_ms": 1403.16, "tts_first_audio_ms": 1150.19, "first_audio_ms": 1403.17, "tts_server_stages": [ { "generation_ms": 1698, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1870, "performance": { "device_seconds": 1.3051309804916382, "raw_audio_seconds": 2.8, "aggregate_rtf": 0.4661182073184423, "audio_seconds_per_second": 2.1453785419646154, "batches": [ { "batch_size": 1, "device_seconds": 1.3051309804916382, "audio_seconds": [ 2.8 ], "model_wall_seconds": 1.5115912109613419, "first_audio_seconds": 0.7318548583984374, "graph_capture_seconds": 0, "live_callback_seconds": 0.20131730288267136 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3318, "alignment_ms": 71, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3518, "performance": { "device_seconds": 2.6611100687980653, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4685052938024763, "audio_seconds_per_second": 2.1344476001195494, "batches": [ { "batch_size": 1, "device_seconds": 2.6611100687980653, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1393518820405006, "first_audio_seconds": 0.7850762329101563, "graph_capture_seconds": 0, "live_callback_seconds": 0.47048160433769226 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 1879.16, "audio_s": 2.75 }, { "sentence": 1, "total_ms": 3531.39, "audio_s": 5.68 } ] }, "at": 41662.29999923706 }, { "type": "pipeline_metrics", "reply_id": 2, "metrics": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.76, "llm_first_sentence_ms": 251.47, "finish_reason": "stop", "usage": { "prompt_tokens": 1577, "completion_tokens": 67, "total_tokens": 1644, "cost": 0.0016444494, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00166106, "upstream_inference_prompt_cost": 0.00156123, "upstream_inference_completions_cost": 9.983e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 308.78, "response_complete": true, "answer_first_audio_ms": 1403.16, "tts_first_audio_ms": 1150.19, "first_audio_ms": 1403.17, "tts_server_stages": [ { "generation_ms": 1698, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1870, "performance": { "device_seconds": 1.3051309804916382, "raw_audio_seconds": 2.8, "aggregate_rtf": 0.4661182073184423, "audio_seconds_per_second": 2.1453785419646154, "batches": [ { "batch_size": 1, "device_seconds": 1.3051309804916382, "audio_seconds": [ 2.8 ], "model_wall_seconds": 1.5115912109613419, "first_audio_seconds": 0.7318548583984374, "graph_capture_seconds": 0, "live_callback_seconds": 0.20131730288267136 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3318, "alignment_ms": 71, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3518, "performance": { "device_seconds": 2.6611100687980653, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4685052938024763, "audio_seconds_per_second": 2.1344476001195494, "batches": [ { "batch_size": 1, "device_seconds": 2.6611100687980653, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1393518820405006, "first_audio_seconds": 0.7850762329101563, "graph_capture_seconds": 0, "live_callback_seconds": 0.47048160433769226 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3245, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3422, "performance": { "device_seconds": 2.5105804800987244, "raw_audio_seconds": 5.28, "aggregate_rtf": 0.47548872729142505, "audio_seconds_per_second": 2.1030992799690584, "batches": [ { "batch_size": 1, "device_seconds": 2.5105804800987244, "audio_seconds": [ 5.28 ], "model_wall_seconds": 3.0796079114079475, "first_audio_seconds": 0.790518798828125, "graph_capture_seconds": 0, "live_callback_seconds": 0.5653996989130974 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 1879.16, "audio_s": 2.75 }, { "sentence": 1, "total_ms": 3531.39, "audio_s": 5.68 }, { "sentence": 2, "total_ms": 3434.36, "audio_s": 4.91 } ], "total_ms": 9099.59, "audio_s": 13.34, "sentences": 3, "generation_rtf": 0.6821 }, "at": 43931.69999694824 }, { "type": "turn_done", "barged": false, "at": 50026.29999923706 }, { "type": "sync", "arrival_order_error_ms": null, "method": "capture timestamps + actual scheduled PCM sample positions", "hardware_accuracy": "not independently measured; inspect device latency availability", "at": 50027.09999847412 } ], "audio": [ { "at": 36226.59999847412, "reply_id": 2, "samples": 14390, "previous_end": 30.7661224489796, "now": 32.58920634920635, "next": 33.50330215419501, "underflows": 0 }, { "at": 36927.59999847412, "reply_id": 2, "samples": 51600, "previous_end": 33.50330215419501, "now": 33.28580498866213, "next": 35.65330215419501, "underflows": 0 }, { "at": 36960.39999771118, "reply_id": 2, "samples": 10, "previous_end": 35.65330215419501, "now": 33.32063492063492, "next": 35.65371882086168, "underflows": 0 }, { "at": 38159.69999694824, "reply_id": 2, "samples": 14390, "previous_end": 35.65371882086168, "now": 34.51646258503401, "next": 36.25330215419501, "underflows": 0 }, { "at": 38929.09999847412, "reply_id": 2, "samples": 38400, "previous_end": 36.25330215419501, "now": 35.28272108843537, "next": 37.85330215419501, "underflows": 0 }, { "at": 39798, "reply_id": 2, "samples": 38400, "previous_end": 37.85330215419501, "now": 36.1534693877551, "next": 39.453302154195015, "underflows": 0 }, { "at": 40412.89999771118, "reply_id": 2, "samples": 21120, "previous_end": 39.453302154195015, "now": 36.768798185941044, "next": 40.33330215419502, "underflows": 0 }, { "at": 40493.59999847412, "reply_id": 2, "samples": 24000, "previous_end": 40.33330215419502, "now": 36.85006802721088, "next": 41.33330215419502, "underflows": 0 }, { "at": 40494.39999771118, "reply_id": 2, "samples": 10, "previous_end": 41.33330215419502, "now": 36.85006802721088, "next": 41.333718820861684, "underflows": 0 }, { "at": 41665.39999771118, "reply_id": 2, "samples": 14390, "previous_end": 41.333718820861684, "now": 38.02267573696145, "next": 41.93330215419502, "underflows": 0 }, { "at": 42528.29999923706, "reply_id": 2, "samples": 38400, "previous_end": 41.93330215419502, "now": 38.88181405895691, "next": 43.53330215419502, "underflows": 0 }, { "at": 43427.5, "reply_id": 2, "samples": 38400, "previous_end": 43.53330215419502, "now": 39.787392290249436, "next": 45.13330215419502, "underflows": 0 }, { "at": 43864.29999923706, "reply_id": 2, "samples": 11520, "previous_end": 45.13330215419502, "now": 40.22857142857143, "next": 45.61330215419502, "underflows": 0 }, { "at": 43931.29999923706, "reply_id": 2, "samples": 15120, "previous_end": 45.61330215419502, "now": 40.28662131519275, "next": 46.24330215419502, "underflows": 0 }, { "at": 43931.69999694824, "reply_id": 2, "samples": 10, "previous_end": 46.24330215419502, "now": 40.28662131519275, "next": 46.24371882086169, "underflows": 0 } ], "metrics": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.76, "llm_first_sentence_ms": 251.47, "finish_reason": "stop", "usage": { "prompt_tokens": 1577, "completion_tokens": 67, "total_tokens": 1644, "cost": 0.0016444494, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00166106, "upstream_inference_prompt_cost": 0.00156123, "upstream_inference_completions_cost": 9.983e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 308.78, "response_complete": true, "answer_first_audio_ms": 1403.16, "tts_first_audio_ms": 1150.19, "first_audio_ms": 1403.17, "tts_server_stages": [ { "generation_ms": 1698, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1870, "performance": { "device_seconds": 1.3051309804916382, "raw_audio_seconds": 2.8, "aggregate_rtf": 0.4661182073184423, "audio_seconds_per_second": 2.1453785419646154, "batches": [ { "batch_size": 1, "device_seconds": 1.3051309804916382, "audio_seconds": [ 2.8 ], "model_wall_seconds": 1.5115912109613419, "first_audio_seconds": 0.7318548583984374, "graph_capture_seconds": 0, "live_callback_seconds": 0.20131730288267136 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3318, "alignment_ms": 71, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3518, "performance": { "device_seconds": 2.6611100687980653, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4685052938024763, "audio_seconds_per_second": 2.1344476001195494, "batches": [ { "batch_size": 1, "device_seconds": 2.6611100687980653, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1393518820405006, "first_audio_seconds": 0.7850762329101563, "graph_capture_seconds": 0, "live_callback_seconds": 0.47048160433769226 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3245, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3422, "performance": { "device_seconds": 2.5105804800987244, "raw_audio_seconds": 5.28, "aggregate_rtf": 0.47548872729142505, "audio_seconds_per_second": 2.1030992799690584, "batches": [ { "batch_size": 1, "device_seconds": 2.5105804800987244, "audio_seconds": [ 5.28 ], "model_wall_seconds": 3.0796079114079475, "first_audio_seconds": 0.790518798828125, "graph_capture_seconds": 0, "live_callback_seconds": 0.5653996989130974 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 1879.16, "audio_s": 2.75 }, { "sentence": 1, "total_ms": 3531.39, "audio_s": 5.68 }, { "sentence": 2, "total_ms": 3434.36, "audio_s": 4.91 } ], "total_ms": 9099.59, "audio_s": 13.34, "sentences": 3, "generation_rtf": 0.6821 }, "text": "Das ist eine große Frage, oder?\nIch denke, der Sinn liegt vielleicht darin, kleine Momente wahrzunehmen, wie dieses Gespräch.\nOder darin, gemeinsam zu lachen, wenn ein Roboter versucht, Wolken zu streicheln.", "first_pcm_ms": 1489.4, "new_underflows": 0, "scheduled_playback_ms": 1803.91, "max_gap_ms": 0, "passed": true }, { "input": "Schreib ein Gedicht mit vier Zeilen über einen kleinen Roboter.", "enabled": true, "events": [ { "type": "latency_start", "reply_id": 3, "input": { "kind": "typed" }, "at": 50261.69999694824 }, { "type": "reply", "text": "Ein kleiner Roboter rollt durch den Tag,", "reply_id": 3, "performance": { "sentences": [ { "emotion": "interest", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "interest, soft", "text": "Ein kleiner Roboter rollt durch den Tag,", "start_s": 0, "end_s": 2.38 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 0, "sentence_text": "Ein kleiner Roboter rollt durch den Tag,", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "sentence_ready_ms": 253.65, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "interest.soft" }, "sources": [], "lookups": [], "at": 50527.39999771118 }, { "type": "latency_update", "reply_id": 3, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "finish_reason": "stop", "usage": { "prompt_tokens": 1688, "completion_tokens": 68, "total_tokens": 1756, "cost": 0.0017547156, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00177244, "upstream_inference_prompt_cost": 0.00167112, "upstream_inference_completions_cost": 0.00010132 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "requested_units": 4, "completed_units": 4, "completion_repairs": 0, "llm_total_ms": 309.54, "response_complete": true, "answer_first_audio_ms": 1425.91, "tts_first_audio_ms": 1172.17, "first_audio_ms": 1425.92 }, "at": 51678.29999923706 }, { "type": "reply", "text": "Ein kleiner Roboter rollt durch den Tag,\nsein Herz ist ein Funke aus warmem Licht,", "reply_id": 3, "performance": { "sentences": [ { "emotion": "interest", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "interest, soft", "text": "Ein kleiner Roboter rollt durch den Tag,", "start_s": 0, "end_s": 2.59 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.72, "direction": "amusement, playful", "text": "sein Herz ist ein Funke aus warmem Licht,", "start_s": 2.59, "end_s": 5.3100000000000005 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 1, "sentence_text": "sein Herz ist ein Funke aus warmem Licht,", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "finish_reason": "stop", "usage": { "prompt_tokens": 1688, "completion_tokens": 68, "total_tokens": 1756, "cost": 0.0017547156, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00177244, "upstream_inference_prompt_cost": 0.00167112, "upstream_inference_completions_cost": 0.00010132 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "requested_units": 4, "completed_units": 4, "completion_repairs": 0, "llm_total_ms": 309.54, "response_complete": true, "answer_first_audio_ms": 1425.91, "tts_first_audio_ms": 1172.17, "first_audio_ms": 1425.92, "tts_server_stages": [ { "generation_ms": 1792, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2017, "performance": { "device_seconds": 1.4520742769241333, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.46540842209106836, "audio_seconds_per_second": 2.1486504165675067, "batches": [ { "batch_size": 1, "device_seconds": 1.4520742769241333, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.6174849420785904, "first_audio_seconds": 0.7600752563476563, "graph_capture_seconds": 0, "live_callback_seconds": 0.16299603879451752 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2029.78, "audio_s": 2.59 } ], "sentence_ready_ms": 2284.5, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "amusement.soft" }, "sources": [], "lookups": [], "at": 52544.29999923706 }, { "type": "latency_update", "reply_id": 3, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "finish_reason": "stop", "usage": { "prompt_tokens": 1688, "completion_tokens": 68, "total_tokens": 1756, "cost": 0.0017547156, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00177244, "upstream_inference_prompt_cost": 0.00167112, "upstream_inference_completions_cost": 0.00010132 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "requested_units": 4, "completed_units": 4, "completion_repairs": 0, "llm_total_ms": 309.54, "response_complete": true, "answer_first_audio_ms": 1425.91, "tts_first_audio_ms": 1172.17, "first_audio_ms": 1425.92, "tts_server_stages": [ { "generation_ms": 1792, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2017, "performance": { "device_seconds": 1.4520742769241333, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.46540842209106836, "audio_seconds_per_second": 2.1486504165675067, "batches": [ { "batch_size": 1, "device_seconds": 1.4520742769241333, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.6174849420785904, "first_audio_seconds": 0.7600752563476563, "graph_capture_seconds": 0, "live_callback_seconds": 0.16299603879451752 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2029.78, "audio_s": 2.59 } ] }, "at": 53693.79999923706 }, { "type": "reply", "text": "Ein kleiner Roboter rollt durch den Tag,\nsein Herz ist ein Funke aus warmem Licht,\ner sammelt die Träume der Menschen im Sack,", "reply_id": 3, "performance": { "sentences": [ { "emotion": "interest", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "interest, soft", "text": "Ein kleiner Roboter rollt durch den Tag,", "start_s": 0, "end_s": 2.59 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.72, "direction": "amusement, playful", "text": "sein Herz ist ein Funke aus warmem Licht,", "start_s": 2.59, "end_s": 7.9 }, { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "contentment, warm", "text": "er sammelt die Träume der Menschen im Sack,", "start_s": 7.9, "end_s": 10.620000000000001 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 2, "sentence_text": "er sammelt die Träume der Menschen im Sack,", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "finish_reason": "stop", "usage": { "prompt_tokens": 1688, "completion_tokens": 68, "total_tokens": 1756, "cost": 0.0017547156, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00177244, "upstream_inference_prompt_cost": 0.00167112, "upstream_inference_completions_cost": 0.00010132 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "requested_units": 4, "completed_units": 4, "completion_repairs": 0, "llm_total_ms": 309.54, "response_complete": true, "answer_first_audio_ms": 1425.91, "tts_first_audio_ms": 1172.17, "first_audio_ms": 1425.92, "tts_server_stages": [ { "generation_ms": 1792, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2017, "performance": { "device_seconds": 1.4520742769241333, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.46540842209106836, "audio_seconds_per_second": 2.1486504165675067, "batches": [ { "batch_size": 1, "device_seconds": 1.4520742769241333, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.6174849420785904, "first_audio_seconds": 0.7600752563476563, "graph_capture_seconds": 0, "live_callback_seconds": 0.16299603879451752 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3142, "alignment_ms": 67, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3334, "performance": { "device_seconds": 2.482361765384674, "raw_audio_seconds": 5.36, "aggregate_rtf": 0.46312719503445404, "audio_seconds_per_second": 2.1592340305682236, "batches": [ { "batch_size": 1, "device_seconds": 2.482361765384674, "audio_seconds": [ 5.36 ], "model_wall_seconds": 2.974411927163601, "first_audio_seconds": 0.756806640625, "graph_capture_seconds": 0, "live_callback_seconds": 0.48415323346853256 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2029.78, "audio_s": 2.59 }, { "sentence": 1, "total_ms": 3349.21, "audio_s": 5.31 } ], "sentence_ready_ms": 5634.59, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "contentment.soft" }, "sources": [], "lookups": [], "at": 55897.89999771118 }, { "type": "latency_update", "reply_id": 3, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "finish_reason": "stop", "usage": { "prompt_tokens": 1688, "completion_tokens": 68, "total_tokens": 1756, "cost": 0.0017547156, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00177244, "upstream_inference_prompt_cost": 0.00167112, "upstream_inference_completions_cost": 0.00010132 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "requested_units": 4, "completed_units": 4, "completion_repairs": 0, "llm_total_ms": 309.54, "response_complete": true, "answer_first_audio_ms": 1425.91, "tts_first_audio_ms": 1172.17, "first_audio_ms": 1425.92, "tts_server_stages": [ { "generation_ms": 1792, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2017, "performance": { "device_seconds": 1.4520742769241333, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.46540842209106836, "audio_seconds_per_second": 2.1486504165675067, "batches": [ { "batch_size": 1, "device_seconds": 1.4520742769241333, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.6174849420785904, "first_audio_seconds": 0.7600752563476563, "graph_capture_seconds": 0, "live_callback_seconds": 0.16299603879451752 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3142, "alignment_ms": 67, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3334, "performance": { "device_seconds": 2.482361765384674, "raw_audio_seconds": 5.36, "aggregate_rtf": 0.46312719503445404, "audio_seconds_per_second": 2.1592340305682236, "batches": [ { "batch_size": 1, "device_seconds": 2.482361765384674, "audio_seconds": [ 5.36 ], "model_wall_seconds": 2.974411927163601, "first_audio_seconds": 0.756806640625, "graph_capture_seconds": 0, "live_callback_seconds": 0.48415323346853256 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2029.78, "audio_s": 2.59 }, { "sentence": 1, "total_ms": 3349.21, "audio_s": 5.31 } ] }, "at": 57064.39999771118 }, { "type": "reply", "text": "Ein kleiner Roboter rollt durch den Tag,\nsein Herz ist ein Funke aus warmem Licht,\ner sammelt die Träume der Menschen im Sack,\nund tanzt mit den Sternen im stillen Gesicht.", "reply_id": 3, "performance": { "sentences": [ { "emotion": "interest", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "interest, soft", "text": "Ein kleiner Roboter rollt durch den Tag,", "start_s": 0, "end_s": 2.59 }, { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.72, "direction": "amusement, playful", "text": "sein Herz ist ein Funke aus warmem Licht,", "start_s": 2.59, "end_s": 7.9 }, { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.72, "direction": "contentment, warm", "text": "er sammelt die Träume der Menschen im Sack,", "start_s": 7.9, "end_s": 11.13 }, { "emotion": "hope", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "hope, gentle", "text": "und tanzt mit den Sternen im stillen Gesicht.", "start_s": 11.13, "end_s": 13.850000000000001 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 3, "sentence_text": "und tanzt mit den Sternen im stillen Gesicht.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "finish_reason": "stop", "usage": { "prompt_tokens": 1688, "completion_tokens": 68, "total_tokens": 1756, "cost": 0.0017547156, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00177244, "upstream_inference_prompt_cost": 0.00167112, "upstream_inference_completions_cost": 0.00010132 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "requested_units": 4, "completed_units": 4, "completion_repairs": 0, "llm_total_ms": 309.54, "response_complete": true, "answer_first_audio_ms": 1425.91, "tts_first_audio_ms": 1172.17, "first_audio_ms": 1425.92, "tts_server_stages": [ { "generation_ms": 1792, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2017, "performance": { "device_seconds": 1.4520742769241333, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.46540842209106836, "audio_seconds_per_second": 2.1486504165675067, "batches": [ { "batch_size": 1, "device_seconds": 1.4520742769241333, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.6174849420785904, "first_audio_seconds": 0.7600752563476563, "graph_capture_seconds": 0, "live_callback_seconds": 0.16299603879451752 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3142, "alignment_ms": 67, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3334, "performance": { "device_seconds": 2.482361765384674, "raw_audio_seconds": 5.36, "aggregate_rtf": 0.46312719503445404, "audio_seconds_per_second": 2.1592340305682236, "batches": [ { "batch_size": 1, "device_seconds": 2.482361765384674, "audio_seconds": [ 5.36 ], "model_wall_seconds": 2.974411927163601, "first_audio_seconds": 0.756806640625, "graph_capture_seconds": 0, "live_callback_seconds": 0.48415323346853256 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2480, "alignment_ms": 83, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2683, "performance": { "device_seconds": 2.0267013144493107, "raw_audio_seconds": 4.32, "aggregate_rtf": 0.46914382278919226, "audio_seconds_per_second": 2.131542506634145, "batches": [ { "batch_size": 1, "device_seconds": 2.0267013144493107, "audio_seconds": [ 4.32 ], "model_wall_seconds": 2.297702930867672, "first_audio_seconds": 0.772443115234375, "graph_capture_seconds": 0, "live_callback_seconds": 0.2639308124780655 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2029.78, "audio_s": 2.59 }, { "sentence": 1, "total_ms": 3349.21, "audio_s": 5.31 }, { "sentence": 2, "total_ms": 2689.11, "audio_s": 3.23 } ], "sentence_ready_ms": 8324.68, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "hope.soft" }, "sources": [], "lookups": [], "at": 58578.79999923706 }, { "type": "latency_update", "reply_id": 3, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "finish_reason": "stop", "usage": { "prompt_tokens": 1688, "completion_tokens": 68, "total_tokens": 1756, "cost": 0.0017547156, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00177244, "upstream_inference_prompt_cost": 0.00167112, "upstream_inference_completions_cost": 0.00010132 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "requested_units": 4, "completed_units": 4, "completion_repairs": 0, "llm_total_ms": 309.54, "response_complete": true, "answer_first_audio_ms": 1425.91, "tts_first_audio_ms": 1172.17, "first_audio_ms": 1425.92, "tts_server_stages": [ { "generation_ms": 1792, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2017, "performance": { "device_seconds": 1.4520742769241333, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.46540842209106836, "audio_seconds_per_second": 2.1486504165675067, "batches": [ { "batch_size": 1, "device_seconds": 1.4520742769241333, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.6174849420785904, "first_audio_seconds": 0.7600752563476563, "graph_capture_seconds": 0, "live_callback_seconds": 0.16299603879451752 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3142, "alignment_ms": 67, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3334, "performance": { "device_seconds": 2.482361765384674, "raw_audio_seconds": 5.36, "aggregate_rtf": 0.46312719503445404, "audio_seconds_per_second": 2.1592340305682236, "batches": [ { "batch_size": 1, "device_seconds": 2.482361765384674, "audio_seconds": [ 5.36 ], "model_wall_seconds": 2.974411927163601, "first_audio_seconds": 0.756806640625, "graph_capture_seconds": 0, "live_callback_seconds": 0.48415323346853256 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2480, "alignment_ms": 83, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2683, "performance": { "device_seconds": 2.0267013144493107, "raw_audio_seconds": 4.32, "aggregate_rtf": 0.46914382278919226, "audio_seconds_per_second": 2.131542506634145, "batches": [ { "batch_size": 1, "device_seconds": 2.0267013144493107, "audio_seconds": [ 4.32 ], "model_wall_seconds": 2.297702930867672, "first_audio_seconds": 0.772443115234375, "graph_capture_seconds": 0, "live_callback_seconds": 0.2639308124780655 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2029.78, "audio_s": 2.59 }, { "sentence": 1, "total_ms": 3349.21, "audio_s": 5.31 }, { "sentence": 2, "total_ms": 2689.11, "audio_s": 3.23 } ] }, "at": 59694.19999694824 }, { "type": "pipeline_metrics", "reply_id": 3, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "finish_reason": "stop", "usage": { "prompt_tokens": 1688, "completion_tokens": 68, "total_tokens": 1756, "cost": 0.0017547156, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00177244, "upstream_inference_prompt_cost": 0.00167112, "upstream_inference_completions_cost": 0.00010132 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "requested_units": 4, "completed_units": 4, "completion_repairs": 0, "llm_total_ms": 309.54, "response_complete": true, "answer_first_audio_ms": 1425.91, "tts_first_audio_ms": 1172.17, "first_audio_ms": 1425.92, "tts_server_stages": [ { "generation_ms": 1792, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2017, "performance": { "device_seconds": 1.4520742769241333, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.46540842209106836, "audio_seconds_per_second": 2.1486504165675067, "batches": [ { "batch_size": 1, "device_seconds": 1.4520742769241333, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.6174849420785904, "first_audio_seconds": 0.7600752563476563, "graph_capture_seconds": 0, "live_callback_seconds": 0.16299603879451752 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3142, "alignment_ms": 67, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3334, "performance": { "device_seconds": 2.482361765384674, "raw_audio_seconds": 5.36, "aggregate_rtf": 0.46312719503445404, "audio_seconds_per_second": 2.1592340305682236, "batches": [ { "batch_size": 1, "device_seconds": 2.482361765384674, "audio_seconds": [ 5.36 ], "model_wall_seconds": 2.974411927163601, "first_audio_seconds": 0.756806640625, "graph_capture_seconds": 0, "live_callback_seconds": 0.48415323346853256 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2480, "alignment_ms": 83, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2683, "performance": { "device_seconds": 2.0267013144493107, "raw_audio_seconds": 4.32, "aggregate_rtf": 0.46914382278919226, "audio_seconds_per_second": 2.131542506634145, "batches": [ { "batch_size": 1, "device_seconds": 2.0267013144493107, "audio_seconds": [ 4.32 ], "model_wall_seconds": 2.297702930867672, "first_audio_seconds": 0.772443115234375, "graph_capture_seconds": 0, "live_callback_seconds": 0.2639308124780655 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2150, "alignment_ms": 73, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2348, "performance": { "device_seconds": 1.6854095640182496, "raw_audio_seconds": 3.52, "aggregate_rtf": 0.4788095352324573, "audio_seconds_per_second": 2.088513127698073, "batches": [ { "batch_size": 1, "device_seconds": 1.6854095640182496, "audio_seconds": [ 3.52 ], "model_wall_seconds": 1.9760848134756088, "first_audio_seconds": 0.7567421264648437, "graph_capture_seconds": 0, "live_callback_seconds": 0.28535042703151703 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2029.78, "audio_s": 2.59 }, { "sentence": 1, "total_ms": 3349.21, "audio_s": 5.31 }, { "sentence": 2, "total_ms": 2689.11, "audio_s": 3.23 }, { "sentence": 3, "total_ms": 2358.76, "audio_s": 2.91 } ], "total_ms": 10683.65, "audio_s": 14.04, "sentences": 4, "generation_rtf": 0.7609 }, "at": 60937 }, { "type": "turn_done", "barged": false, "at": 66135.29999923706 }, { "type": "sync", "arrival_order_error_ms": null, "method": "capture timestamps + actual scheduled PCM sample positions", "hardware_accuracy": "not independently measured; inspect device latency availability", "at": 66135.39999771118 } ], "audio": [ { "at": 51682.29999923706, "reply_id": 3, "samples": 14390, "previous_end": 46.24371882086169, "now": 48.042086167800456, "next": 48.95618197278912, "underflows": 0 }, { "at": 52462.5, "reply_id": 3, "samples": 36480, "previous_end": 48.95618197278912, "now": 48.81995464852608, "next": 50.47618197278912, "underflows": 0 }, { "at": 52536.5, "reply_id": 3, "samples": 11280, "previous_end": 50.47618197278912, "now": 48.901224489795915, "next": 50.94618197278912, "underflows": 0 }, { "at": 52544, "reply_id": 3, "samples": 10, "previous_end": 50.94618197278912, "now": 48.901224489795915, "next": 50.946598639455786, "underflows": 0 }, { "at": 53694.09999847412, "reply_id": 3, "samples": 14390, "previous_end": 50.946598639455786, "now": 50.05061224489796, "next": 51.54618197278912, "underflows": 0 }, { "at": 54479.29999923706, "reply_id": 3, "samples": 38400, "previous_end": 51.54618197278912, "now": 50.84009070294785, "next": 53.14618197278912, "underflows": 0 }, { "at": 55363.59999847412, "reply_id": 3, "samples": 38400, "previous_end": 53.14618197278912, "now": 51.72244897959184, "next": 54.746181972789124, "underflows": 0 }, { "at": 55811.69999694824, "reply_id": 3, "samples": 13440, "previous_end": 54.746181972789124, "now": 52.17523809523809, "next": 55.306181972789126, "underflows": 0 }, { "at": 55897.5, "reply_id": 3, "samples": 22800, "previous_end": 55.306181972789126, "now": 52.25650793650794, "next": 56.25618197278913, "underflows": 0 }, { "at": 55897.69999694824, "reply_id": 3, "samples": 10, "previous_end": 56.25618197278913, "now": 52.25650793650794, "next": 56.256598639455795, "underflows": 0 }, { "at": 57064.69999694824, "reply_id": 3, "samples": 14390, "previous_end": 56.256598639455795, "now": 53.4291156462585, "next": 56.85618197278913, "underflows": 0 }, { "at": 57820.59999847412, "reply_id": 3, "samples": 38400, "previous_end": 56.85618197278913, "now": 54.1837641723356, "next": 58.45618197278913, "underflows": 0 }, { "at": 58495.59999847412, "reply_id": 3, "samples": 24720, "previous_end": 58.45618197278913, "now": 54.857142857142854, "next": 59.48618197278913, "underflows": 0 }, { "at": 58578.29999923706, "reply_id": 3, "samples": 10, "previous_end": 59.48618197278913, "now": 54.9384126984127, "next": 59.4865986394558, "underflows": 0 }, { "at": 59697.39999771118, "reply_id": 3, "samples": 14390, "previous_end": 59.4865986394558, "now": 56.05297052154195, "next": 60.086181972789134, "underflows": 0 }, { "at": 60528.09999847412, "reply_id": 3, "samples": 38400, "previous_end": 60.086181972789134, "now": 56.888888888888886, "next": 61.686181972789136, "underflows": 0 }, { "at": 60850.89999771118, "reply_id": 3, "samples": 7680, "previous_end": 61.686181972789136, "now": 57.213968253968254, "next": 62.006181972789136, "underflows": 0 }, { "at": 60936.39999771118, "reply_id": 3, "samples": 9360, "previous_end": 62.006181972789136, "now": 57.2952380952381, "next": 62.39618197278914, "underflows": 0 }, { "at": 60936.69999694824, "reply_id": 3, "samples": 10, "previous_end": 62.39618197278914, "now": 57.2952380952381, "next": 62.3965986394558, "underflows": 0 } ], "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 225.52, "llm_first_sentence_ms": 250.95, "finish_reason": "stop", "usage": { "prompt_tokens": 1688, "completion_tokens": 68, "total_tokens": 1756, "cost": 0.0017547156, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00177244, "upstream_inference_prompt_cost": 0.00167112, "upstream_inference_completions_cost": 0.00010132 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "requested_units": 4, "completed_units": 4, "completion_repairs": 0, "llm_total_ms": 309.54, "response_complete": true, "answer_first_audio_ms": 1425.91, "tts_first_audio_ms": 1172.17, "first_audio_ms": 1425.92, "tts_server_stages": [ { "generation_ms": 1792, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2017, "performance": { "device_seconds": 1.4520742769241333, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.46540842209106836, "audio_seconds_per_second": 2.1486504165675067, "batches": [ { "batch_size": 1, "device_seconds": 1.4520742769241333, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.6174849420785904, "first_audio_seconds": 0.7600752563476563, "graph_capture_seconds": 0, "live_callback_seconds": 0.16299603879451752 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3142, "alignment_ms": 67, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3334, "performance": { "device_seconds": 2.482361765384674, "raw_audio_seconds": 5.36, "aggregate_rtf": 0.46312719503445404, "audio_seconds_per_second": 2.1592340305682236, "batches": [ { "batch_size": 1, "device_seconds": 2.482361765384674, "audio_seconds": [ 5.36 ], "model_wall_seconds": 2.974411927163601, "first_audio_seconds": 0.756806640625, "graph_capture_seconds": 0, "live_callback_seconds": 0.48415323346853256 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2480, "alignment_ms": 83, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2683, "performance": { "device_seconds": 2.0267013144493107, "raw_audio_seconds": 4.32, "aggregate_rtf": 0.46914382278919226, "audio_seconds_per_second": 2.131542506634145, "batches": [ { "batch_size": 1, "device_seconds": 2.0267013144493107, "audio_seconds": [ 4.32 ], "model_wall_seconds": 2.297702930867672, "first_audio_seconds": 0.772443115234375, "graph_capture_seconds": 0, "live_callback_seconds": 0.2639308124780655 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 2150, "alignment_ms": 73, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2348, "performance": { "device_seconds": 1.6854095640182496, "raw_audio_seconds": 3.52, "aggregate_rtf": 0.4788095352324573, "audio_seconds_per_second": 2.088513127698073, "batches": [ { "batch_size": 1, "device_seconds": 1.6854095640182496, "audio_seconds": [ 3.52 ], "model_wall_seconds": 1.9760848134756088, "first_audio_seconds": 0.7567421264648437, "graph_capture_seconds": 0, "live_callback_seconds": 0.28535042703151703 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2029.78, "audio_s": 2.59 }, { "sentence": 1, "total_ms": 3349.21, "audio_s": 5.31 }, { "sentence": 2, "total_ms": 2689.11, "audio_s": 3.23 }, { "sentence": 3, "total_ms": 2358.76, "audio_s": 2.91 } ], "total_ms": 10683.65, "audio_s": 14.04, "sentences": 4, "generation_rtf": 0.7609 }, "text": "Ein kleiner Roboter rollt durch den Tag,\nsein Herz ist ein Funke aus warmem Licht,\ner sammelt die Träume der Menschen im Sack,\nund tanzt mit den Sternen im stillen Gesicht.", "first_pcm_ms": 1487.0, "new_underflows": 0, "scheduled_playback_ms": 1801.51, "max_gap_ms": 0, "passed": true }, { "input": "Wie entsteht ein Regenbogen?", "enabled": false, "events": [ { "type": "latency_start", "reply_id": 4, "input": { "kind": "typed" }, "at": 66460.59999847412 }, { "type": "reply", "text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.", "reply_id": 4, "performance": { "sentences": [ { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.", "start_s": 0, "end_s": 5.44 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 0, "sentence_text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "sentence_ready_ms": 272.64, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "delivery.calm.soft" }, "sources": [], "lookups": [], "at": 66727.5 }, { "type": "latency_update", "reply_id": 4, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "finish_reason": "stop", "usage": { "prompt_tokens": 1657, "completion_tokens": 110, "total_tokens": 1767, "cost": 0.0017862867, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00180433, "upstream_inference_prompt_cost": 0.00164043, "upstream_inference_completions_cost": 0.0001639 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 346.3, "response_complete": true, "answer_first_audio_ms": 1455.15, "tts_first_audio_ms": 1182.4, "first_audio_ms": 1455.16 }, "at": 67897.09999847412 }, { "type": "reply", "text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.\nDas Licht bricht dabei an den Tropfen, wird abgelenkt und zerlegt in seine verschiedenen Farben.", "reply_id": 4, "performance": { "sentences": [ { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.", "start_s": 0, "end_s": 6.56 }, { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Das Licht bricht dabei an den Tropfen, wird abgelenkt und zerlegt in seine verschiedenen Farben.", "start_s": 6.56, "end_s": 11.66 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 1, "sentence_text": "Das Licht bricht dabei an den Tropfen, wird abgelenkt und zerlegt in seine verschiedenen Farben.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "finish_reason": "stop", "usage": { "prompt_tokens": 1657, "completion_tokens": 110, "total_tokens": 1767, "cost": 0.0017862867, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00180433, "upstream_inference_prompt_cost": 0.00164043, "upstream_inference_completions_cost": 0.0001639 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 346.3, "response_complete": true, "answer_first_audio_ms": 1455.15, "tts_first_audio_ms": 1182.4, "first_audio_ms": 1455.16, "tts_server_stages": [ { "generation_ms": 4049, "alignment_ms": 96, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4265, "performance": { "device_seconds": 3.1464290723800663, "raw_audio_seconds": 6.56, "aggregate_rtf": 0.47963857810671745, "audio_seconds_per_second": 2.084903186785581, "batches": [ { "batch_size": 1, "device_seconds": 3.1464290723800663, "audio_seconds": [ 6.56 ], "model_wall_seconds": 3.870021916925907, "first_audio_seconds": 0.7871559448242188, "graph_capture_seconds": 0, "live_callback_seconds": 0.7200600579380989 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 4282.63, "audio_s": 6.56 } ], "sentence_ready_ms": 4556.27, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "delivery.calm.soft" }, "sources": [], "lookups": [], "at": 71003.09999847412 }, { "type": "latency_update", "reply_id": 4, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "finish_reason": "stop", "usage": { "prompt_tokens": 1657, "completion_tokens": 110, "total_tokens": 1767, "cost": 0.0017862867, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00180433, "upstream_inference_prompt_cost": 0.00164043, "upstream_inference_completions_cost": 0.0001639 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 346.3, "response_complete": true, "answer_first_audio_ms": 1455.15, "tts_first_audio_ms": 1182.4, "first_audio_ms": 1455.16, "tts_server_stages": [ { "generation_ms": 4049, "alignment_ms": 96, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4265, "performance": { "device_seconds": 3.1464290723800663, "raw_audio_seconds": 6.56, "aggregate_rtf": 0.47963857810671745, "audio_seconds_per_second": 2.084903186785581, "batches": [ { "batch_size": 1, "device_seconds": 3.1464290723800663, "audio_seconds": [ 6.56 ], "model_wall_seconds": 3.870021916925907, "first_audio_seconds": 0.7871559448242188, "graph_capture_seconds": 0, "live_callback_seconds": 0.7200600579380989 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 4282.63, "audio_s": 6.56 } ] }, "at": 72088.19999694824 }, { "type": "reply", "text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.\nDas Licht bricht dabei an den Tropfen, wird abgelenkt und zerlegt in seine verschiedenen Farben.\nVon außen sehen wir genau die Farben, die von den Tropfen in unsere Richtung zurückgeworfen werden.", "reply_id": 4, "performance": { "sentences": [ { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.", "start_s": 0, "end_s": 6.56 }, { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Das Licht bricht dabei an den Tropfen, wird abgelenkt und zerlegt in seine verschiedenen Farben.", "start_s": 6.56, "end_s": 12.35 }, { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Von außen sehen wir genau die Farben, die von den Tropfen in unsere Richtung zurückgeworfen werden.", "start_s": 12.35, "end_s": 17.79 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 2, "sentence_text": "Von außen sehen wir genau die Farben, die von den Tropfen in unsere Richtung zurückgeworfen werden.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "finish_reason": "stop", "usage": { "prompt_tokens": 1657, "completion_tokens": 110, "total_tokens": 1767, "cost": 0.0017862867, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00180433, "upstream_inference_prompt_cost": 0.00164043, "upstream_inference_completions_cost": 0.0001639 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 346.3, "response_complete": true, "answer_first_audio_ms": 1455.15, "tts_first_audio_ms": 1182.4, "first_audio_ms": 1455.16, "tts_server_stages": [ { "generation_ms": 4049, "alignment_ms": 96, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4265, "performance": { "device_seconds": 3.1464290723800663, "raw_audio_seconds": 6.56, "aggregate_rtf": 0.47963857810671745, "audio_seconds_per_second": 2.084903186785581, "batches": [ { "batch_size": 1, "device_seconds": 3.1464290723800663, "audio_seconds": [ 6.56 ], "model_wall_seconds": 3.870021916925907, "first_audio_seconds": 0.7871559448242188, "graph_capture_seconds": 0, "live_callback_seconds": 0.7200600579380989 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3382, "alignment_ms": 66, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3571, "performance": { "device_seconds": 2.6530347571372985, "raw_audio_seconds": 5.92, "aggregate_rtf": 0.4481477630299491, "audio_seconds_per_second": 2.2314068762475814, "batches": [ { "batch_size": 1, "device_seconds": 2.6530347571372985, "audio_seconds": [ 5.92 ], "model_wall_seconds": 3.2132324278354645, "first_audio_seconds": 0.6819090576171875, "graph_capture_seconds": 0, "live_callback_seconds": 0.5520960539579391 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 4282.63, "audio_s": 6.56 }, { "sentence": 1, "total_ms": 3583.95, "audio_s": 5.79 } ], "sentence_ready_ms": 8141.09, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "delivery.calm.soft" }, "sources": [], "lookups": [], "at": 74585.29999923706 }, { "type": "latency_update", "reply_id": 4, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "finish_reason": "stop", "usage": { "prompt_tokens": 1657, "completion_tokens": 110, "total_tokens": 1767, "cost": 0.0017862867, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00180433, "upstream_inference_prompt_cost": 0.00164043, "upstream_inference_completions_cost": 0.0001639 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 346.3, "response_complete": true, "answer_first_audio_ms": 1455.15, "tts_first_audio_ms": 1182.4, "first_audio_ms": 1455.16, "tts_server_stages": [ { "generation_ms": 4049, "alignment_ms": 96, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4265, "performance": { "device_seconds": 3.1464290723800663, "raw_audio_seconds": 6.56, "aggregate_rtf": 0.47963857810671745, "audio_seconds_per_second": 2.084903186785581, "batches": [ { "batch_size": 1, "device_seconds": 3.1464290723800663, "audio_seconds": [ 6.56 ], "model_wall_seconds": 3.870021916925907, "first_audio_seconds": 0.7871559448242188, "graph_capture_seconds": 0, "live_callback_seconds": 0.7200600579380989 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3382, "alignment_ms": 66, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3571, "performance": { "device_seconds": 2.6530347571372985, "raw_audio_seconds": 5.92, "aggregate_rtf": 0.4481477630299491, "audio_seconds_per_second": 2.2314068762475814, "batches": [ { "batch_size": 1, "device_seconds": 2.6530347571372985, "audio_seconds": [ 5.92 ], "model_wall_seconds": 3.2132324278354645, "first_audio_seconds": 0.6819090576171875, "graph_capture_seconds": 0, "live_callback_seconds": 0.5520960539579391 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 4282.63, "audio_s": 6.56 }, { "sentence": 1, "total_ms": 3583.95, "audio_s": 5.79 } ] }, "at": 75768.79999923706 }, { "type": "reply", "text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.\nDas Licht bricht dabei an den Tropfen, wird abgelenkt und zerlegt in seine verschiedenen Farben.\nVon außen sehen wir genau die Farben, die von den Tropfen in unsere Richtung zurückgeworfen werden.\nDeshalb sieht jeder Beobachter eigentlich einen eigenen Bogen aus seiner eigenen Perspektive.", "reply_id": 4, "performance": { "sentences": [ { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.", "start_s": 0, "end_s": 6.56 }, { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Das Licht bricht dabei an den Tropfen, wird abgelenkt und zerlegt in seine verschiedenen Farben.", "start_s": 6.56, "end_s": 12.35 }, { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Von außen sehen wir genau die Farben, die von den Tropfen in unsere Richtung zurückgeworfen werden.", "start_s": 12.35, "end_s": 18.46 }, { "emotion": "contentment", "secondary": null, "blend": 0, "intensity": 0.55, "direction": "calm, soft", "text": "Deshalb sieht jeder Beobachter eigentlich einen eigenen Bogen aus seiner eigenen Perspektive.", "start_s": 18.46, "end_s": 22.54 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 3, "sentence_text": "Deshalb sieht jeder Beobachter eigentlich einen eigenen Bogen aus seiner eigenen Perspektive.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "finish_reason": "stop", "usage": { "prompt_tokens": 1657, "completion_tokens": 110, "total_tokens": 1767, "cost": 0.0017862867, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00180433, "upstream_inference_prompt_cost": 0.00164043, "upstream_inference_completions_cost": 0.0001639 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 346.3, "response_complete": true, "answer_first_audio_ms": 1455.15, "tts_first_audio_ms": 1182.4, "first_audio_ms": 1455.16, "tts_server_stages": [ { "generation_ms": 4049, "alignment_ms": 96, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4265, "performance": { "device_seconds": 3.1464290723800663, "raw_audio_seconds": 6.56, "aggregate_rtf": 0.47963857810671745, "audio_seconds_per_second": 2.084903186785581, "batches": [ { "batch_size": 1, "device_seconds": 3.1464290723800663, "audio_seconds": [ 6.56 ], "model_wall_seconds": 3.870021916925907, "first_audio_seconds": 0.7871559448242188, "graph_capture_seconds": 0, "live_callback_seconds": 0.7200600579380989 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3382, "alignment_ms": 66, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3571, "performance": { "device_seconds": 2.6530347571372985, "raw_audio_seconds": 5.92, "aggregate_rtf": 0.4481477630299491, "audio_seconds_per_second": 2.2314068762475814, "batches": [ { "batch_size": 1, "device_seconds": 2.6530347571372985, "audio_seconds": [ 5.92 ], "model_wall_seconds": 3.2132324278354645, "first_audio_seconds": 0.6819090576171875, "graph_capture_seconds": 0, "live_callback_seconds": 0.5520960539579391 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3554, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3782, "performance": { "device_seconds": 2.8810144653320315, "raw_audio_seconds": 6.16, "aggregate_rtf": 0.4676971534629921, "audio_seconds_per_second": 2.138135741463579, "batches": [ { "batch_size": 1, "device_seconds": 2.8810144653320315, "audio_seconds": [ 6.16 ], "model_wall_seconds": 3.3861640617251396, "first_audio_seconds": 0.7867689208984375, "graph_capture_seconds": 0, "live_callback_seconds": 0.4999021738767624 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 4282.63, "audio_s": 6.56 }, { "sentence": 1, "total_ms": 3583.95, "audio_s": 5.79 }, { "sentence": 2, "total_ms": 3794.58, "audio_s": 6.11 } ], "sentence_ready_ms": 11936.65, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "delivery.calm.soft" }, "sources": [], "lookups": [], "at": 78387 }, { "type": "latency_update", "reply_id": 4, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "finish_reason": "stop", "usage": { "prompt_tokens": 1657, "completion_tokens": 110, "total_tokens": 1767, "cost": 0.0017862867, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00180433, "upstream_inference_prompt_cost": 0.00164043, "upstream_inference_completions_cost": 0.0001639 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 346.3, "response_complete": true, "answer_first_audio_ms": 1455.15, "tts_first_audio_ms": 1182.4, "first_audio_ms": 1455.16, "tts_server_stages": [ { "generation_ms": 4049, "alignment_ms": 96, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4265, "performance": { "device_seconds": 3.1464290723800663, "raw_audio_seconds": 6.56, "aggregate_rtf": 0.47963857810671745, "audio_seconds_per_second": 2.084903186785581, "batches": [ { "batch_size": 1, "device_seconds": 3.1464290723800663, "audio_seconds": [ 6.56 ], "model_wall_seconds": 3.870021916925907, "first_audio_seconds": 0.7871559448242188, "graph_capture_seconds": 0, "live_callback_seconds": 0.7200600579380989 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3382, "alignment_ms": 66, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3571, "performance": { "device_seconds": 2.6530347571372985, "raw_audio_seconds": 5.92, "aggregate_rtf": 0.4481477630299491, "audio_seconds_per_second": 2.2314068762475814, "batches": [ { "batch_size": 1, "device_seconds": 2.6530347571372985, "audio_seconds": [ 5.92 ], "model_wall_seconds": 3.2132324278354645, "first_audio_seconds": 0.6819090576171875, "graph_capture_seconds": 0, "live_callback_seconds": 0.5520960539579391 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3554, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3782, "performance": { "device_seconds": 2.8810144653320315, "raw_audio_seconds": 6.16, "aggregate_rtf": 0.4676971534629921, "audio_seconds_per_second": 2.138135741463579, "batches": [ { "batch_size": 1, "device_seconds": 2.8810144653320315, "audio_seconds": [ 6.16 ], "model_wall_seconds": 3.3861640617251396, "first_audio_seconds": 0.7867689208984375, "graph_capture_seconds": 0, "live_callback_seconds": 0.4999021738767624 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 4282.63, "audio_s": 6.56 }, { "sentence": 1, "total_ms": 3583.95, "audio_s": 5.79 }, { "sentence": 2, "total_ms": 3794.58, "audio_s": 6.11 } ] }, "at": 79528.5 }, { "type": "pipeline_metrics", "reply_id": 4, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "finish_reason": "stop", "usage": { "prompt_tokens": 1657, "completion_tokens": 110, "total_tokens": 1767, "cost": 0.0017862867, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00180433, "upstream_inference_prompt_cost": 0.00164043, "upstream_inference_completions_cost": 0.0001639 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 346.3, "response_complete": true, "answer_first_audio_ms": 1455.15, "tts_first_audio_ms": 1182.4, "first_audio_ms": 1455.16, "tts_server_stages": [ { "generation_ms": 4049, "alignment_ms": 96, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4265, "performance": { "device_seconds": 3.1464290723800663, "raw_audio_seconds": 6.56, "aggregate_rtf": 0.47963857810671745, "audio_seconds_per_second": 2.084903186785581, "batches": [ { "batch_size": 1, "device_seconds": 3.1464290723800663, "audio_seconds": [ 6.56 ], "model_wall_seconds": 3.870021916925907, "first_audio_seconds": 0.7871559448242188, "graph_capture_seconds": 0, "live_callback_seconds": 0.7200600579380989 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3382, "alignment_ms": 66, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3571, "performance": { "device_seconds": 2.6530347571372985, "raw_audio_seconds": 5.92, "aggregate_rtf": 0.4481477630299491, "audio_seconds_per_second": 2.2314068762475814, "batches": [ { "batch_size": 1, "device_seconds": 2.6530347571372985, "audio_seconds": [ 5.92 ], "model_wall_seconds": 3.2132324278354645, "first_audio_seconds": 0.6819090576171875, "graph_capture_seconds": 0, "live_callback_seconds": 0.5520960539579391 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3554, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3782, "performance": { "device_seconds": 2.8810144653320315, "raw_audio_seconds": 6.16, "aggregate_rtf": 0.4676971534629921, "audio_seconds_per_second": 2.138135741463579, "batches": [ { "batch_size": 1, "device_seconds": 2.8810144653320315, "audio_seconds": [ 6.16 ], "model_wall_seconds": 3.3861640617251396, "first_audio_seconds": 0.7867689208984375, "graph_capture_seconds": 0, "live_callback_seconds": 0.4999021738767624 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3182, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3363, "performance": { "device_seconds": 2.4594121899604797, "raw_audio_seconds": 5.28, "aggregate_rtf": 0.4657977632500908, "audio_seconds_per_second": 2.1468544482105885, "batches": [ { "batch_size": 1, "device_seconds": 2.4594121899604797, "audio_seconds": [ 5.28 ], "model_wall_seconds": 3.0127760618925095, "first_audio_seconds": 0.7673385009765625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5421648398041725 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 4282.63, "audio_s": 6.56 }, { "sentence": 1, "total_ms": 3583.95, "audio_s": 5.79 }, { "sentence": 2, "total_ms": 3794.58, "audio_s": 6.11 }, { "sentence": 3, "total_ms": 3371.58, "audio_s": 5.07 } ], "total_ms": 15308.4, "audio_s": 23.53, "sentences": 4, "generation_rtf": 0.6506 }, "at": 81750.59999847412 }, { "type": "turn_done", "barged": false, "at": 91861.79999923706 }, { "type": "sync", "arrival_order_error_ms": null, "method": "capture timestamps + actual scheduled PCM sample positions", "hardware_accuracy": "not independently measured; inspect device latency availability", "at": 91862.69999694824 } ], "audio": [ { "at": 67901, "reply_id": 4, "samples": 14390, "previous_end": 62.3965986394558, "now": 64.26122448979592, "next": 65.17532029478457, "underflows": 0 }, { "at": 68770, "reply_id": 4, "samples": 38400, "previous_end": 65.17532029478457, "now": 65.13197278911565, "next": 66.77532029478456, "underflows": 0 }, { "at": 69645.19999694824, "reply_id": 4, "samples": 38400, "previous_end": 66.77532029478456, "now": 66.00272108843538, "next": 68.37532029478456, "underflows": 0 }, { "at": 70544.69999694824, "reply_id": 4, "samples": 38400, "previous_end": 68.37532029478456, "now": 66.90829931972789, "next": 69.97532029478455, "underflows": 0 }, { "at": 70886.19999694824, "reply_id": 4, "samples": 3840, "previous_end": 69.97532029478455, "now": 67.24498866213152, "next": 70.13532029478455, "underflows": 0 }, { "at": 71000.5, "reply_id": 4, "samples": 24000, "previous_end": 70.13532029478455, "now": 67.36108843537414, "next": 71.13532029478455, "underflows": 0 }, { "at": 71002.89999771118, "reply_id": 4, "samples": 10, "previous_end": 71.13532029478455, "now": 67.36108843537414, "next": 71.13573696145122, "underflows": 0 }, { "at": 72091.69999694824, "reply_id": 4, "samples": 14390, "previous_end": 71.13573696145122, "now": 68.45242630385488, "next": 71.73532029478454, "underflows": 0 }, { "at": 72950.39999771118, "reply_id": 4, "samples": 38400, "previous_end": 71.73532029478454, "now": 69.31156462585034, "next": 73.33532029478454, "underflows": 0 }, { "at": 73845.09999847412, "reply_id": 4, "samples": 38400, "previous_end": 73.33532029478454, "now": 70.20553287981859, "next": 74.93532029478453, "underflows": 0 }, { "at": 74513.79999923706, "reply_id": 4, "samples": 26880, "previous_end": 74.93532029478453, "now": 70.87891156462585, "next": 76.05532029478454, "underflows": 0 }, { "at": 74584.79999923706, "reply_id": 4, "samples": 20880, "previous_end": 76.05532029478454, "now": 70.94857142857143, "next": 76.92532029478454, "underflows": 0 }, { "at": 74585.29999923706, "reply_id": 4, "samples": 10, "previous_end": 76.92532029478454, "now": 70.94857142857143, "next": 76.92573696145121, "underflows": 0 }, { "at": 75777.5, "reply_id": 4, "samples": 14390, "previous_end": 76.92573696145121, "now": 72.14439909297052, "next": 77.52532029478454, "underflows": 0 }, { "at": 76592.89999771118, "reply_id": 4, "samples": 38400, "previous_end": 77.52532029478454, "now": 72.95709750566894, "next": 79.12532029478453, "underflows": 0 }, { "at": 77510.39999771118, "reply_id": 4, "samples": 38400, "previous_end": 79.12532029478453, "now": 73.87428571428572, "next": 80.72532029478452, "underflows": 0 }, { "at": 78280.69999694824, "reply_id": 4, "samples": 32640, "previous_end": 80.72532029478452, "now": 74.64054421768708, "next": 82.08532029478452, "underflows": 0 }, { "at": 78386.29999923706, "reply_id": 4, "samples": 22800, "previous_end": 82.08532029478452, "now": 74.74503401360545, "next": 83.03532029478453, "underflows": 0 }, { "at": 78386.79999923706, "reply_id": 4, "samples": 10, "previous_end": 83.03532029478453, "now": 74.74503401360545, "next": 83.03573696145119, "underflows": 0 }, { "at": 79532.29999923706, "reply_id": 4, "samples": 14390, "previous_end": 83.03573696145119, "now": 75.89442176870749, "next": 83.63532029478452, "underflows": 0 }, { "at": 80381.69999694824, "reply_id": 4, "samples": 38400, "previous_end": 83.63532029478452, "now": 76.74195011337868, "next": 85.23532029478451, "underflows": 0 }, { "at": 81244.79999923706, "reply_id": 4, "samples": 38400, "previous_end": 85.23532029478451, "now": 77.6126984126984, "next": 86.83532029478451, "underflows": 0 }, { "at": 81701.19999694824, "reply_id": 4, "samples": 30480, "previous_end": 86.83532029478451, "now": 78.06548752834468, "next": 88.1053202947845, "underflows": 0 }, { "at": 81750.19999694824, "reply_id": 4, "samples": 10, "previous_end": 88.1053202947845, "now": 78.11192743764173, "next": 88.10573696145117, "underflows": 0 } ], "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 224.89, "llm_first_sentence_ms": 271.5, "finish_reason": "stop", "usage": { "prompt_tokens": 1657, "completion_tokens": 110, "total_tokens": 1767, "cost": 0.0017862867, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00180433, "upstream_inference_prompt_cost": 0.00164043, "upstream_inference_completions_cost": 0.0001639 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 346.3, "response_complete": true, "answer_first_audio_ms": 1455.15, "tts_first_audio_ms": 1182.4, "first_audio_ms": 1455.16, "tts_server_stages": [ { "generation_ms": 4049, "alignment_ms": 96, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 4265, "performance": { "device_seconds": 3.1464290723800663, "raw_audio_seconds": 6.56, "aggregate_rtf": 0.47963857810671745, "audio_seconds_per_second": 2.084903186785581, "batches": [ { "batch_size": 1, "device_seconds": 3.1464290723800663, "audio_seconds": [ 6.56 ], "model_wall_seconds": 3.870021916925907, "first_audio_seconds": 0.7871559448242188, "graph_capture_seconds": 0, "live_callback_seconds": 0.7200600579380989 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3382, "alignment_ms": 66, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3571, "performance": { "device_seconds": 2.6530347571372985, "raw_audio_seconds": 5.92, "aggregate_rtf": 0.4481477630299491, "audio_seconds_per_second": 2.2314068762475814, "batches": [ { "batch_size": 1, "device_seconds": 2.6530347571372985, "audio_seconds": [ 5.92 ], "model_wall_seconds": 3.2132324278354645, "first_audio_seconds": 0.6819090576171875, "graph_capture_seconds": 0, "live_callback_seconds": 0.5520960539579391 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3554, "alignment_ms": 99, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3782, "performance": { "device_seconds": 2.8810144653320315, "raw_audio_seconds": 6.16, "aggregate_rtf": 0.4676971534629921, "audio_seconds_per_second": 2.138135741463579, "batches": [ { "batch_size": 1, "device_seconds": 2.8810144653320315, "audio_seconds": [ 6.16 ], "model_wall_seconds": 3.3861640617251396, "first_audio_seconds": 0.7867689208984375, "graph_capture_seconds": 0, "live_callback_seconds": 0.4999021738767624 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } }, { "generation_ms": 3182, "alignment_ms": 54, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3363, "performance": { "device_seconds": 2.4594121899604797, "raw_audio_seconds": 5.28, "aggregate_rtf": 0.4657977632500908, "audio_seconds_per_second": 2.1468544482105885, "batches": [ { "batch_size": 1, "device_seconds": 2.4594121899604797, "audio_seconds": [ 5.28 ], "model_wall_seconds": 3.0127760618925095, "first_audio_seconds": 0.7673385009765625, "graph_capture_seconds": 0, "live_callback_seconds": 0.5421648398041725 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 4282.63, "audio_s": 6.56 }, { "sentence": 1, "total_ms": 3583.95, "audio_s": 5.79 }, { "sentence": 2, "total_ms": 3794.58, "audio_s": 6.11 }, { "sentence": 3, "total_ms": 3371.58, "audio_s": 5.07 } ], "total_ms": 15308.4, "audio_s": 23.53, "sentences": 4, "generation_rtf": 0.6506 }, "text": "Ein Regenbogen entsteht, wenn Sonnenlicht in die Luft trifft, in der noch viele kleine Wassertröpfchen schweben.\nDas Licht bricht dabei an den Tropfen, wird abgelenkt und zerlegt in seine verschiedenen Farben.\nVon außen sehen wir genau die Farben, die von den Tropfen in unsere Richtung zurückgeworfen werden.\nDeshalb sieht jeder Beobachter eigentlich einen eigenen Bogen aus seiner eigenen Perspektive.", "first_pcm_ms": 1567.6, "new_underflows": 0, "scheduled_playback_ms": 1882.11, "max_gap_ms": 0, "passed": true } ], "layout": { "all_static_controls_retained": true, "boxes": { "avatar": { "x": 140, "y": 119.3125, "width": 470, "height": 385 }, "conversation": { "x": 699, "y": 119.3125, "width": 584, "height": 346.640625 }, "latency": { "x": 83, "y": 662.3125, "width": 1200, "height": 156 }, "models": { "x": 689, "y": 1077.28125, "width": 598, "height": 51.09375 } }, "camera_requests": 0 }, "cancellation": { "events": [ { "type": "latency_start", "reply_id": 5, "input": { "kind": "typed" }, "at": 91994.79999923706 }, { "type": "reply", "text": "Es war einmal ein kleiner Roboter namens Knopf, der in einer verstaubten Werkstatt lebte.", "reply_id": 5, "performance": { "sentences": [ { "emotion": "amusement", "secondary": null, "blend": 0, "intensity": 0.72, "direction": "eager, playful", "text": "Es war einmal ein kleiner Roboter namens Knopf, der in einer verstaubten Werkstatt lebte.", "start_s": 0, "end_s": 4.76 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "sentence_index": 0, "sentence_text": "Es war einmal ein kleiner Roboter namens Knopf, der in einer verstaubten Werkstatt lebte.", "sentence_role": "answer", "speaker_id": null, "speaker_name": null, "speaker_update": null, "timing": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 219.16, "llm_first_sentence_ms": 246.87, "sentence_ready_ms": 247.94, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "delivery.playful.soft" }, "sources": [], "lookups": [], "at": 92229 }, { "type": "latency_update", "reply_id": 5, "metrics": { "quick_reply_status": "disabled_or_ineligible", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 219.16, "llm_first_sentence_ms": 246.87, "answer_first_audio_ms": 1414.76, "tts_first_audio_ms": 1166.73, "first_audio_ms": 1414.76 }, "at": 93392.5 }, { "type": "reply_cancelled", "reply_id": 5, "at": 93490.5 }, { "type": "barge", "yield": 1, "at": 93490.69999694824 }, { "type": "turn_done", "barged": true, "at": 93492.39999771118 }, { "type": "sync", "arrival_order_error_ms": null, "method": "capture timestamps + actual scheduled PCM sample positions", "hardware_accuracy": "not independently measured; inspect device latency availability", "at": 93492.5 } ], "sources_remaining": 0, "passed": true }, "off_setting_persists": true, "mobile_controls_accessible": true, "errors": [], "source_unchanged": true }