diff --git a/docs/reference-matrix.md b/docs/reference-matrix.md index d701e4c6..df895df0 100644 --- a/docs/reference-matrix.md +++ b/docs/reference-matrix.md @@ -447,6 +447,7 @@ the candidate mechanisms. | --- | --- | --- | | [Issue #60 report](https://github.com/InferMatrixLab/InferRSI/issues/60#issuecomment-6310920198), 2026-10-21; eight Git source inventories in the [owner metadata](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json) | Task/quality/workload separation across text, agentic coding, work agents, TTS, full-duplex interaction and image/video; selected pinned AIPerf/realtime/FDB/video sources rechecked and the current InferRSI metric contract inspected | Extend [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) with [task selection](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/task-benchmark-selection.md) and [command templates](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/benchmark-commands.md). Candidate guidance; source review and bounded smoke do not qualify all tasks or establish a performance gain. Replay scores, reference parity and task quality retain separate meanings. | | [NVIDIA/NeMo-Speech.cpp @ 20999cc9](https://github.com/NVIDIA/NeMo-Speech.cpp/tree/20999cc921d2b14aa58708f61756a8b2e787ba20), inspected 2026-11-01 | Selected cache-aware encoder/config, runner, SDK bounds, endpoint policy, geometry test/CMake and ASR benchmark; complete 14-file readset and Apache-2.0 / parakeet MIT attribution retained in the reference | Extend [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) with [cache-aware ASR stream boundaries](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/cache-aware-asr-stream-boundaries.md): first/steady mel geometry and decoded clock, stream-slot reset/reuse, EOU/EOF/reset and model-test SKIP boundaries. Source-derived candidate / empirical not_run; benchmark figures remain author-reported with precision, FE and trial limits. No model/device/quality/long-audio/load/service qualification. | +| [yuekaizhang/faster-cosyvoice @ f1049c59](https://github.com/yuekaizhang/faster-cosyvoice/tree/f1049c59196c2c1a890079abc56056b43f27cd1d), inspected 2026-11-03; overall source rights UNKNOWN, selected file headers retain Apache-2.0 and unresolved inherited attribution | Bounded PCM client/server-credit, chunk planner, session/worker, Token2Wav and timing/quality-report inspection; [exact readset and per-file rights](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/faster-cosyvoice-source.json) | Extend [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) with [dependency-aware TTS playout](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/dependency-aware-tts-playout.md): route-origin credit versus client zero-buffer playback, same-session Mel dependencies, cumulative vocoding and partial-cost/token-derived denominators. Original analysis with immutable links only; candidate / empirical not_run / efficacy not_evaluated. Author reports remain separate; no copied implementation/media, new runtime or hardware/quality/service qualification. | ## SGLang versioned entry selection diff --git a/docs/searchspace-matrix.md b/docs/searchspace-matrix.md index 747a43b4..6d102b20 100644 --- a/docs/searchspace-matrix.md +++ b/docs/searchspace-matrix.md @@ -640,3 +640,16 @@ NOASSERTION and InferRSI's repository license. | Resource | Stage / scoped question | Canonical owner | Evidence boundary | | --- | --- | --- | --- | | SSD / Storage + RAM / Host | Diagnose / Verify: independent tensor-byte identity after aligned/coalesced reads, resident versus pending slot lifetime, ordered consumer accumulation and actual oracle callback coverage | [weight-loading-optimization](../src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md) ([expert-cache oracle reference](../src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-oracles.md)) and [source record](../src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-source.json); DeepSeek source at `d0402056d0da4bf7edfd2d4977de479a0db9d201` | Source-derived candidate / empirical not_run; byte oracle shares tensor metadata, resident snapshot covers w1, and tiny PyTorch/batched checks are source definitions. No measured resource gain, full-model quality, native/device execution or agent efficacy is established. | + +## Faster CosyVoice evaluation source context + +The existing [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) +uses [dependency-aware TTS playout](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/dependency-aware-tts-playout.md) +and its [source record](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/faster-cosyvoice-source.json) +as methodology/source context for server PCM credit, client playback +observations, chunk dependencies and partial cost denominators. This adds no +numbered resource-layer mapping, lifecycle-stage mapping, supplementary +hardware coverage or hardware prerequisite. The owner's existing N/A reason +remains unchanged; source use of GPU, TensorRT, Flow or HiFT does not alter +that boundary. Source-derived candidate / empirical not_run / efficacy +not_evaluated; author reports are not local quality or service results. diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md index 8fb7558b..961d26b6 100644 --- a/docs/skill-coverage.md +++ b/docs/skill-coverage.md @@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained matrices. **Do not edit by hand**; regenerate after any registry, skill, matrix or frozen-plan change. -- Input digest: `sha256:e26cf7b1c819cfe6381ac0cba56b3bfe1efca71f8fa97453d110c47602fe954b` +- Input digest: `sha256:05eed1b67b40115092850127c608e2a63978d50cf5ea4c9117671e9662a56321` - Regenerate: `python tools/build_skill_coverage.py` - Verify without writing: `python tools/build_skill_coverage.py --check` @@ -132,7 +132,7 @@ optimization is fully covered or that any procedure has measured benefit. | [choose-parallel-training-plan](../src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md) | common | Compare stage, device-mesh and operator-sharding candidates for static distributed training while accounting for ... | Comparing stage, device-mesh and operator-sharding candidates for static distributed ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Alpa — [2201.12023v3](https://arxiv.org/abs/2201.12023v3)) | candidate · not_run · 1 readset(s) | no-eval | | [dense-performance-evidence](../src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md) | common | Taxonomy of performance evidence an optimization agent needs, ordered by collection cost and diagnostic power. Maps ... | Deciding which performance evidence to collect next and which claim each evidence level ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Observe / Profile / Verify
2 RAM/Host · Verify
4 Network · Verify | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 0 readset(s) | — | | [evaluate-elastic-serving-placement](../src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md) | common | Compare migrating an LLM request between model replicas with changing its sequence-parallel execution group, including ... | Comparing migration of a running LLM request between compatible model replicas with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify
2 RAM/Host · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Llumnix — [2406.03243v1](https://arxiv.org/abs/2406.03243v1)) | candidate · not_run · 1 readset(s) | no-eval | -| [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) | common | Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents ... | Selecting task quality gates, reference parity checks and performance workloads for text ... | failure_modes, input_contract, produced_artifact, refusal_point
7 advice | N/A: SKILL.md §§ Select the task and its evaluation contract, Statistical requirements, Shared-machine measurement discipline, and Search integrity and final-state verification design comparable evaluation; hardware identity and allocation isolation are comparison controls, not a hardware-resource mechanism. | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references ([Neroued/ninfer](https://github.com/Neroued/ninfer/tree/1882c6adbc9804 ...)
2 · Methodology and corpus references ([Xe-Forge repo + paper v1](https://github.com/IntelLabs/Xe-Forge/tree/ ...)
Task-specific benchmark selection (issue #60) ([Issue #60 report](https://github.com/InferMatrixLab/InferRSI/issues/6 ...)
Task-specific benchmark selection (issue #60) ([NVIDIA/NeMo-Speech.cpp @ 20999cc9](https://github.com/NVIDIA/NeMo-Spe ...) | candidate · not_run · 11 readset(s) | — | +| [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) | common | Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents ... | Selecting task quality gates, reference parity checks and performance workloads for text ... | failure_modes, input_contract, produced_artifact, refusal_point
7 advice | N/A: SKILL.md §§ Select the task and its evaluation contract, Statistical requirements, Shared-machine measurement discipline, and Search integrity and final-state verification design comparable evaluation; hardware identity and allocation isolation are comparison controls, not a hardware-resource mechanism. | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references ([Neroued/ninfer](https://github.com/Neroued/ninfer/tree/1882c6adbc9804 ...)
2 · Methodology and corpus references ([Xe-Forge repo + paper v1](https://github.com/IntelLabs/Xe-Forge/tree/ ...)
Task-specific benchmark selection (issue #60) ([Issue #60 report](https://github.com/InferMatrixLab/InferRSI/issues/6 ...)
Task-specific benchmark selection (issue #60) ([NVIDIA/NeMo-Speech.cpp @ 20999cc9](https://github.com/NVIDIA/NeMo-Spe ...)
Task-specific benchmark selection (issue #60) ([yuekaizhang/faster-cosyvoice @ f1049c59](https://github.com/yuekaizha ...) | candidate · not_run · 13 readset(s) | — | | [inference-optimization-casebook](../src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md) | common | How to structure and use a casebook of past optimization decisions as a decision-support tool. Extracts the casebook ... | Structuring and querying a casebook of past optimization decisions as decision support ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Optimize | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-eval | | [kernel-writing-guide](../src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md) | common | Decision guide for writing custom inference kernels: when to write one, which framework to choose (Triton, CuTe DSL ... | Deciding whether to write a custom inference kernel, routing a kernel pattern to Triton ... | failure_modes, input_contract, produced_artifact, refusal_point
6 advice | 1 Accelerator · Optimize / Diagnose / Configure / Verify | 1 · Systems sources (Dao-AILab/causal-conv1d @ `cd81f0413cad2fc1e6f17e785ac39f59aae690cd` ...)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (huggingface/transformers @ `0dad7b822255a0ae261ec45ae937371e859ffd1a` ...)
1 · Systems sources (mindspore-ai/akg @ `5aa15f3fe4d856eb60cd3b27447e04f5f1fb4dff` ...)
1 · Systems sources (sgl-project/sglang @ `16176278d0180f1a1a578e53e6eef432577876c2` ...)
1 · Systems sources (vllm-project/vllm @ `537a79f0fc5ba564d2bef43d54915b2f80f9f77d` (root ...) | candidate · not_run · 7 readset(s) | no-eval | | [optimization-search-policy](../src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md) | common | Policy for how an optimization agent should select, order, and stop optimization experiments. Covers candidate ... | Selecting, ordering and stopping optimization experiments: candidate generation ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Optimize | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (RRSI — arXiv 2609.24972v3)
2 · Methodology and corpus references (arXiv 2608.01867 (CRISP))
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 9 readset(s) | no-eval | diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/_provenance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/_provenance.json index e649e78f..392adf36 100644 --- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/_provenance.json +++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/_provenance.json @@ -1,7 +1,7 @@ { "schema_version": 1, "repository": "https://github.com/InferMatrixLab/InferRSI", - "source_revision": "56c500367d5495fdd9f0e35030c9cd957f70eac5", + "source_revision": "b2d62f8192903ab31f8e8828f4e630a1bec776b7", "registry": { "path": "src/inferrsi/data/skill_library/skill-index.json", "sha256": "7904275c3383380db179287171e1430617aaa9ac4f439291eff8a71426987fbb" diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json index 24a44193..08d697df 100644 --- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json +++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json @@ -45,12 +45,14 @@ }, "checkout_ref": "checkout:inference-benchmark-design", "source_files": { - "src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md": "9deb10a23233949f8edc7673c77932f7a4871c9c8b3e978147eef100418792a5", - "src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json": "b9f8de0cdc8676f1fa9cb119125b2238fb7c377687222d2bfab1b37f8794dbe1", + "src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md": "fa7d91bf2b1c38299a7b3743cc362cb0399c4874d0cc42618161b77238e83a7a", + "src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json": "061e07be83be10623b0500ef067709d05dd15fdd2c9d9aeb1e4404ef09fd6ca2", "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/autotunebench-source.json": "bc343ece2bf40c4e8aa4bd8d84f32465d00633255557feda5fe7799ce24e8688", "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/autotunebench-trial-validity.md": "cb3a7a5d94d056f868af07f92a16a6fef31eafd3199ed07ae79e93798e320b1d", "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/benchmark-commands.md": "0898424e0bff3a6246e1eeeda6abd6c5bd974c047fe32f3bbd5ab9d31ce17cd3", "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/cache-aware-asr-stream-boundaries.md": "4e0ea9c1bf6b096a49a91ad5266e90b8a456f4f5aca15eba8ac270a53c2309b5", + "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/dependency-aware-tts-playout.md": "e94866e718adc3bd7e272a4856e9a90b6f27cbc0ff3a44101e68e6995ed79e2c", + "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/faster-cosyvoice-source.json": "9efe8918208adddfb9a31d2abed080923a5ae0dfb6332438179af0de66f54e13", "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/justgoal-evaluation-contracts.md": "d25df795fd301e8a1bb436e5f8122d0bac2441c43104e59a3b310fd5a64f2f3e", "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/justgoal-source.json": "60f2784384ea1a3b08a2ecafb60207d8e3707247dc7bc177b95b463d81f1cc16", "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/ninfer-source.json": "ccda0fc6781e3c4dceb41521ab952445840b4afcc9f5a7d968857f76e8db6272", @@ -391,6 +393,49 @@ "readset": "references/ninfer-source.json", "license": "Apache-2.0", "scope": "Selected Qwen3.5 GDN/MTP state-restoration, accepted-prefix and completion evidence contracts; historical results retain their separate reported revision. Source-only, no native execution." + }, + { + "type": "git", + "url": "https://github.com/yuekaizhang/faster-cosyvoice", + "commit": "f1049c59196c2c1a890079abc56056b43f27cd1d", + "paths": [ + "README.md", + "README_zh.md", + "docs/benchmark.md", + "docs/benchmark_zh.md", + "docs/usage.md", + "docs/usage_zh.md", + "examples/offline_inference.py", + "examples/stream_client.py", + "faster_cosyvoice/config/settings.py", + "faster_cosyvoice/llm/engine.py", + "faster_cosyvoice/server/app.py", + "faster_cosyvoice/server/arguments.py", + "faster_cosyvoice/server/audio_encode.py", + "faster_cosyvoice/server/leading_silence.py", + "faster_cosyvoice/server/openai_speech.py", + "faster_cosyvoice/server/protocol.py", + "faster_cosyvoice/streaming/chunker.py", + "faster_cosyvoice/streaming/session.py", + "faster_cosyvoice/streaming/token2wav_batcher.py", + "faster_cosyvoice/streaming/token2wav_scheduler.py", + "faster_cosyvoice/token2wav/builders.py", + "faster_cosyvoice/token2wav/cosyvoice/flow/DiT/modules.py", + "faster_cosyvoice/token2wav/cosyvoice/flow/flow.py", + "faster_cosyvoice/token2wav/cosyvoice/hifigan/generator.py", + "faster_cosyvoice/token2wav/flashinfer_dit.py", + "faster_cosyvoice/token2wav/flashinfer_flow.py", + "faster_cosyvoice/token2wav/frontend.py", + "faster_cosyvoice/token2wav/matcha_audio.py", + "faster_cosyvoice/token2wav/token2wav.py", + "pyproject.toml", + "tests/test_session.py", + "tests/test_stream_client.py", + "tests/test_token2wav_batcher.py" + ], + "readset": "references/faster-cosyvoice-source.json", + "license": "UNKNOWN overall; selected file-level Apache-2.0 headers and unresolved inherited attribution retained", + "scope": "Selected PCM/playout, routed-credit, chunk-dependency and timing-denominator source audit; original analysis and immutable links only, no code/media copying or native/model/quality/service execution" } ], "empirical_validation": "not_run", @@ -437,7 +482,8 @@ "JustGoal contributes source-only controlled-evaluation and kernel-acceptance guidance; its repaired/retested timings, released-pass counts and protocol differences are author-reported, not locally reproduced or a complete campaign runner.", "The AutoTuneBench reference is source-only: original operational checkpoint/prompt receipts remain unavailable, source contradictions remain explicit, and no tuning, device/performance or agent-efficacy validation was run.", "The Xe-Forge reference is source-only and limited to inspected host/manual/Claude trial paths; named-verdict update predicates and caller defaults remain explicit. Native tests, backend/model/quality/performance/service behavior and agent efficacy were not evaluated, and source/package checks do not promote whole-model support.", - "The NInfer reference inspects selected Qwen3.5 GDN/MTP source contracts and unexecuted test definitions; its historical author measurements use a different engine revision. Native state/model/device/quality/performance and efficacy remain unvalidated." + "The NInfer reference inspects selected Qwen3.5 GDN/MTP source contracts and unexecuted test definitions; its historical author measurements use a different engine revision. Native state/model/device/quality/performance and efficacy remain unvalidated.", + "The Faster CosyVoice reference is source-only: server route credit and client simulated playback remain distinct, reported costs/quality retain their boundaries, overall source rights are UNKNOWN, and native/model/device/audio quality/service/performance are not_run with efficacy not_evaluated." ], "provenance_readsets": [ "references/task-benchmark-selection.md", @@ -450,7 +496,9 @@ "references/xeforge-trial-evidence.md", "references/xeforge-source.json", "references/ninfer-state-and-completion-contracts.md", - "references/ninfer-source.json" + "references/ninfer-source.json", + "references/dependency-aware-tts-playout.md", + "references/faster-cosyvoice-source.json" ], "references": [ "../inference-optimization-casebook/cases/admission-drain-occupancy.md", @@ -508,7 +556,8 @@ "JustGoal contributes source-only controlled-evaluation and kernel-acceptance guidance; its repaired/retested timings, released-pass counts and protocol differences are author-reported, not locally reproduced or a complete campaign runner.", "The AutoTuneBench reference is source-only: original operational checkpoint/prompt receipts remain unavailable, source contradictions remain explicit, and no tuning, device/performance or agent-efficacy validation was run.", "The Xe-Forge reference is source-only and limited to inspected host/manual/Claude trial paths; named-verdict update predicates and caller defaults remain explicit. Native tests, backend/model/quality/performance/service behavior and agent efficacy were not evaluated, and source/package checks do not promote whole-model support.", - "The NInfer reference inspects selected Qwen3.5 GDN/MTP source contracts and unexecuted test definitions; its historical author measurements use a different engine revision. Native state/model/device/quality/performance and efficacy remain unvalidated." + "The NInfer reference inspects selected Qwen3.5 GDN/MTP source contracts and unexecuted test definitions; its historical author measurements use a different engine revision. Native state/model/device/quality/performance and efficacy remain unvalidated.", + "The Faster CosyVoice reference is source-only: server route credit and client simulated playback remain distinct, reported costs/quality retain their boundaries, overall source rights are UNKNOWN, and native/model/device/audio quality/service/performance are not_run with efficacy not_evaluated." ], "verification": { "locally_verified": false, diff --git a/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md b/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md index fb899bdf..03369757 100644 --- a/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md +++ b/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md @@ -24,6 +24,12 @@ before selecting tests or comparing chunk latency. It separates startup/steady g progress, stream-slot reuse, reset, EOU and EOF; its source tests and author benchmark reports are not local model or serving qualification. +For streaming TTS measurements tied to playback deadlines or dependent chunks, +read [dependency-aware TTS playout](references/dependency-aware-tts-playout.md). +It traces server PCM credit, client playback events, independent batching work +and partial pipeline costs at one pinned source; it adds no hardware coverage, +runtime or locally reproduced quality/performance result. + For treatment-enabled tuning comparisons or claimed protocol integrity, read [treatment delivery and trial validity](references/autotunebench-trial-validity.md). Trace feedback to the eligible consumer input and apply the frozen arm-validity diff --git a/src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json b/src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json index 394fa62f..1060dde9 100644 --- a/src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json +++ b/src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json @@ -328,6 +328,49 @@ "readset": "references/ninfer-source.json", "license": "Apache-2.0", "scope": "Selected Qwen3.5 GDN/MTP state-restoration, accepted-prefix and completion evidence contracts; historical results retain their separate reported revision. Source-only, no native execution." + }, + { + "type": "git", + "url": "https://github.com/yuekaizhang/faster-cosyvoice", + "commit": "f1049c59196c2c1a890079abc56056b43f27cd1d", + "paths": [ + "README.md", + "README_zh.md", + "docs/benchmark.md", + "docs/benchmark_zh.md", + "docs/usage.md", + "docs/usage_zh.md", + "examples/offline_inference.py", + "examples/stream_client.py", + "faster_cosyvoice/config/settings.py", + "faster_cosyvoice/llm/engine.py", + "faster_cosyvoice/server/app.py", + "faster_cosyvoice/server/arguments.py", + "faster_cosyvoice/server/audio_encode.py", + "faster_cosyvoice/server/leading_silence.py", + "faster_cosyvoice/server/openai_speech.py", + "faster_cosyvoice/server/protocol.py", + "faster_cosyvoice/streaming/chunker.py", + "faster_cosyvoice/streaming/session.py", + "faster_cosyvoice/streaming/token2wav_batcher.py", + "faster_cosyvoice/streaming/token2wav_scheduler.py", + "faster_cosyvoice/token2wav/builders.py", + "faster_cosyvoice/token2wav/cosyvoice/flow/DiT/modules.py", + "faster_cosyvoice/token2wav/cosyvoice/flow/flow.py", + "faster_cosyvoice/token2wav/cosyvoice/hifigan/generator.py", + "faster_cosyvoice/token2wav/flashinfer_dit.py", + "faster_cosyvoice/token2wav/flashinfer_flow.py", + "faster_cosyvoice/token2wav/frontend.py", + "faster_cosyvoice/token2wav/matcha_audio.py", + "faster_cosyvoice/token2wav/token2wav.py", + "pyproject.toml", + "tests/test_session.py", + "tests/test_stream_client.py", + "tests/test_token2wav_batcher.py" + ], + "readset": "references/faster-cosyvoice-source.json", + "license": "UNKNOWN overall; selected file-level Apache-2.0 headers and unresolved inherited attribution retained", + "scope": "Selected PCM/playout, routed-credit, chunk-dependency and timing-denominator source audit; original analysis and immutable links only, no code/media copying or native/model/quality/service execution" } ], "empirical_validation": "not_run", @@ -374,7 +417,8 @@ "JustGoal contributes source-only controlled-evaluation and kernel-acceptance guidance; its repaired/retested timings, released-pass counts and protocol differences are author-reported, not locally reproduced or a complete campaign runner.", "The AutoTuneBench reference is source-only: original operational checkpoint/prompt receipts remain unavailable, source contradictions remain explicit, and no tuning, device/performance or agent-efficacy validation was run.", "The Xe-Forge reference is source-only and limited to inspected host/manual/Claude trial paths; named-verdict update predicates and caller defaults remain explicit. Native tests, backend/model/quality/performance/service behavior and agent efficacy were not evaluated, and source/package checks do not promote whole-model support.", - "The NInfer reference inspects selected Qwen3.5 GDN/MTP source contracts and unexecuted test definitions; its historical author measurements use a different engine revision. Native state/model/device/quality/performance and efficacy remain unvalidated." + "The NInfer reference inspects selected Qwen3.5 GDN/MTP source contracts and unexecuted test definitions; its historical author measurements use a different engine revision. Native state/model/device/quality/performance and efficacy remain unvalidated.", + "The Faster CosyVoice reference is source-only: server route credit and client simulated playback remain distinct, reported costs/quality retain their boundaries, overall source rights are UNKNOWN, and native/model/device/audio quality/service/performance are not_run with efficacy not_evaluated." ], "provenance_readsets": [ "references/task-benchmark-selection.md", @@ -387,7 +431,9 @@ "references/xeforge-trial-evidence.md", "references/xeforge-source.json", "references/ninfer-state-and-completion-contracts.md", - "references/ninfer-source.json" + "references/ninfer-source.json", + "references/dependency-aware-tts-playout.md", + "references/faster-cosyvoice-source.json" ], "references": [ "../inference-optimization-casebook/cases/admission-drain-occupancy.md", diff --git a/src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/dependency-aware-tts-playout.md b/src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/dependency-aware-tts-playout.md new file mode 100644 index 00000000..af7c7e7f --- /dev/null +++ b/src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/dependency-aware-tts-playout.md @@ -0,0 +1,196 @@ +# Dependency-aware TTS playout + +Use this reference when designing or auditing a streaming TTS benchmark whose +chunks share model state or are scheduled against estimated playback time. +It derives from [Faster CosyVoice][source] at +`f1049c59196c2c1a890079abc56056b43f27cd1d`, inspected 2026-11-03. The +[source record](faster-cosyvoice-source.json) preserves the exact bounded +readset, file identities and rights. This is source-derived **candidate** +guidance: native/model/device/quality/service execution is **not_run** and +efficacy is **not_evaluated**. + +Keep the [parent method](../SKILL.md) and +[task benchmark selection](task-benchmark-selection.md) for quality gates, +reference parity, comparable workloads, statistics and evidence verdicts. This +reference adds the source-specific playout and cost-accounting prerequisites; +it adds no runtime, hardware prerequisite or measured optimization. + +## 1. Establish the client event and buffer policy + +Before comparing first-audio measurements, record sample rate, channel count, +sample width, encoding, leading-silence policy and the client's playback-buffer +policy. The source client requests mono, 24 kHz, little-endian PCM16. It carries +an odd byte into the next HTTP read and records an audio observation only when +at least one complete sample exists. HTTP reads can split or combine model +chunks; their boundaries are not model chunk identities. [Client][client] + +| Source field | Event or quantity | Interpretation limit | +| --- | --- | --- | +| `ttfb_ms` | First nonempty raw response-body read after request start | An odd byte can arrive before any complete PCM sample; a WAV header is a different payload again | +| `first_playable_ms` | First sample-aligned PCM observation | Receipt of complete samples, not a physical speaker measurement or a guaranteed startup buffer | +| `audible_ttfa_ms` | Detected onset placed on the simulated playback clock | Depends on both the onset detector and the playback policy; can be absent when no onset is detected | +| `underruns` | Later arrival events that exceed the simulated playback deadline | Event count, not total missing-audio duration or measured device underruns | +| `audio_s` | Complete received PCM sample count divided by sample rate | Delivered content duration; preserve it alongside latency and completion | + +The client's zero-buffer simulation starts playback at the first complete +payload arrival. Each payload extends the deadline by its audio duration. If a +later payload arrives after that deadline, the simulation records an underrun +and restarts playback at the later arrival before adding its duration. Audible +onset uses this rebased clock, rather than adding leading silence directly to +first-body time. [Playback calculation][playback] + +Retain the onset detector's parameters when comparing audible latency. This +source uses 20 ms RMS frames, 10 ms hops, per-frame DC removal, a -45 dBFS +threshold and two consecutive active frames. These are the source's detector +settings, not general quality or acceptance gates. Its tests define an onset +case and a late-arrival case; they are test definitions, not executed audio or +service evidence. [Detector][detector] · [Client test definitions][client-tests] + +If a supplied report records only HTTP headers, token events or first-body +time, keep first-playable/audible latency unknown until the required event is +available. Do not infer physical playback from this simulation. + +## 2. Separate server credit from client playback + +Trace the actual PCM bytes credited by the serving path before interpreting a +deadline scheduler. Here, each session retains its first route timestamp and +accumulates routed sample duration. The scheduler estimates its playback +deadline as that timestamp plus the accumulated duration. +[Session accounting][session] · [Deadline policy][scheduler] + +The generator marks credit immediately before yielding a nonempty encoded, +optionally trimmed PCM payload. This is the server response-generator boundary; +it is not acknowledgment of network receipt or client consumption. The server's +monotonic clock and the client's request-relative arrival/playback clock must +remain distinct. No client buffer feedback is present in these selected paths. +[PCM production and credit][route] + +The source prioritizes established streams within its configured reserve, +then streams without a first PCM payload, then established streams with the +earliest estimated deadline, with admission order resolving ties. The default +reserve is 100 ms in this source. Treat it as source context, not a universal +bar or a guarantee that a client will avoid underruns. Record actual arrival +and playback observations before claiming continuity. [Policy][scheduler] · +[CLI conversion and validation][arguments] + +When leading-silence trimming is enabled, initial output can be buffered and +released only after its bounded onset/buffer decision. Credit uses the resulting +payload length. Record trimming, pre-roll and startup-buffer settings with both +first-payload and audible measurements. Changing initial PCM content does not +establish reduced model computation or a latency benefit for every request. +[Trimmer][detector] · [Documented scope][trim] + +## 3. Identify independent batching work and accumulated cost + +Inspect the selected planner and session state before treating chunks as +independent jobs. For this model, the first non-final hop includes prompt-token +alignment padding and requires lookahead tokens; lookahead is resent without +being consumed. Later hops grow to the configured cap, and completion drains +the remaining tokens. Nominal hop size alone does not identify the delivered +PCM duration. Use actual output samples for playback credit. +[Chunk planning][planner] · [Model rates][builders] + +The serving coroutine awaits each session's submitted chunk. The packed worker +also allows at most one chunk from a given session in a batch because the next +chunk depends on that session's Mel cache. It forms batches from ready jobs +across different sessions, subject to its maximum size and scheduling order. +This is the source's dependency boundary, not proof that an arbitrary caller's +jobs or changed model are independent. [Submission][route] · [Worker][worker] + +Flow runs across the selected sessions, after which HiFT finishes each session +separately. Each finish updates accumulated Mel state, vocodes that cache and +returns only waveform samples beyond the previous speech offset. The visible +streaming path also recomputes Flow from the available token prefix. Newly +delivered audio is therefore not a measure of all work performed for that +chunk. [Token2Wav implementation][token2wav] + +When auditing a batching or chunk-size claim, retain prompt and lookahead +context, prefix length, accumulated Mel length, actual batch membership and +per-session output duration. Keep queueing, conditioning, vocoding, conversion +and delivery in the declared cost boundary. A packed Flow call does not show +batched HiFT, constant work per emitted chunk, numerical parity or a service +throughput gain. Queue tests using fake workers define selection/dependency +cases only; no native model result follows from them. [Worker test definitions][worker-tests] + +## 4. Audit the reported timing and quality boundary + +Resolve registered-voice and reference-cache state before comparing cost. A +registered voice reuses prepared conditions; a request-level reference follows +audio decoding/frontend processing and can itself hit the frontend cache. +Record which preparation actually belongs to the timed request. +[Condition resolution][condition] · [Frontend cache][frontend] + +| Source measurement or mode | Included boundary | Required interpretation | +| --- | --- | --- | +| Streaming `request_done.ttfa_ms` | Generator-local start through its first nonempty PCM payload | The streaming HTTP handler resolves conditioning before this timer starts; it excludes that earlier preparation and client delivery | +| `stream=false` HTTP response | The same incremental synthesis collected before response delivery | It changes buffering, not the backend into the separate offline static-batch implementation; internal route credit does not observe buffered-response playback | +| Offline `rtf` | Sum of LLM and Token2Wav stage wall times divided by token-derived audio seconds | Frontend time is reported separately; loading, other preparation and output delivery are not a full-pipeline denominator | +| Offline `audio_seconds` | Counted generated token IDs divided by the source's 25-token/s rate | Not measured saved/received PCM duration; the counter is updated before filtering outputs without valid speech tokens | +| Documented RTFx | Audio duration received within a measurement window divided by window duration | Author-reported aggregate throughput; do not replace it with inverse mean per-request RTF or the offline estimate | + +Source paths: [HTTP handler][handler], [generator timer][route], +[offline accounting][offline] and [reported metrics][benchmark]. Keep an absent +common timing or audio-duration boundary unresolved rather than comparing +similarly named fields. + +The author performance report uses a fixed registered voice/reference, one +NVIDIA Hopper GPU, fixed concurrency, 15 s warmup, a 60 s measurement window, +seed 0 and disabled trimming. Its concurrency-one cumulative ablations report +complete requests without underruns and equal 760 ms first chunks. Preserve +these conditions when citing the report; the exact Hopper SKU, raw timing +records and independent reproduction are not established here. +[Reported workload][reported] · [Ablation conditions][benchmark] + +The quality table reports a target-only versus DSpark comparison over 2,020 +paired Chinese Seed-TTS samples decoded with Paraformer. It does not establish +speaker similarity, perceptual quality, every cumulative optimization's quality +or another workload's result. Keep its character-error protocol and per-row +conditioning scope distinct from fixed-voice performance. The inspected release +tree contains no raw benchmark/evaluator receipts; that does not establish +whether external receipts exist. [Author quality report][quality] + +Return a benchmark contract or audit note identifying the event/buffer policy, +server-credit origin, dependent state, actual timed stages, delivered-duration +denominator, quality scope and missing evidence. If a required event, native +dispatch, output, evaluator or common boundary is missing, withhold the affected +comparison or gain claim. Use the parent method's frozen gates and verdicts for +any separately authorized campaign. + +## Source and rights limits + +The pinned tree has no repository-level license/notice file and its package +metadata declares no license, so overall source rights remain **UNKNOWN**. +Selected NVIDIA and Alibaba files have Apache-2.0 headers; those do not license +other files. The vendored HiFT generator names unpinned HiFi-GAN, +ParallelWaveGAN and BigVGAN origins, and Nari-compatible comments do not resolve +implementation lineage. Exact header scopes, unresolved inheritance and other +unlicensed files remain in the [source record](faster-cosyvoice-source.json). + +This reference contains original analysis and immutable links. It copies no +upstream implementation, patch, audio, model, dataset, paper or figure. Source +dependency/model names are inspection context, not a tested stack or permission +to install or execute it. Recheck changed source contracts before transferring +the procedure; no native, quality, performance or efficacy result is implied. + +[source]: https://github.com/yuekaizhang/faster-cosyvoice/tree/f1049c59196c2c1a890079abc56056b43f27cd1d +[client]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/examples/stream_client.py#L85-L134 +[playback]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/examples/stream_client.py#L27-L51 +[detector]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/leading_silence.py#L1-L88 +[client-tests]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/tests/test_stream_client.py#L6-L22 +[session]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/streaming/session.py#L17-L37 +[scheduler]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/streaming/token2wav_scheduler.py#L46-L122 +[route]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/openai_speech.py#L52-L128 +[arguments]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/arguments.py#L273-L336 +[trim]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/docs/usage.md#L198-L202 +[planner]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/streaming/chunker.py#L20-L55 +[builders]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/builders.py#L16-L50 +[worker]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/streaming/token2wav_batcher.py#L79-L183 +[token2wav]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/token2wav.py#L153-L226 +[worker-tests]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/tests/test_token2wav_batcher.py#L57-L320 +[condition]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/openai_speech.py#L31-L49 +[frontend]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/frontend.py#L91-L139 +[handler]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/app.py#L143-L173 +[offline]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/examples/offline_inference.py#L240-L308 +[benchmark]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/docs/benchmark.md#L9-L49 +[reported]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/README.md#L120-L135 +[quality]: https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/docs/benchmark.md#L74-L84 diff --git a/src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/faster-cosyvoice-source.json b/src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/faster-cosyvoice-source.json new file mode 100644 index 00000000..d250c092 --- /dev/null +++ b/src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/faster-cosyvoice-source.json @@ -0,0 +1,776 @@ +{ + "schema_version": 1, + "record_kind": "source-readset", + "source": { + "type": "git", + "url": "https://github.com/yuekaizhang/faster-cosyvoice", + "repository": "yuekaizhang/faster-cosyvoice", + "commit": "f1049c59196c2c1a890079abc56056b43f27cd1d", + "tree": "a5950f9eaa500141d08c29e29dc24e8d1f313f00", + "tree_entries": 117, + "truncated": false + }, + "inspected_at": "2026-11-03", + "status": "candidate", + "empirical_validation": "not_run", + "efficacy": "not_evaluated", + "scope": "Bounded source-only interpretation of PCM receipt/playout, server route credit, dependency-limited cross-session batching and partial pipeline/audio denominators; author performance and DSpark quality reports remain separate from execution.", + "access_and_identity": "Read-only GitHub connector commit/tree/file calls, with per-file blob identity checked against the complete pinned tree", + "rights": { + "repository_license": "UNKNOWN", + "pinned_tree_license_notice_paths": [], + "pyproject_license_declaration": "absent", + "no_default_branch_license_inference": true, + "file_headers_are_not_repository_license": true, + "redistribution_boundary": "Only original explanatory analysis and immutable links are proposed. No source implementation, patches, audio, models, datasets, papers, screenshots or figures are copied or adapted. Unpinned inherited provenance remains unresolved." + }, + "readset": [ + { + "path": "README.md", + "git_blob_sha": "131c56b961e0cb045997df712ffe314cdf4ef957", + "tree_size_bytes": 5675, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 147 + ] + ], + "role": "Source overview, reported fixed-registered-voice benchmark and acknowledgments", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/README.md", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "Names Nari or vLLM-derived/compatible behavior without pinned source lineage; this does not grant copying rights" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "README_zh.md", + "git_blob_sha": "a93f11ca7ac90ea736af78ffc12b3f83a8225fe2", + "tree_size_bytes": 5398, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 36 + ], + [ + 124, + 124 + ] + ], + "role": "Bounded Chinese counterpart overview and attribution cross-check", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/README_zh.md", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "Names Nari or vLLM-derived/compatible behavior without pinned source lineage; this does not grant copying rights" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "docs/benchmark.md", + "git_blob_sha": "a4d127435287776df29b90019dd6878a0e8e209b", + "tree_size_bytes": 3574, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 84 + ] + ], + "role": "Author-reported metric, workload, cumulative ablation and DSpark-only CER scope", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/docs/benchmark.md", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "docs/benchmark_zh.md", + "git_blob_sha": "8c9d1c9fa5a48173b14628ad0849bdc975c1abe0", + "tree_size_bytes": 3161, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 76 + ] + ], + "role": "Chinese counterpart metric/workload/quality cross-check", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/docs/benchmark_zh.md", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "docs/usage.md", + "git_blob_sha": "46cf5bb01c7d6bad93219866a5050b4039e21c76", + "tree_size_bytes": 8103, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 228 + ] + ], + "role": "Documented API modes, conditioning cache, chunk/graph/trim scopes and CLI defaults", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/docs/usage.md", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "docs/usage_zh.md", + "git_blob_sha": "328a4b2c587415a4dd733a22b0a13b08e9a0c3c2", + "tree_size_bytes": 7498, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 78, + 95 + ], + [ + 160, + 205 + ] + ], + "role": "Bounded Chinese counterpart API, graph-duration and trim cross-check", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/docs/usage_zh.md", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "examples/offline_inference.py", + "git_blob_sha": "ab98b25448cd0d61f76439799dab5947b850cb86", + "tree_size_bytes": 14364, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 100 + ], + [ + 150, + 321 + ] + ], + "role": "Selected per-row provenance, stage timing and token-derived audio/RTF denominator; loader/CLI lines 101-149 excluded", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/examples/offline_inference.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "examples/stream_client.py", + "git_blob_sha": "491c8e9e0de7d619839c01a274fe73c0b27ae243", + "tree_size_bytes": 4833, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 142 + ] + ], + "role": "PCM16 sample reconstruction, client event clocks and zero-buffer playback diagnostics", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/examples/stream_client.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "Names Nari or vLLM-derived/compatible behavior without pinned source lineage; this does not grant copying rights" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/config/settings.py", + "git_blob_sha": "fa5a4fb7e5936deeb94ea295602b014ee9574c62", + "tree_size_bytes": 3127, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 79 + ] + ], + "role": "Typed source defaults and optional feature scopes", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/config/settings.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/llm/engine.py", + "git_blob_sha": "8ecae2c648f6709e3163ca6d0bc94a48d60903ae", + "tree_size_bytes": 5294, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 14, + 141 + ] + ], + "role": "Selected draft configuration dependency, token deltas and source length-stop warning; no upstream vLLM runtime validation", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/llm/engine.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/server/app.py", + "git_blob_sha": "91a917ba12f8d45a5457eeed74b94b9b39a01bb5", + "tree_size_bytes": 8000, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 212 + ] + ], + "role": "Initialization and shared objects, condition resolution, streaming/buffered response and voice-cache paths", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/app.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/server/arguments.py", + "git_blob_sha": "126514f5c3a7df3a93900201e21a68482b631aeb", + "tree_size_bytes": 11598, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 115 + ], + [ + 195, + 336 + ] + ], + "role": "Selected effective CLI defaults, estimator/batch validation and parameter-to-runtime conversion", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/arguments.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/server/audio_encode.py", + "git_blob_sha": "76f7dd30ea10beb7a4a1b838002aae692c888706", + "tree_size_bytes": 970, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 36 + ] + ], + "role": "PCM16 encoding and header-only WAV prefix boundary", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/audio_encode.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "Names Nari or vLLM-derived/compatible behavior without pinned source lineage; this does not grant copying rights" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/server/leading_silence.py", + "git_blob_sha": "ee2303e377f96dbbcd4803a038b1e04d18a78d7e", + "tree_size_bytes": 3741, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 88 + ] + ], + "role": "Declared onset rule and bounded content buffering/trimming; attributed Nari origin not independently source-matched", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/leading_silence.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "Names Nari or vLLM-derived/compatible behavior without pinned source lineage; this does not grant copying rights" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/server/openai_speech.py", + "git_blob_sha": "6f937211c5327c19de847943255c9dd40adda214", + "tree_size_bytes": 5751, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 137 + ] + ], + "role": "Condition/prompt/token/chunk pipeline and routed-PCM credit publication", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/openai_speech.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/server/protocol.py", + "git_blob_sha": "0abd7892098b17c0e1ae323a765b29e6176fd6df", + "tree_size_bytes": 3222, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 98 + ] + ], + "role": "Request format/mode fields and reference audio decoding boundary", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/server/protocol.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "Names Nari or vLLM-derived/compatible behavior without pinned source lineage; this does not grant copying rights" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/streaming/chunker.py", + "git_blob_sha": "c0d202f1d8fe577bf4cfc83946c4090d9eb97234", + "tree_size_bytes": 2255, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 55 + ] + ], + "role": "Initial prompt-alignment padding, lookahead, growing hops and final remainder", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/streaming/chunker.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/streaming/session.py", + "git_blob_sha": "7bbb690e1e59fff0ae0a1e5315a0aeaab6a00205", + "tree_size_bytes": 1351, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 37 + ] + ], + "role": "Session-owned Mel/audio state and route-origin audio credit", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/streaming/session.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "Names Nari or vLLM-derived/compatible behavior without pinned source lineage; this does not grant copying rights" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/streaming/token2wav_batcher.py", + "git_blob_sha": "12dccc32558f84c34ecd440943ff46baae2947e5", + "tree_size_bytes": 7552, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 183 + ] + ], + "role": "Ready-job selection, one GPU thread, one chunk per session in packed batches and pending/failure behavior", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/streaming/token2wav_batcher.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/streaming/token2wav_scheduler.py", + "git_blob_sha": "4324eac7bfaaa5fc125c695339c05d1ef7ea209b", + "tree_size_bytes": 4553, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 137 + ] + ], + "role": "Server-estimated playback deadline and urgent/startup/earliest/FIFO queue ordering", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/streaming/token2wav_scheduler.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/token2wav/builders.py", + "git_blob_sha": "9d774feefa3f8bcb9d6b61c4aba96894d0c4040f", + "tree_size_bytes": 2491, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 50 + ] + ], + "role": "Model-local token/Mel/sample rates and component dependency identities", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/builders.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/token2wav/cosyvoice/flow/DiT/modules.py", + "git_blob_sha": "be8caecb8a2acd290f4f6e7e1d756dff67f78e5b", + "tree_size_bytes": 20551, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": { + "first_line": 1, + "last_line": 55 + }, + "inspected_ranges": [ + [ + 1, + 55 + ] + ], + "role": "Header/import scan only; source lineage and whole-file rights unresolved", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/cosyvoice/flow/DiT/modules.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/token2wav/cosyvoice/flow/flow.py", + "git_blob_sha": "a00e2d1a7ff6fc6cf7627f64fabf544ff54b47e4", + "tree_size_bytes": 19346, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": { + "first_line": 1, + "last_line": 45 + }, + "inspected_ranges": [ + [ + 1, + 45 + ] + ], + "role": "File-level rights/header and dependency declaration only; algorithm body excluded", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/cosyvoice/flow/flow.py", + "rights": { + "license": "Apache-2.0 file header observed", + "copyright": "2024 Alibaba Inc; authors Xiang Lyu, Zhihao Du", + "exception": "Only lines 1-45 inspected; deeper inherited material and modifications not separately rights-audited" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/token2wav/cosyvoice/hifigan/generator.py", + "git_blob_sha": "53344d78dfbfded36da1f79f12d886e14a659aa0", + "tree_size_bytes": 30935, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": { + "first_line": 1, + "last_line": 45 + }, + "inspected_ranges": [ + [ + 1, + 45 + ] + ], + "role": "File-level rights and unpinned HiFi-GAN/ParallelWaveGAN/BigVGAN attribution only; algorithm body excluded", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/cosyvoice/hifigan/generator.py", + "rights": { + "license": "Apache-2.0 file header observed", + "copyright": "2024 Alibaba Inc; authors Xiang Lyu, Kai Hu", + "exception": "Lines 39-41 identify modifications from HiFi-GAN, ParallelWaveGAN and BigVGAN without pins; those inherited rights are unresolved, not replaced by the outer header" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/token2wav/flashinfer_dit.py", + "git_blob_sha": "2fc3bb419dd421d21d1f5a892bffbf1602c5f09c", + "tree_size_bytes": 39248, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": { + "first_line": 1, + "last_line": 95 + }, + "inspected_ranges": [ + [ + 1, + 95 + ] + ], + "role": "Header, imports and class declarations only; no kernel, graph dispatch or mask-correctness audit", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/flashinfer_dit.py", + "rights": { + "license": "Apache-2.0 file header observed", + "copyright": "2026 NVIDIA CORPORATION", + "exception": "Only lines 1-95 inspected; no repository-wide license or full body/lineage audit" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/token2wav/flashinfer_flow.py", + "git_blob_sha": "a51f2991e7960d2b70fa4e2296f2dc2cdb321acb", + "tree_size_bytes": 11048, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 110 + ], + [ + 207, + 319 + ] + ], + "role": "Header and selected packed streaming interface/returned per-request Mel lengths; middle helpers excluded", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/flashinfer_flow.py", + "rights": { + "license": "Apache-2.0 file header observed", + "copyright": "2026 NVIDIA CORPORATION", + "exception": "Copyright/license notice search across complete retrieved file found only the header; algorithm review limited to declared ranges" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/token2wav/frontend.py", + "git_blob_sha": "7df23a44301baad34a47b45d74c5db8742cfced4", + "tree_size_bytes": 6021, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 146 + ] + ], + "role": "Reference token/features/speaker conditions, cache-hit and per-request preparation cost", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/frontend.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/token2wav/matcha_audio.py", + "git_blob_sha": "d257f0d652822851714c47826dc4b1b69fb09fcd", + "tree_size_bytes": 2311, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": { + "first_line": 1, + "last_line": 45 + }, + "inspected_ranges": [ + [ + 1, + 45 + ] + ], + "role": "Header/import scan only; whole-file rights/lineage unresolved", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/matcha_audio.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "faster_cosyvoice/token2wav/token2wav.py", + "git_blob_sha": "7afdc1c6dd002d0c5757bbadeec880644c8b9e34", + "tree_size_bytes": 11040, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 226 + ] + ], + "role": "Single/packed Flow execution boundary, cumulative Mel vocoding and newly emitted CPU waveform slice", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/faster_cosyvoice/token2wav/token2wav.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "pyproject.toml", + "git_blob_sha": "5a3d8a1ced57a487bdb1b25a30d42ed852711c18", + "tree_size_bytes": 1690, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 59 + ] + ], + "role": "Dependency/build/source revision constraints and absent project-license declaration", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/pyproject.toml", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "tests/test_session.py", + "git_blob_sha": "9623c8c473542e3223830a288d0cf6d33314eaf8", + "tree_size_bytes": 710, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 18 + ] + ], + "role": "Source test definition for cumulative routed-sample accounting, not executed", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/tests/test_session.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "tests/test_stream_client.py", + "git_blob_sha": "235ae7f74e20284ff9de7b79c152f3263ff87889", + "tree_size_bytes": 848, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 22 + ] + ], + "role": "Source test definitions for audible onset and one late arrival, not executed", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/tests/test_stream_client.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + }, + { + "path": "tests/test_token2wav_batcher.py", + "git_blob_sha": "80f3fb580a947929bace3a32fb109dd5c65eb031", + "tree_size_bytes": 10630, + "identity_matches_complete_pinned_tree": true, + "retrieval_scope": "complete_file", + "inspected_ranges": [ + [ + 1, + 320 + ] + ], + "role": "Source test definitions using fake workers for queueing/dependency/cancellation/error contracts, not executed or native qualification", + "immutable_url": "https://github.com/yuekaizhang/faster-cosyvoice/blob/f1049c59196c2c1a890079abc56056b43f27cd1d/tests/test_token2wav_batcher.py", + "rights": { + "license": "UNKNOWN", + "copyright": "No explicit copyright notice in retrieved material", + "exception": "No repository-level license at the pinned tree; no license inference from dependencies, file path or default branch" + }, + "copy_action": "reference_only; original analysis; no upstream code or media copied" + } + ], + "exclusions": [ + "All docs/assets SVG bodies and any chart extraction/reproduction", + "uv.lock dependency body (tree identity only)", + "GPU/native test bodies; no upstream test execution", + "Actual CUDA, FlashInfer, Triton, vLLM fork execution and numerical/dispatch qualification", + "Target/draft/Flow/HiFT/model downloads, audio/data, speaker verification, listening or ASR evaluation", + "Uninspected algorithm portions of vendor files and FlashInfer DiT; no copied implementation or presumed inherited rights", + "Repositories outside this source, including dependencies and acknowledged implementations; their names do not establish inspected lineage or inherited copying rights" + ], + "evidence_limits": { + "source_observation": "Selected source text and code contracts only", + "deduction": "Measurement/ownership implications are labeled separately", + "author_report": "Benchmark and DSpark CER report only", + "cpu_execution": "not_run", + "device_execution": "not_run", + "quality": "not_run", + "service_performance": "not_run", + "efficacy": "not_evaluated" + } +}