|
6 | 6 | "source_path": "src/inferrsi/data/skill_library/en/strata/sub-skills/tune-execution/SKILL.md", |
7 | 7 | "content_sha256": "cdf51c537e7237296d52ef6cda9f2e4c9d9ce3fc87265a369cd9eb7497c53512", |
8 | 8 | "lesson": "Use resolved settings, controlled one-plane experiments, repeated service measurements and explicit regressions.", |
9 | | - "application": "Separate MPS, encoder capture and microbatch controls; retest MPS after speaker extraction changes the bottleneck. No Strata speedup is imported." |
| 9 | + "application": "Separate MPS, encoder capture and microbatch controls; retest MPS after speaker extraction changes the bottleneck. No Strata speedup is imported.", |
| 10 | + "source_revision": "5b9d269", |
| 11 | + "source_url": "https://github.com/InferMatrixLab/InferRSI/blob/5b9d269/src/inferrsi/data/skill_library/en/strata/sub-skills/tune-execution/SKILL.md", |
| 12 | + "current_path": "src/inferrsi/data/skill_library/repositories/strata/sub-skills/tune-execution/SKILL.md" |
10 | 13 | }, |
11 | 14 | { |
12 | 15 | "knowledge_scope": "vllm_omni", |
13 | 16 | "source_path": "src/inferrsi/data/skill_library/en/vllm-omni/sub-skills/omni-validation/SKILL.md", |
14 | 17 | "content_sha256": "81c2a8cbdccaa86ff560311ec9d33ba191ecc6e3121ca4e9f751f006875223c9", |
15 | 18 | "lesson": "Escalate native tests through accelerator parity, real speech quality and matched service performance.", |
16 | | - "application": "67 CPU regressions, real CAM++ numerical checks, repeated service results and independent speech gates. Component PASS never overrides speech FAIL." |
| 19 | + "application": "67 CPU regressions, real CAM++ numerical checks, repeated service results and independent speech gates. Component PASS never overrides speech FAIL.", |
| 20 | + "source_revision": "5b9d269", |
| 21 | + "source_url": "https://github.com/InferMatrixLab/InferRSI/blob/5b9d269/src/inferrsi/data/skill_library/en/vllm-omni/sub-skills/omni-validation/SKILL.md", |
| 22 | + "current_path": "src/inferrsi/data/skill_library/repositories/vllm-omni/sub-skills/omni-validation/SKILL.md" |
17 | 23 | }, |
18 | 24 | { |
19 | 25 | "knowledge_scope": "sglang_omni", |
20 | 26 | "source_path": "src/inferrsi/data/skill_library/en/sglang-omni/sub-skills/sglang-omni-architecture/SKILL.md", |
21 | 27 | "content_sha256": "766b2d66d2056dd779abd9e492cb3af6e25899298f44964f565509f130a7cb24", |
22 | 28 | "lesson": "Trace concrete stages and terminal outputs at the pinned source rather than infer behavior from model name.", |
23 | | - "application": "Identify terminal waveform semantics, FP16 compiled Code2Wav and reference_workers=8; retain native scorer implementation and a separate CPU parallel preparation prototype." |
| 29 | + "application": "Identify terminal waveform semantics, FP16 compiled Code2Wav and reference_workers=8; retain native scorer implementation and a separate CPU parallel preparation prototype.", |
| 30 | + "source_revision": "5b9d269", |
| 31 | + "source_url": "https://github.com/InferMatrixLab/InferRSI/blob/5b9d269/src/inferrsi/data/skill_library/en/sglang-omni/sub-skills/sglang-omni-architecture/SKILL.md", |
| 32 | + "current_path": "src/inferrsi/data/skill_library/repositories/sglang-omni/sub-skills/sglang-omni-architecture/SKILL.md" |
24 | 33 | }, |
25 | 34 | { |
26 | 35 | "knowledge_scope": "minicpmo_single_gpu_engine", |
27 | 36 | "source_path": "src/inferrsi/data/knowledge_packs/peer/minicpmo_single_gpu_engine/cuda/audio/streaming-transfer-constraints.json", |
28 | 37 | "content_sha256": "58ebe6ac09cbbb774216acd0fdaef29409e685cd7a7521501154b67b963565fa", |
29 | 38 | "lesson": "Preserve audio-delivery metrics, disjoint pools, request-owned state and quality-constrained transfer.", |
30 | | - "application": "Reuse pinned adapters/client/workload, test MPS and graph/batching hypotheses, and explicitly distinguish this steady-state campaign from strict cold-cache certification." |
| 39 | + "application": "Reuse pinned adapters/client/workload, test MPS and graph/batching hypotheses, and explicitly distinguish this steady-state campaign from strict cold-cache certification.", |
| 40 | + "source_revision": "5b9d269", |
| 41 | + "source_url": "https://github.com/InferMatrixLab/InferRSI/blob/5b9d269/src/inferrsi/data/knowledge_packs/peer/minicpmo_single_gpu_engine/cuda/audio/streaming-transfer-constraints.json" |
31 | 42 | } |
32 | 43 | ] |
33 | 44 | } |
0 commit comments