diff --git a/docs/reference-matrix.md b/docs/reference-matrix.md
index d0804590e..06e80ea90 100644
--- a/docs/reference-matrix.md
+++ b/docs/reference-matrix.md
@@ -554,3 +554,9 @@ create a repo runtime descriptor, backend port or measured resource effect.
| Source / revision | Inspected scope | Extraction and qualification boundary |
| --- | --- | --- |
| vllm-project/vllm-ascend at `3b31886237ed65c435e0965b10e8002ff5769766` | Release table, plugin hooks, DeepSeek-V4/DSpark evolution, Kimi K3/MiniMax-M3/GLM5+ optimization, DSA-CP/FlashComm1/Fused-MC2 and fusion dispatch, reversals and numerical oracles. [History screening and semantic scope](../src/inferrsi/data/skill_library/repositories/vllm-ascend/references/history-scope.md). [Source locators](../src/inferrsi/data/skill_library/repositories/vllm-ascend/references/sources.md) | [Release-specific diagnosis](../src/inferrsi/data/skill_library/repositories/vllm-ascend/SKILL.md). Candidate; device execution and downstream efficacy unverified. Older ACLGraph and checkpoint-adaptation owners retain their own source revisions. |
+
+## DeepSeek streamed expert-cache oracle contracts
+
+| Source | Inspected scope | Canonical extraction and boundary |
+| --- | --- | --- |
+| [ronak-create/deepseek-v4-in-c](https://github.com/ronak-create/deepseek-v4-in-c/tree/d0402056d0da4bf7edfd2d4977de479a0db9d201) @ `d0402056d0da4bf7edfd2d4977de479a0db9d201` (Apache-2.0 root and selected-file declarations; third-party/model rights excluded) | Bounded source inspection, 2026-11-03: byte/fixture geometry, begin/end slot lifetime, actual model/oracle callback wiring and tiny-reference limits; [selected hashes, regions and rights](../src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-source.json) | [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)): independent buffered bytes, usable residents during misses, ordered accumulation and oracle reachability. Source-derived candidate/not_run; historical corruption reports are not reproduced tests, and tiny fixtures do not establish released-model quality or performance. |
diff --git a/docs/searchspace-matrix.md b/docs/searchspace-matrix.md
index 3c4c2f4b7..94c79f250 100644
--- a/docs/searchspace-matrix.md
+++ b/docs/searchspace-matrix.md
@@ -628,3 +628,9 @@ NOASSERTION and InferRSI's repository license.
| Accelerator / Host | Diagnose engine-process plugin discovery and MiniMax-M3 mixed KV storage from hardware capability through per-layer resolution | [vllm-ascend](../src/inferrsi/data/skill_library/repositories/vllm-ascend/SKILL.md), [cache path](../src/inferrsi/data/skill_library/repositories/vllm-ascend/references/minimax-cache.md) | Entry-point metadata and cache specs can be inspected separately from device dispatch and numerics. |
| Communication / Accelerator | Verify DSpark draft budget, effective P/D ranks and target capture tiers | [vllm-ascend](../src/inferrsi/data/skill_library/repositories/vllm-ascend/SKILL.md), [DSpark PD](../src/inferrsi/data/skill_library/repositories/vllm-ascend/references/dspark-pd.md) | TP4 acceptance-length fixture and two-node 950DT deployment have separate evidence requirements. |
| Accelerator / RAM / Host | Diagnose fused preprocessing, grouped GEMM/activation/quant, residual specialization, KV index reuse and reverted optimizations | [vllm-ascend](../src/inferrsi/data/skill_library/repositories/vllm-ascend/SKILL.md), [fusion](../src/inferrsi/data/skill_library/repositories/vllm-ascend/references/fused-operators.md), [model performance](../src/inferrsi/data/skill_library/repositories/vllm-ascend/references/model-performance.md) | Whole-history discovery and selected semantic review remain separate from native performance validation. |
+
+## DeepSeek streamed expert-cache verification
+
+| 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. |
diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md
index 6d152f768..57bf903c5 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: `sha256:19d575629009048019f2fd2dfcf1f4c4afd1d529fa4cfc27e65dfb3103768f6f`
+- Input digest: `sha256:2cf9a15b2055e4bbaf10377e776bbda22239eec7ca313be517995ae33276c231`
- Regenerate: `python tools/build_skill_coverage.py`
- Verify without writing: `python tools/build_skill_coverage.py --check`
@@ -143,7 +143,7 @@ optimization is fully covered or that any procedure has measured benefit.
| [training-input-pipeline](../src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md) | common | Diagnose training input stalls and choose bounded queue, shard-cache and cache-ownership changes from DALI, Mosaic ... | Diagnosing training input stalls and choosing bounded queue, shard-cache and ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Observe / Configure / Optimize
2 RAM/Host · Observe / Diagnose / Configure / Optimize
3 SSD · Diagnose / Configure / Optimize
4 Network · Configure / Optimize
5 OS · Configure / Optimize | 1 · Systems sources (B1 / NVIDIA DALI @ `6fe47c48d3b0ca335f5e1afd3c77c750b2c7e041` ...)
1 · Systems sources (B2 / Mosaic Streaming @ `d99bf9c7cdf2dd4ed62a4960ed35270b93337a5f` ...)
1 · Systems sources (B3 / FFCV @ `3a12966b3afe3a81733a732e633317d747bfaac7` ...) | candidate · not_run · 1 readset(s) | no-eval |
| [training-resource-budget](../src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md) | common | Choose training offload and resource-budget changes using pinned DeepSpeed configuration, attributed expert diagnosis ... | Choosing training offload and resource-budget changes from pinned DeepSpeed ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Diagnose / Configure / Optimize / Verify
3 SSD · Configure / Optimize
4 Network · Observe / Diagnose / Verify
5 OS · Diagnose / Configure | 1 · Systems sources (CR1 / DeepSpeed ZeRO-Infinity @ `5f266e26d216a705d9c64cac88fb813ce ...)
1 · Systems sources (CR5 / Ascend MindSpeed adaptive memory — historical `core_r0.8.0` ...)
1 · Systems sources (ZeRO-Offload — [2101.06840v1](https://arxiv.org/abs/2101.06840v1))
2 · Methodology and corpus references (CR6 / Stas Bekman ML Engineering Open Book @ ...) | candidate · not_run · 9 readset(s) | no-eval |
| [video-diffusion-optimization](../src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md) | common | Run and verify goal-driven video diffusion optimization, backend-agnostic. Use when the user asks for latency, RTF ... | Backend-agnostic video diffusion inference and serving optimization | — | 1 Accelerator † guidance: Video diffusion campaign guidance (unvalidated) | — | candidate · not_run · 0 readset(s) | no-Reference-Matrix-row, no-eval |
-| [weight-loading-optimization](../src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md) | common | Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source ... | Diagnosing the serving-level weight load path from storage to device and choosing changes ... | failure_modes, input_contract, produced_artifact, refusal_point
5 advice | 1 Accelerator † guidance: Flare preparation and complete-session source contracts
2 RAM/Host † guidance: Flare preparation and complete-session source contracts
3 SSD · Observe / Profile / Diagnose / Optimize / Verify † guidance: Flare preparation and complete-session source contracts
4 Network · Configure / Optimize / Verify | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c)
1 · Systems sources (foundation-model-stack/fastsafetensors @ 219e256)
1 · Systems sources (ggml-org/llama.cpp @ `3e390af18de1131dda27913c00511725b4274c8a` (MIT ...)
1 · Systems sources (mindie-agent/mindie-agent @ 6a81b41) | candidate · not_run · 3 readset(s) | no-eval |
+| [weight-loading-optimization](../src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md) | common | Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source ... | Diagnosing the serving-level weight load path from storage to device and choosing changes ... | failure_modes, input_contract, produced_artifact, refusal_point
6 advice | 1 Accelerator † guidance: Flare preparation and complete-session source contracts
2 RAM/Host † guidance: Flare preparation and complete-session source contracts; DeepSeek streamed expert-cache verification
3 SSD · Observe / Profile / Diagnose / Optimize / Verify † guidance: Flare preparation and complete-session source contracts; DeepSeek streamed expert-cache verification
4 Network · Configure / Optimize / Verify | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c)
1 · Systems sources (foundation-model-stack/fastsafetensors @ 219e256)
1 · Systems sources (ggml-org/llama.cpp @ `3e390af18de1131dda27913c00511725b4274c8a` (MIT ...)
1 · Systems sources (mindie-agent/mindie-agent @ 6a81b41)
DeepSeek streamed expert-cache oracle contracts ([ronak-create/deepseek-v4-in-c](https://github.com/ronak-create/deepse ...) | candidate · not_run · 5 readset(s) | no-eval |
## Registered: Workflows (6)
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
index 1edc8e137..a5713edfb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
@@ -15,7 +15,7 @@ These are knowledge abstracts (`--kind knowledge`), with namespaced package
IDs distinct from explicitly selected `checkout_ref` skills. Ascend/CUDA
procedures use their corresponding backend; shared methods use `common`.
-Source snapshot: [5898e717015272257fdd14e06d8ae6b542d2c3ce](https://github.com/InferMatrixLab/InferRSI/tree/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library).
+Source snapshot: [630bb57a54b4ddda8338e0e529664436f982d94e](https://github.com/InferMatrixLab/InferRSI/tree/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library).
```bash
inferrsi knowledge search 'profiling' --repo-type inferrsi --limit 50
@@ -29,7 +29,7 @@ contain the current input paths and bytes (Git checkout filters are respected).
```bash
python tools/build_skill_index.py --source-revision
-python tools/build_skill_index.py --source-revision 5898e717015272257fdd14e06d8ae6b542d2c3ce --check
+python tools/build_skill_index.py --source-revision 630bb57a54b4ddda8338e0e529664436f982d94e --check
```
`--check` detects missing, stale and unexpected generated records without writing.
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
index 5eda780f0..f47ef9d2f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
index 1a38912f2..eb6e6bf8f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
index 8fe8c8c66..6744edd91 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
index 725c9d822..2692aeacc 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
index 1013a7d0d..5b04e1f1a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
index 61903cb35..230f8fcaf 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
index c81002898..0414fa51f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
index 74d0620e5..e365666aa 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
index d6e91f321..18d17853a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
index 1c694b348..a848b4f41 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
index e0974aaa4..4e6bf1537 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
index b83d538d2..dcc13d94f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json
index f238bc640..c6856a87d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
index 075e2a40c..eb4fa1840 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
index b77d38521..6851a212a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/vllm-ascend.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/vllm-ascend.json
index 7bbb2a888..96a96a147 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/vllm-ascend.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/vllm-ascend.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/vllm-ascend/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/vllm-ascend/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-ascend/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm-ascend/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
index e30de6a0e..20ab08590 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
index d1732807f..73c915946 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
index c507d381c..b2a4aa83b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
index 86912635c..da4276275 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
index b2c73cc8b..4d8a0fa02 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
index 478fb0d66..126d936ce 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
index ef2fd4123..e556c35b0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
index 2195b9449..e43a3b607 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
index fcecbf080..c67dcff68 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
index 37c1fc7db..39c0eb407 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
index 866982bb8..a7c0035b6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
index a6320dcfc..ea1400018 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
index 631a567db..7d4fe393a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
index ea64dfbca..d48ce21f5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
index f83ed38b8..e207b981e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
index c4e6a81d2..1d7131867 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
index 9c71ff9f4..e55a74952 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
index 1354c963b..cfe5f515c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
index daef448c1..6f5a4ddd9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
index 7d6f8716f..03325a6ea 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
index 16b756168..9d7266296 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
index 107e42384..3f52660e9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
index c785e5c23..352ea56f6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
index 6b566f808..f1231cf70 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
index df3ff1058..7fd39e46c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
index 835ccbf77..be0f3de43 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
index b65b41915..0379a2570 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
index b49365b44..a3ceac31b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
index 992a12da8..cd8d95e7d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
index 9095df9e2..6a71caa95 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
index 450ece071..b3efbc187 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
"src/inferrsi/data/skill_library/repositories/husky/metadata.json",
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 d0567fade..d7435a8e3 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
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
index ac591b8c8..a710c376a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/admission-drain-occupancy.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
index d56505e89..ee91df5dc 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/causal-conv1d-dispatch-state.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
index 9cedc87b5..2e8ced1d1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
index 824f3e670..3fd78e12f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
index 8bbc728a0..74e6ff8c2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/lily/LICENSE",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
index ae5ebc055..04b62d6fd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
index a86d9770f..32b960a82 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
"src/inferrsi/data/skill_library/repositories/local-dream/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
index a2f44eabc..db325fc65 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
index 2ca8f483c..34750d6e0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
index f56aba383..51a083d41 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
index e517b9350..90a3ca439 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
index 86774f567..396059b19 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
index 2f3a08df4..8a97e0811 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
index 36ff84e75..e615bdf5b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
index a4007f684..45975c590 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
index d5de2182a..7a40642ac 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json
index 471373075..0e19756aa 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/LICENSE",
"src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
index 9d68d21d1..d4eb6b8a6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
index c1f7e95bd..fd46fb94d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
index fda70b895..791745a71 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
index 83715e8ce..b5ed67f7c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/splash/LICENSE",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
index 78f652e81..f3d39e43c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
index 4cc81d45c..e2d79100d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
index a0049e497..da63b1c48 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
index 6fa371999..301eeb0b7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
index 739b27f32..4ec290e80 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
index 631a90c36..49c52d6cf 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
@@ -8,7 +8,7 @@
"component": "skills",
"scenario": "skills.discovery",
"status": "imported",
- "summary": "Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source, format, placement, duplication), account startup fixed overhead separately from per-request cost with bytes-and-counts lines and hit-criteria warmup validation, verify sharded weight downloads against the safetensors index, apply device-side weight prefetch only on profiled memory-bound hot spots, and order CPU-tier offload candidates by change surface. Includes source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics and fastsafetensors broadcast/scatter/coalesced delivery. Use when startup/loading latency dominates; route DCP asynchronous save performance and distributed recovery validity to their separate skills.\n",
+ "summary": "Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source, format, placement, duplication), account startup fixed overhead separately from per-request cost with bytes-and-counts lines and hit-criteria warmup validation, verify sharded weight downloads against the safetensors index, apply device-side weight prefetch only on profiled memory-bound hot spots, and order CPU-tier offload candidates by change surface. Includes source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics and fastsafetensors broadcast/scatter/coalesced delivery; also audit streamed expert-cache byte correctness and begin/end overlap contracts. Use when startup/loading latency dominates or a streamed expert-cache change needs correctness verification; route DCP asynchronous save performance and distributed recovery validity to their separate skills.\n",
"tags": [
"storage",
"weight-loading",
@@ -21,7 +21,9 @@
"tensor-parallel",
"Flare",
"model-preparation",
- "prepared-runtime-package"
+ "prepared-runtime-package",
+ "expert-cache",
+ "loader-correctness"
],
"origin": [
"external"
@@ -29,17 +31,19 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json",
+ "src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-oracles.md",
+ "src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-source.json",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/fastsafetensors-transfer-readset.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/flare-preparation-boundary.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/llama-loading-readset.md",
"src/inferrsi/data/skill_library/skill-index.json"
],
- "content_sha256": "a1bf0e2ec066d5f30225c6cc5fb603f830f48eb5e2aee3fc3d084ed9d8aebe2f"
+ "content_sha256": "b970ef27da2de14c20d3fe734cc5ee0cf8445c396f53dab588bd180943921833"
}
],
"index_type": "packaged-skill",
@@ -48,8 +52,10 @@
},
"checkout_ref": "checkout:weight-loading-optimization",
"source_files": {
- "src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md": "a1bf0e2ec066d5f30225c6cc5fb603f830f48eb5e2aee3fc3d084ed9d8aebe2f",
- "src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json": "ebcc5425d7b6409e5c509122e9dc76bed1caab09d73daf013d96086d7faaba2b",
+ "src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md": "b970ef27da2de14c20d3fe734cc5ee0cf8445c396f53dab588bd180943921833",
+ "src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json": "007142544806e5b065b8c48b6e467f49c65b4a55465d09d7a88e0997287bb212",
+ "src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-oracles.md": "5e42c4a01f9ac810663979dee0658c4cbe87060594c16622ae6ec0b854e44b29",
+ "src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-source.json": "65bf0623616b4f633d8f51b30134798cbf71176716c8a8613209e2fc48ff84bd",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/fastsafetensors-transfer-readset.md": "5a39f14b102b0cee304541e011aa31e44e6b13527d1bee9987c3f3b66c064d3d",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/flare-preparation-boundary.md": "1d7922785a196bdbafdc9791fe79428465a5acc0292ca8f99446f8c764b35454",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/llama-loading-readset.md": "6c640f28ee3f0253a17b81461c865cb9f947367002cb75bf72550c4572d02e97",
@@ -57,7 +63,7 @@
},
"skill_metadata": {
"name": "weight-loading-optimization",
- "description": "Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source, format, placement, duplication), account startup fixed overhead separately from per-request cost with bytes-and-counts lines and hit-criteria warmup validation, verify sharded weight downloads against the safetensors index, apply device-side weight prefetch only on profiled memory-bound hot spots, and order CPU-tier offload candidates by change surface. Includes source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics and fastsafetensors broadcast/scatter/coalesced delivery. Use when startup/loading latency dominates; route DCP asynchronous save performance and distributed recovery validity to their separate skills.\n",
+ "description": "Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source, format, placement, duplication), account startup fixed overhead separately from per-request cost with bytes-and-counts lines and hit-criteria warmup validation, verify sharded weight downloads against the safetensors index, apply device-side weight prefetch only on profiled memory-bound hot spots, and order CPU-tier offload candidates by change surface. Includes source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics and fastsafetensors broadcast/scatter/coalesced delivery; also audit streamed expert-cache byte correctness and begin/end overlap contracts. Use when startup/loading latency dominates or a streamed expert-cache change needs correctness verification; route DCP asynchronous save performance and distributed recovery validity to their separate skills.\n",
"tags": [
"storage",
"weight-loading",
@@ -70,7 +76,9 @@
"tensor-parallel",
"Flare",
"model-preparation",
- "prepared-runtime-package"
+ "prepared-runtime-package",
+ "expert-cache",
+ "loader-correctness"
],
"status": "candidate",
"profile": "local-skill-v1",
@@ -133,14 +141,38 @@
],
"readset": "references/flare-preparation-boundary.md",
"license": "Apache-2.0"
+ },
+ {
+ "type": "git",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c",
+ "commit": "d0402056d0da4bf7edfd2d4977de479a0db9d201",
+ "paths": [
+ "LICENSE",
+ "README.md",
+ "Makefile",
+ "src/cache/dsv4_cache.c",
+ "src/cache/dsv4_cache.h",
+ "src/io/dsv4_st.c",
+ "src/io/dsv4_st.h",
+ "src/model/dsv4_layer.h",
+ "src/model/dsv4_layer.c",
+ "src/cli/dsv4_run.c",
+ "tests/unit/test_cache.c",
+ "tests/unit/test_model_oracle.c",
+ "tools/make_st_fixture.py",
+ "tools/make_tiny_checkpoint.py",
+ "tools/dsv4_ref.py"
+ ],
+ "readset": "references/deepseek-expert-cache-oracles.md",
+ "license": "Apache-2.0 (repository declaration and selected-file notices; third-party/model rights excluded)"
}
],
"empirical_validation": "not_run",
"applicability": {
- "scope": "Diagnosing the serving-level weight load path from storage to device and choosing changes to it when startup or loading latency dominates: locating the cost by source, format, placement and duplication, accounting once-per-process loading and warmup apart from per-request cost, verifying sharded downloads, restricting device-side prefetch to profiled memory-bound hot spots and ordering CPU-tier offload; also covers source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics, fastsafetensors broadcast/scatter/coalesced delivery and the separation of model preparation from serving. Request-time host overhead belongs to host-overhead-profiling, DCP asynchronous save performance to dcp-async-save-performance and distributed recovery validity to checkpoint-recovery-consistency",
+ "scope": "Diagnosing the serving-level weight load path from storage to device and choosing changes to it when startup or loading latency dominates: locating the cost by source, format, placement and duplication, accounting once-per-process loading and warmup apart from per-request cost, verifying sharded downloads, restricting device-side prefetch to profiled memory-bound hot spots and ordering CPU-tier offload; also covers source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics, fastsafetensors broadcast/scatter/coalesced delivery and the separation of model preparation from serving. Request-time host overhead belongs to host-overhead-profiling, DCP asynchronous save performance to dcp-async-save-performance and distributed recovery validity to checkpoint-recovery-consistency; for disk-streamed routed experts, audit independent byte identity, resident/pending slot ownership, ordered accumulation and actual oracle callback coverage; the DeepSeek source slice describes POSIX storage, C/OpenMP and pthread ownership; target validation must identify its actual direct/buffered and reader-pool paths",
"prerequisites": "Flag names and API entry paths are re-verified against the pinned source revisions, and torch_npu.npu_prefetch details against the target CANN version, before they are relied on; card state is checked before launch and process placement fixed before measuring, through ascend-device-health and os-process-placement; the Diffusers enable_model_cpu_offload path requires accelerate",
- "input_contract": "The current load path's byte source, file format, host-first or direct device placement and duplicated copies across processes or instances; startup timings for loading and warmup with byte and access counts per segment, and the expected warmup hit criteria; the safetensors index weight_map and the downloaded shard set; a profile showing memory-bound hot spots and the adjacent compute or communication segments; free device memory and whether an OOM occurs during construction or inference; for cross-rank delivery, the file-to-loader-group-rank map, tensor names, shapes, dtypes and call order, the required replicated or sliced output, the model's expected shard layout and the collective backend",
- "produced_artifact": "The located load-path cost on the four dimensions with its matching lever, a startup ledger that keeps fixed overhead apart from per-request cost, a structural shard-completeness result naming the missing shards, a prefetch proposal behind an enable_prefetch switch defaulting to False or a recorded decision not to prefetch, an offload candidate order, the recorded loading mode, a distributed delivery choice with its proposed verification, and a preparation/serving record of serialized and rebuilt state, live resource owners and ABI/prepared-contract versions; proposed changes and verification targets, not a measured startup or delivery gain",
+ "input_contract": "The current load path's byte source, file format, host-first or direct device placement and duplicated copies across processes or instances; startup timings for loading and warmup with byte and access counts per segment, and the expected warmup hit criteria; the safetensors index weight_map and the downloaded shard set; a profile showing memory-bound hot spots and the adjacent compute or communication segments; free device memory and whether an OOM occurs during construction or inference; for cross-rank delivery, the file-to-loader-group-rank map, tensor names, shapes, dtypes and call order, the required replicated or sliced output, the model's expected shard layout and the collective backend; for streamed expert-cache changes, the selected direct/buffered I/O and reader-pool path, tensor offsets and cache-slot budget, requested keys, source callback table, independent byte/numerical reference and checked fixture coverage",
+ "produced_artifact": "The located load-path cost on the four dimensions with its matching lever, a startup ledger that keeps fixed overhead apart from per-request cost, a structural shard-completeness result naming the missing shards, a prefetch proposal behind an enable_prefetch switch defaulting to False or a recorded decision not to prefetch, an offload candidate order, the recorded loading mode, a distributed delivery choice with its proposed verification, and a preparation/serving record of serialized and rebuilt state, live resource owners and ABI/prepared-contract versions; proposed changes and verification targets, not a measured startup or delivery gain; a source-bound expert-cache byte/state audit and oracle-path coverage record, with synthetic-fixture, fallback and SKIP limits",
"failure_modes": [
"Loading and warmup are amortized into per-request latency, or a host segment without timing points is treated as costing nothing",
"Warmup is accepted on total time while mismatched warmup shapes have warmed a different set of cache entries",
@@ -150,9 +182,11 @@
"A free-memory probe runs on an mmap-backed load, or a run whose first mlock failure latched with only a warning is treated as fully resident",
"Disk counters that include cache and pipes are read as physical writes, or a run without a page-cache drop is reported as a cold start",
"The ceil-block scatter partition is used without checking it against the model's expected shard layout, or ranks run different orders or plans so their collectives mismatch",
- "A faster H2D copy, local clone or rank-local filtered read is counted as a cross-rank communication gain"
+ "A faster H2D copy, local clone or rank-local filtered read is counted as a cross-rank communication gain",
+ "Serial and concurrent cache paths agree on identically corrupted placement bytes, or a contiguous fixture misses a separated-run alignment failure",
+ "Resident bytes or ordered per-k accumulation change while misses are pending, or an exit-zero SKIP/non-overlapped oracle is counted as overlap validation"
],
- "refusal_point": "Withholds a load-path win until the verification checklist holds and keeps the distributed-delivery checks as verification targets rather than results; does not force prefetch without a safe dependency window and disables it when neither a smaller size nor a moved position removes a regression; records unsupported save behaviors as unknown; rejects mismatched execution ABI or prepared-contract versions instead of editing their tags; and implies no Ascend support from the llama.cpp or fastsafetensors mechanisms",
+ "refusal_point": "Withholds a load-path win until the verification checklist holds and keeps the distributed-delivery checks as verification targets rather than results; does not force prefetch without a safe dependency window and disables it when neither a smaller size nor a moved position removes a regression; records unsupported save behaviors as unknown; rejects mismatched execution ABI or prepared-contract versions instead of editing their tags; and implies no Ascend support from the llama.cpp or fastsafetensors mechanisms; withholds expert-cache correctness/overlap conclusions when byte coverage, actual I/O/pool dispatch or model-oracle callback reachability is missing",
"hardware": "NPU for the device-side prefetch API (torch_npu.npu_prefetch, checked against the target CANN version) and the Diffusers offload playbook; the prefetch and fastsafetensors methodology transfers while the fastsafetensors and llama.cpp APIs do not, and cross-rank completion follows each collective backend's own wait contract"
},
"recommendations": [
@@ -160,27 +194,31 @@
"Place each prefetch behind an enable_prefetch switch defaulting to False, start max_size at 50% of the theoretical size, verify a single operator first and step up to 0.7 and then 1.0 only after verified wins.",
"Try offload candidates in order of change surface: batch size, resolution, CFG/feature flags, CPU or layered offload, tensor-parallel degree, then activation checkpointing; re-estimate the un-overlapped transfer share after adopting each tier.",
"Record whole-set atomic publication, asynchronous checkpointing, multi-rank save coordination and crash durability as unknown for the fastsafetensors saver instead of assuming them.",
- "Before adopting coalesced delivery, compare it disabled and enabled with identical checkpoint, rank mapping, backend, topology, dtype, cache state, chunk/queue budgets and consumer retention, then measure complete startup under the same accounting protocol."
+ "Before adopting coalesced delivery, compare it disabled and enabled with identical checkpoint, rank mapping, backend, topology, dtype, cache state, chunk/queue budgets and consumer retention, then measure complete startup under the same accounting protocol.",
+ "For streamed expert-cache changes, compare against independent buffered tensor bytes, exercise separated-run geometry, check resident bytes during pending reads, preserve ordered per-k accumulation and verify that the numerical oracle selects the actual callbacks."
],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Normalized provenance covers the existing llama.cpp loading slice at 3e390af18de1131dda27913c00511725b4274c8a and the fastsafetensors distributed-delivery slice at 219e2561a41ede2d5d8eba611c248bb0effbf580; other inherited claims remain unnormalized with original identities and limits retained.",
"The readsets record static mechanisms and source recommendations, not validated startup, memory, placement or backend performance.",
"The fastsafetensors delivery readset does not revalidate inherited I/O/save claims or qualify uneven scatter, accelerator/NPU compatibility, inter-node transport or agent efficacy.",
- "Flare adds source-only preparation/serving ownership and ABI/contract checks; live exporter handles, native backend execution, startup gains and model/dependency rights are not qualified."
+ "Flare adds source-only preparation/serving ownership and ABI/contract checks; live exporter handles, native backend execution, startup gains and model/dependency rights are not qualified.",
+ "The DeepSeek increment records source/test definitions only: its buffered byte oracle shares tensor metadata, its resident snapshot covers w1 bytes, and its tiny PyTorch/batched checks do not qualify the released model, backend performance or agent efficacy."
],
"provenance_readsets": [
"references/llama-loading-readset.md",
"references/fastsafetensors-transfer-readset.md",
- "references/flare-preparation-boundary.md"
+ "references/flare-preparation-boundary.md",
+ "references/deepseek-expert-cache-oracles.md",
+ "references/deepseek-expert-cache-source.json"
],
"consumption_mode": "packaged"
},
"applicability": {
- "scope": "Diagnosing the serving-level weight load path from storage to device and choosing changes to it when startup or loading latency dominates: locating the cost by source, format, placement and duplication, accounting once-per-process loading and warmup apart from per-request cost, verifying sharded downloads, restricting device-side prefetch to profiled memory-bound hot spots and ordering CPU-tier offload; also covers source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics, fastsafetensors broadcast/scatter/coalesced delivery and the separation of model preparation from serving. Request-time host overhead belongs to host-overhead-profiling, DCP asynchronous save performance to dcp-async-save-performance and distributed recovery validity to checkpoint-recovery-consistency",
+ "scope": "Diagnosing the serving-level weight load path from storage to device and choosing changes to it when startup or loading latency dominates: locating the cost by source, format, placement and duplication, accounting once-per-process loading and warmup apart from per-request cost, verifying sharded downloads, restricting device-side prefetch to profiled memory-bound hot spots and ordering CPU-tier offload; also covers source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics, fastsafetensors broadcast/scatter/coalesced delivery and the separation of model preparation from serving. Request-time host overhead belongs to host-overhead-profiling, DCP asynchronous save performance to dcp-async-save-performance and distributed recovery validity to checkpoint-recovery-consistency; for disk-streamed routed experts, audit independent byte identity, resident/pending slot ownership, ordered accumulation and actual oracle callback coverage; the DeepSeek source slice describes POSIX storage, C/OpenMP and pthread ownership; target validation must identify its actual direct/buffered and reader-pool paths",
"prerequisites": "Flag names and API entry paths are re-verified against the pinned source revisions, and torch_npu.npu_prefetch details against the target CANN version, before they are relied on; card state is checked before launch and process placement fixed before measuring, through ascend-device-health and os-process-placement; the Diffusers enable_model_cpu_offload path requires accelerate",
- "input_contract": "The current load path's byte source, file format, host-first or direct device placement and duplicated copies across processes or instances; startup timings for loading and warmup with byte and access counts per segment, and the expected warmup hit criteria; the safetensors index weight_map and the downloaded shard set; a profile showing memory-bound hot spots and the adjacent compute or communication segments; free device memory and whether an OOM occurs during construction or inference; for cross-rank delivery, the file-to-loader-group-rank map, tensor names, shapes, dtypes and call order, the required replicated or sliced output, the model's expected shard layout and the collective backend",
- "produced_artifact": "The located load-path cost on the four dimensions with its matching lever, a startup ledger that keeps fixed overhead apart from per-request cost, a structural shard-completeness result naming the missing shards, a prefetch proposal behind an enable_prefetch switch defaulting to False or a recorded decision not to prefetch, an offload candidate order, the recorded loading mode, a distributed delivery choice with its proposed verification, and a preparation/serving record of serialized and rebuilt state, live resource owners and ABI/prepared-contract versions; proposed changes and verification targets, not a measured startup or delivery gain",
+ "input_contract": "The current load path's byte source, file format, host-first or direct device placement and duplicated copies across processes or instances; startup timings for loading and warmup with byte and access counts per segment, and the expected warmup hit criteria; the safetensors index weight_map and the downloaded shard set; a profile showing memory-bound hot spots and the adjacent compute or communication segments; free device memory and whether an OOM occurs during construction or inference; for cross-rank delivery, the file-to-loader-group-rank map, tensor names, shapes, dtypes and call order, the required replicated or sliced output, the model's expected shard layout and the collective backend; for streamed expert-cache changes, the selected direct/buffered I/O and reader-pool path, tensor offsets and cache-slot budget, requested keys, source callback table, independent byte/numerical reference and checked fixture coverage",
+ "produced_artifact": "The located load-path cost on the four dimensions with its matching lever, a startup ledger that keeps fixed overhead apart from per-request cost, a structural shard-completeness result naming the missing shards, a prefetch proposal behind an enable_prefetch switch defaulting to False or a recorded decision not to prefetch, an offload candidate order, the recorded loading mode, a distributed delivery choice with its proposed verification, and a preparation/serving record of serialized and rebuilt state, live resource owners and ABI/prepared-contract versions; proposed changes and verification targets, not a measured startup or delivery gain; a source-bound expert-cache byte/state audit and oracle-path coverage record, with synthetic-fixture, fallback and SKIP limits",
"failure_modes": [
"Loading and warmup are amortized into per-request latency, or a host segment without timing points is treated as costing nothing",
"Warmup is accepted on total time while mismatched warmup shapes have warmed a different set of cache entries",
@@ -190,9 +228,11 @@
"A free-memory probe runs on an mmap-backed load, or a run whose first mlock failure latched with only a warning is treated as fully resident",
"Disk counters that include cache and pipes are read as physical writes, or a run without a page-cache drop is reported as a cold start",
"The ceil-block scatter partition is used without checking it against the model's expected shard layout, or ranks run different orders or plans so their collectives mismatch",
- "A faster H2D copy, local clone or rank-local filtered read is counted as a cross-rank communication gain"
+ "A faster H2D copy, local clone or rank-local filtered read is counted as a cross-rank communication gain",
+ "Serial and concurrent cache paths agree on identically corrupted placement bytes, or a contiguous fixture misses a separated-run alignment failure",
+ "Resident bytes or ordered per-k accumulation change while misses are pending, or an exit-zero SKIP/non-overlapped oracle is counted as overlap validation"
],
- "refusal_point": "Withholds a load-path win until the verification checklist holds and keeps the distributed-delivery checks as verification targets rather than results; does not force prefetch without a safe dependency window and disables it when neither a smaller size nor a moved position removes a regression; records unsupported save behaviors as unknown; rejects mismatched execution ABI or prepared-contract versions instead of editing their tags; and implies no Ascend support from the llama.cpp or fastsafetensors mechanisms",
+ "refusal_point": "Withholds a load-path win until the verification checklist holds and keeps the distributed-delivery checks as verification targets rather than results; does not force prefetch without a safe dependency window and disables it when neither a smaller size nor a moved position removes a regression; records unsupported save behaviors as unknown; rejects mismatched execution ABI or prepared-contract versions instead of editing their tags; and implies no Ascend support from the llama.cpp or fastsafetensors mechanisms; withholds expert-cache correctness/overlap conclusions when byte coverage, actual I/O/pool dispatch or model-oracle callback reachability is missing",
"hardware": "NPU for the device-side prefetch API (torch_npu.npu_prefetch, checked against the target CANN version) and the Diffusers offload playbook; the prefetch and fastsafetensors methodology transfers while the fastsafetensors and llama.cpp APIs do not, and cross-rank completion follows each collective backend's own wait contract"
},
"recommendations": [
@@ -200,7 +240,8 @@
"Place each prefetch behind an enable_prefetch switch defaulting to False, start max_size at 50% of the theoretical size, verify a single operator first and step up to 0.7 and then 1.0 only after verified wins.",
"Try offload candidates in order of change surface: batch size, resolution, CFG/feature flags, CPU or layered offload, tensor-parallel degree, then activation checkpointing; re-estimate the un-overlapped transfer share after adopting each tier.",
"Record whole-set atomic publication, asynchronous checkpointing, multi-rank save coordination and crash durability as unknown for the fastsafetensors saver instead of assuming them.",
- "Before adopting coalesced delivery, compare it disabled and enabled with identical checkpoint, rank mapping, backend, topology, dtype, cache state, chunk/queue budgets and consumer retention, then measure complete startup under the same accounting protocol."
+ "Before adopting coalesced delivery, compare it disabled and enabled with identical checkpoint, rank mapping, backend, topology, dtype, cache state, chunk/queue budgets and consumer retention, then measure complete startup under the same accounting protocol.",
+ "For streamed expert-cache changes, compare against independent buffered tensor bytes, exercise separated-run geometry, check resident bytes during pending reads, preserve ordered per-k accumulation and verify that the numerical oracle selects the actual callbacks."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
@@ -211,7 +252,8 @@
"Normalized provenance covers the existing llama.cpp loading slice at 3e390af18de1131dda27913c00511725b4274c8a and the fastsafetensors distributed-delivery slice at 219e2561a41ede2d5d8eba611c248bb0effbf580; other inherited claims remain unnormalized with original identities and limits retained.",
"The readsets record static mechanisms and source recommendations, not validated startup, memory, placement or backend performance.",
"The fastsafetensors delivery readset does not revalidate inherited I/O/save claims or qualify uneven scatter, accelerator/NPU compatibility, inter-node transport or agent efficacy.",
- "Flare adds source-only preparation/serving ownership and ABI/contract checks; live exporter handles, native backend execution, startup gains and model/dependency rights are not qualified."
+ "Flare adds source-only preparation/serving ownership and ABI/contract checks; live exporter handles, native backend execution, startup gains and model/dependency rights are not qualified.",
+ "The DeepSeek increment records source/test definitions only: its buffered byte oracle shares tensor metadata, its resident snapshot covers w1 bytes, and its tiny PyTorch/batched checks do not qualify the released model, backend performance or agent efficacy."
],
"verification": {
"locally_verified": false,
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
index 3b39ef7ba..0f58400eb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
index e1e11b682..0a4f76d2a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
index 41cd06531..d3d620c2b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
index 9eb3557e5..e900ca3bd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
index 8b703ffbe..ef2fbe23f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
@@ -36,8 +36,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/5898e717015272257fdd14e06d8ae6b542d2c3ce/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
- "commit": "5898e717015272257fdd14e06d8ae6b542d2c3ce",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/630bb57a54b4ddda8338e0e529664436f982d94e/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
+ "commit": "630bb57a54b4ddda8338e0e529664436f982d94e",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json",
diff --git a/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md b/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md
index 144cc6aed..73ef82bae 100644
--- a/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md
+++ b/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md
@@ -1,6 +1,6 @@
---
name: weight-loading-optimization
-description: 'Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source, format, placement, duplication), account startup fixed overhead separately from per-request cost with bytes-and-counts lines and hit-criteria warmup validation, verify sharded weight downloads against the safetensors index, apply device-side weight prefetch only on profiled memory-bound hot spots, and order CPU-tier offload candidates by change surface. Includes source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics and fastsafetensors broadcast/scatter/coalesced delivery. Use when startup/loading latency dominates; route DCP asynchronous save performance and distributed recovery validity to their separate skills.
+description: 'Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source, format, placement, duplication), account startup fixed overhead separately from per-request cost with bytes-and-counts lines and hit-criteria warmup validation, verify sharded weight downloads against the safetensors index, apply device-side weight prefetch only on profiled memory-bound hot spots, and order CPU-tier offload candidates by change surface. Includes source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics and fastsafetensors broadcast/scatter/coalesced delivery; also audit streamed expert-cache byte correctness and begin/end overlap contracts. Use when startup/loading latency dominates or a streamed expert-cache change needs correctness verification; route DCP asynchronous save performance and distributed recovery validity to their separate skills.
'
---
@@ -436,6 +436,16 @@ separate; changing language at the boundary does not by itself establish a gain.
For complete runtime KV/recurrent/convolution/tail state, follow that reference
to the Flare session owner. The addition remains candidate and empirical not_run.
+## 11. Verify streamed expert-cache bytes and overlap
+
+For aligned reads, cache-slot placement or overlapping routed-expert loads, use
+the [DeepSeek expert-cache reference](references/deepseek-expert-cache-oracles.md).
+Compare bytes through an independent read path, preserve resident/pending slot
+lifetimes and ordered caller accumulation, and confirm that the numerical oracle
+reaches the actual callbacks. Keep its synthetic-fixture and SKIP boundaries;
+request-time expert reads remain separate from startup loading. These are
+source-derived verification targets, not executed model or performance results.
+
## Verification checklist (before claiming a load-path win)
- [ ] Fixed overhead accounted separately from per-request cost
@@ -448,3 +458,4 @@ to the Flare session owner. The addition remains candidate and empirical not_run
protocol stated (fadvise or page-cache drop)
- [ ] Prefetch changes verified single-op first, switch defaults recorded
- [ ] Offload tier wins re-estimated against un-overlapped transfer share
+- [ ] Streamed expert-cache changes checked for independent bytes, resident/pending lifetime and actual oracle-path coverage
diff --git a/src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json b/src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json
index 5efd71b48..fa7edd96e 100644
--- a/src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json
+++ b/src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json
@@ -11,7 +11,9 @@
"tensor-parallel",
"Flare",
"model-preparation",
- "prepared-runtime-package"
+ "prepared-runtime-package",
+ "expert-cache",
+ "loader-correctness"
],
"status": "candidate",
"profile": "local-skill-v1",
@@ -74,14 +76,38 @@
],
"readset": "references/flare-preparation-boundary.md",
"license": "Apache-2.0"
+ },
+ {
+ "type": "git",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c",
+ "commit": "d0402056d0da4bf7edfd2d4977de479a0db9d201",
+ "paths": [
+ "LICENSE",
+ "README.md",
+ "Makefile",
+ "src/cache/dsv4_cache.c",
+ "src/cache/dsv4_cache.h",
+ "src/io/dsv4_st.c",
+ "src/io/dsv4_st.h",
+ "src/model/dsv4_layer.h",
+ "src/model/dsv4_layer.c",
+ "src/cli/dsv4_run.c",
+ "tests/unit/test_cache.c",
+ "tests/unit/test_model_oracle.c",
+ "tools/make_st_fixture.py",
+ "tools/make_tiny_checkpoint.py",
+ "tools/dsv4_ref.py"
+ ],
+ "readset": "references/deepseek-expert-cache-oracles.md",
+ "license": "Apache-2.0 (repository declaration and selected-file notices; third-party/model rights excluded)"
}
],
"empirical_validation": "not_run",
"applicability": {
- "scope": "Diagnosing the serving-level weight load path from storage to device and choosing changes to it when startup or loading latency dominates: locating the cost by source, format, placement and duplication, accounting once-per-process loading and warmup apart from per-request cost, verifying sharded downloads, restricting device-side prefetch to profiled memory-bound hot spots and ordering CPU-tier offload; also covers source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics, fastsafetensors broadcast/scatter/coalesced delivery and the separation of model preparation from serving. Request-time host overhead belongs to host-overhead-profiling, DCP asynchronous save performance to dcp-async-save-performance and distributed recovery validity to checkpoint-recovery-consistency",
+ "scope": "Diagnosing the serving-level weight load path from storage to device and choosing changes to it when startup or loading latency dominates: locating the cost by source, format, placement and duplication, accounting once-per-process loading and warmup apart from per-request cost, verifying sharded downloads, restricting device-side prefetch to profiled memory-bound hot spots and ordering CPU-tier offload; also covers source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics, fastsafetensors broadcast/scatter/coalesced delivery and the separation of model preparation from serving. Request-time host overhead belongs to host-overhead-profiling, DCP asynchronous save performance to dcp-async-save-performance and distributed recovery validity to checkpoint-recovery-consistency; for disk-streamed routed experts, audit independent byte identity, resident/pending slot ownership, ordered accumulation and actual oracle callback coverage; the DeepSeek source slice describes POSIX storage, C/OpenMP and pthread ownership; target validation must identify its actual direct/buffered and reader-pool paths",
"prerequisites": "Flag names and API entry paths are re-verified against the pinned source revisions, and torch_npu.npu_prefetch details against the target CANN version, before they are relied on; card state is checked before launch and process placement fixed before measuring, through ascend-device-health and os-process-placement; the Diffusers enable_model_cpu_offload path requires accelerate",
- "input_contract": "The current load path's byte source, file format, host-first or direct device placement and duplicated copies across processes or instances; startup timings for loading and warmup with byte and access counts per segment, and the expected warmup hit criteria; the safetensors index weight_map and the downloaded shard set; a profile showing memory-bound hot spots and the adjacent compute or communication segments; free device memory and whether an OOM occurs during construction or inference; for cross-rank delivery, the file-to-loader-group-rank map, tensor names, shapes, dtypes and call order, the required replicated or sliced output, the model's expected shard layout and the collective backend",
- "produced_artifact": "The located load-path cost on the four dimensions with its matching lever, a startup ledger that keeps fixed overhead apart from per-request cost, a structural shard-completeness result naming the missing shards, a prefetch proposal behind an enable_prefetch switch defaulting to False or a recorded decision not to prefetch, an offload candidate order, the recorded loading mode, a distributed delivery choice with its proposed verification, and a preparation/serving record of serialized and rebuilt state, live resource owners and ABI/prepared-contract versions; proposed changes and verification targets, not a measured startup or delivery gain",
+ "input_contract": "The current load path's byte source, file format, host-first or direct device placement and duplicated copies across processes or instances; startup timings for loading and warmup with byte and access counts per segment, and the expected warmup hit criteria; the safetensors index weight_map and the downloaded shard set; a profile showing memory-bound hot spots and the adjacent compute or communication segments; free device memory and whether an OOM occurs during construction or inference; for cross-rank delivery, the file-to-loader-group-rank map, tensor names, shapes, dtypes and call order, the required replicated or sliced output, the model's expected shard layout and the collective backend; for streamed expert-cache changes, the selected direct/buffered I/O and reader-pool path, tensor offsets and cache-slot budget, requested keys, source callback table, independent byte/numerical reference and checked fixture coverage",
+ "produced_artifact": "The located load-path cost on the four dimensions with its matching lever, a startup ledger that keeps fixed overhead apart from per-request cost, a structural shard-completeness result naming the missing shards, a prefetch proposal behind an enable_prefetch switch defaulting to False or a recorded decision not to prefetch, an offload candidate order, the recorded loading mode, a distributed delivery choice with its proposed verification, and a preparation/serving record of serialized and rebuilt state, live resource owners and ABI/prepared-contract versions; proposed changes and verification targets, not a measured startup or delivery gain; a source-bound expert-cache byte/state audit and oracle-path coverage record, with synthetic-fixture, fallback and SKIP limits",
"failure_modes": [
"Loading and warmup are amortized into per-request latency, or a host segment without timing points is treated as costing nothing",
"Warmup is accepted on total time while mismatched warmup shapes have warmed a different set of cache entries",
@@ -91,9 +117,11 @@
"A free-memory probe runs on an mmap-backed load, or a run whose first mlock failure latched with only a warning is treated as fully resident",
"Disk counters that include cache and pipes are read as physical writes, or a run without a page-cache drop is reported as a cold start",
"The ceil-block scatter partition is used without checking it against the model's expected shard layout, or ranks run different orders or plans so their collectives mismatch",
- "A faster H2D copy, local clone or rank-local filtered read is counted as a cross-rank communication gain"
+ "A faster H2D copy, local clone or rank-local filtered read is counted as a cross-rank communication gain",
+ "Serial and concurrent cache paths agree on identically corrupted placement bytes, or a contiguous fixture misses a separated-run alignment failure",
+ "Resident bytes or ordered per-k accumulation change while misses are pending, or an exit-zero SKIP/non-overlapped oracle is counted as overlap validation"
],
- "refusal_point": "Withholds a load-path win until the verification checklist holds and keeps the distributed-delivery checks as verification targets rather than results; does not force prefetch without a safe dependency window and disables it when neither a smaller size nor a moved position removes a regression; records unsupported save behaviors as unknown; rejects mismatched execution ABI or prepared-contract versions instead of editing their tags; and implies no Ascend support from the llama.cpp or fastsafetensors mechanisms",
+ "refusal_point": "Withholds a load-path win until the verification checklist holds and keeps the distributed-delivery checks as verification targets rather than results; does not force prefetch without a safe dependency window and disables it when neither a smaller size nor a moved position removes a regression; records unsupported save behaviors as unknown; rejects mismatched execution ABI or prepared-contract versions instead of editing their tags; and implies no Ascend support from the llama.cpp or fastsafetensors mechanisms; withholds expert-cache correctness/overlap conclusions when byte coverage, actual I/O/pool dispatch or model-oracle callback reachability is missing",
"hardware": "NPU for the device-side prefetch API (torch_npu.npu_prefetch, checked against the target CANN version) and the Diffusers offload playbook; the prefetch and fastsafetensors methodology transfers while the fastsafetensors and llama.cpp APIs do not, and cross-rank completion follows each collective backend's own wait contract"
},
"recommendations": [
@@ -101,19 +129,23 @@
"Place each prefetch behind an enable_prefetch switch defaulting to False, start max_size at 50% of the theoretical size, verify a single operator first and step up to 0.7 and then 1.0 only after verified wins.",
"Try offload candidates in order of change surface: batch size, resolution, CFG/feature flags, CPU or layered offload, tensor-parallel degree, then activation checkpointing; re-estimate the un-overlapped transfer share after adopting each tier.",
"Record whole-set atomic publication, asynchronous checkpointing, multi-rank save coordination and crash durability as unknown for the fastsafetensors saver instead of assuming them.",
- "Before adopting coalesced delivery, compare it disabled and enabled with identical checkpoint, rank mapping, backend, topology, dtype, cache state, chunk/queue budgets and consumer retention, then measure complete startup under the same accounting protocol."
+ "Before adopting coalesced delivery, compare it disabled and enabled with identical checkpoint, rank mapping, backend, topology, dtype, cache state, chunk/queue budgets and consumer retention, then measure complete startup under the same accounting protocol.",
+ "For streamed expert-cache changes, compare against independent buffered tensor bytes, exercise separated-run geometry, check resident bytes during pending reads, preserve ordered per-k accumulation and verify that the numerical oracle selects the actual callbacks."
],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Normalized provenance covers the existing llama.cpp loading slice at 3e390af18de1131dda27913c00511725b4274c8a and the fastsafetensors distributed-delivery slice at 219e2561a41ede2d5d8eba611c248bb0effbf580; other inherited claims remain unnormalized with original identities and limits retained.",
"The readsets record static mechanisms and source recommendations, not validated startup, memory, placement or backend performance.",
"The fastsafetensors delivery readset does not revalidate inherited I/O/save claims or qualify uneven scatter, accelerator/NPU compatibility, inter-node transport or agent efficacy.",
- "Flare adds source-only preparation/serving ownership and ABI/contract checks; live exporter handles, native backend execution, startup gains and model/dependency rights are not qualified."
+ "Flare adds source-only preparation/serving ownership and ABI/contract checks; live exporter handles, native backend execution, startup gains and model/dependency rights are not qualified.",
+ "The DeepSeek increment records source/test definitions only: its buffered byte oracle shares tensor metadata, its resident snapshot covers w1 bytes, and its tiny PyTorch/batched checks do not qualify the released model, backend performance or agent efficacy."
],
"provenance_readsets": [
"references/llama-loading-readset.md",
"references/fastsafetensors-transfer-readset.md",
- "references/flare-preparation-boundary.md"
+ "references/flare-preparation-boundary.md",
+ "references/deepseek-expert-cache-oracles.md",
+ "references/deepseek-expert-cache-source.json"
],
"consumption_mode": "packaged"
}
diff --git a/src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-oracles.md b/src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-oracles.md
new file mode 100644
index 000000000..346c07bea
--- /dev/null
+++ b/src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-oracles.md
@@ -0,0 +1,132 @@
+# Audit streamed expert-cache bytes and overlap
+
+Status: source-derived candidate. Use this reference when changing aligned reads,
+cache-slot placement or overlapped routed-expert loading. The
+[source record](deepseek-expert-cache-source.json) fixes the repository revision,
+inspected regions, file hashes, rights and exclusions. It describes checks to
+perform on an authorized target; it supplies no loader implementation or measured
+gain.
+
+Keep startup loading separate from request-time expert traffic. Before judging a
+change, identify the checkpoint and tensor metadata, effective direct/buffered
+read path, cache-slot budget, requested expert keys, source callback table and
+numerical reference. A higher cache hit rate can accompany corrupted expert
+bytes and changed routing.
+
+## Check bytes through an independent read path
+
+Compare each cached weight and scale with a plain buffered read of the same
+tensor bytes. Serial and concurrent cache loads can agree while sharing the same
+placement error; matching pointers, shapes or the two fast paths is insufficient.
+The source's byte oracle uses the safetensors reader's tensor offsets and buffered
+file descriptors, then bypasses its aligned/coalesced placement code. It does not
+independently validate the shared metadata parser.
+
+Exercise the layout that can expose the error. The source fixture separates an
+expert's scale and weight runs, gives the runs different 4096-byte alignment
+residues and writes nonconstant payload bytes. A single contiguous run cannot
+expose a later widened read overwriting an earlier run. An aligned read starts
+before its payload; reserving space only for the payload leaves that earlier
+storage vulnerable.
+
+For an aligned-placement check, confirm that the actual run selected direct I/O.
+A successful buffered fallback does not exercise the widened-window path. Retain
+the compared tensor set and actual checked count; a skipped or absent tensor is
+not byte-identity evidence.
+
+Sources: [fixture geometry and payload](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tools/make_st_fixture.py#L130-L202),
+[independent byte check](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tests/unit/test_cache.c#L137-L186),
+[run placement](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/cache/dsv4_cache.c#L260-L389),
+[aligned and buffered reads](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/io/dsv4_st.c#L470-L508).
+
+## Trace resident and pending ownership
+
+At this source pin, `begin` makes hit/miss, victim and recency decisions serially
+in request order before starting the reads. Slots already claimed by this call
+cannot be chosen again as victims. It returns usable resident experts immediately;
+misses and duplicate requests for an in-flight expert remain unpublished until
+`end` completes that read.
+
+| State | Bytes and visibility | Required boundary |
+| --- | --- | --- |
+| Returned resident | `out[k]` points to a complete claimed slot. | Keep its bytes usable while pending reads run. |
+| Pending miss | The selected slot is marked invalid while a reader fills it; `out[k]` is null. | Publish its key and output only after successful completion in `end`. |
+| Missing or invalid tensor metadata | Resolution fails before choosing a victim. | Surface failure; do not interpret it as a completed cache load. |
+| Partial I/O failure | The chosen slot can contain partial replacement bytes and stays invalid. | Complete the batch wait and reject the unavailable expert; do not reuse its old key. |
+
+Only one batch may be active per cache. Call `end` before another `begin` or any
+other cache operation. Shutdown stops and joins readers before freeing their slot
+storage. When the reader pool is unavailable, the source performs reads inside
+`end`; the same callback pair then supplies no overlap with caller computation.
+Callback selection alone does not prove the asynchronous path executed.
+
+The resident check warms a subset, calls `begin` with both hits and misses,
+copies resident `w1` bytes during the gap, then compares them after `end` and with
+serial loads. This specifically tests those resident `w1` bytes. It does not
+establish every tensor's stability under all eviction, capacity or timing cases.
+The separate serial/concurrent check compares weight and scale bytes and includes
+a duplicate request.
+
+Sources: [single-batch interface](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/cache/dsv4_cache.h#L220-L270),
+[decide, wait and publish](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/cache/dsv4_cache.c#L563-L701),
+[reader cleanup](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/cache/dsv4_cache.c#L199-L210),
+[concurrent and resident checks](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tests/unit/test_cache.c#L188-L298).
+
+## Follow the model consumer and its accumulation order
+
+Check the callback table used by both the real caller and its oracle. The layer
+selects overlap only when both `begin` and `end` exist; otherwise it uses
+`get_many` or serial `get`. A test wired only to `get` cannot cover the overlap
+branch. The CLI and current model-oracle test both supply the overlapped pair.
+
+The caller computes available residents while misses are being read, waits in
+`end`, and then computes the remaining experts. Each expert writes to its own
+per-`k` output storage, separate from the MoE input. The weighted sum happens
+afterwards in `k` order. Completion order must not change that sum: floating-point
+addition can produce different results when reordered. Expert matmuls remain
+sequential because their working scratch is shared; I/O is what overlaps them.
+
+The batched caller deduplicates expert requests and fetches the union in bounded
+chunks, while retaining a separate output for each `(token, k)` and summing in
+`k` order per token. An unavailable expert after the wait aborts the model path;
+silently skipping it would change the routed computation.
+
+Sources: [source interface and scratch](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/model/dsv4_layer.h#L9-L74),
+[per-token consumer](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/model/dsv4_layer.c#L369-L483),
+[batched consumer](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/model/dsv4_layer.c#L714-L831),
+[CLI wiring](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/cli/dsv4_run.c#L313-L315),
+[oracle wiring](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tests/unit/test_model_oracle.c#L57-L83).
+
+## Interpret the numerical evidence at its actual scope
+
+| Check | Source-defined comparison | Interpretation |
+| --- | --- | --- |
+| Cache byte oracle | Buffered tensor bytes versus cached weights/scales. | Read/placement identity for the checked bytes, with shared tensor metadata. |
+| Tiny per-token model | C logits against a repo-owned PyTorch reference, using a vector-scaled tolerance and matching argmax. | Numerical agreement for the synthetic checkpoint and supplied positions; not bit equality to PyTorch. |
+| Batched prefill | Bitwise C-logit comparison against the per-token run after its PyTorch check, across several chunk sizes and fresh layer states. | Batched/per-token agreement within that anchored fixture. |
+
+The tiny checkpoint draws FP8/FP4 weights from exactly representable values and
+uses the real config/reader/binder/trunk/cache entry points. Its dimensions and
+token sequence are synthetic. The repo-owned PyTorch reimplementation omits
+activation-quantization simulation and Hadamard rotation; it is not an execution
+of the released model's official GPU reference. Missing tiny checkpoint or trunk
+resources print `SKIP` and return exit status 0, so an exit code alone is not a
+model-oracle pass.
+
+Return the selected I/O and callback paths, byte/resident coverage, failure and
+cleanup boundaries, numerical comparison and unresolved gaps. Retain the
+campaign's independent acceptance and workload controls. These source
+definitions and historical reports establish no locally executed C/model check,
+full-model quality, device performance or agent efficacy.
+
+Sources: [fixture/reference producer](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tools/make_tiny_checkpoint.py#L17-L50),
+[reference exclusions](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tools/dsv4_ref.py#L3-L30),
+[model comparisons and cleanup](https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tests/unit/test_model_oracle.c#L145-L390).
+
+## Attribution
+
+Original analysis of Ronak Parmar's `ronak-create/deepseek-v4-in-c` at the pinned
+revision. The repository declares Apache-2.0 and attributes its streaming design
+to Fareed Khan's `kimi-k3-in-c`, also Apache-2.0. Selected-file notices and excluded
+third-party/model material remain in the source record. No upstream source code,
+prose, checkpoint or fixture payload is redistributed.
diff --git a/src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-source.json b/src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-source.json
new file mode 100644
index 000000000..54d074b30
--- /dev/null
+++ b/src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/deepseek-expert-cache-source.json
@@ -0,0 +1,361 @@
+{
+ "title": "DeepSeek streamed expert-cache byte and overlap oracle source record",
+ "repository": {
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c",
+ "id": "1340526651",
+ "commit": "d0402056d0da4bf7edfd2d4977de479a0db9d201",
+ "git_tree": "cec1a0fdeab62419ca70269f86706d1572768b4c",
+ "default_license_declaration": "Apache-2.0",
+ "tree_entries": 91,
+ "tree_truncated": false,
+ "verified_blobs": 74
+ },
+ "inspected_on": "2026-11-03",
+ "files": [
+ {
+ "path": "LICENSE",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/LICENSE",
+ "bytes": 11358,
+ "git_blob_sha1": "d645695673349e3947e8e5ae42332d0ac3164cd7",
+ "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
+ "read_regions_one_based": [
+ [
+ 1,
+ 202
+ ]
+ ],
+ "read_scope": "Complete root license declaration",
+ "license_observation": "Complete Apache-2.0 root license text",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "README.md",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/README.md",
+ "bytes": 38462,
+ "git_blob_sha1": "ceeffc5ffaa5b66c190051d4c11e29d428909cca",
+ "sha256": "352c90deefd9bfabd5a5bb0ab1513efbcfd5f6607e7b87eca2694bda4db5ec60",
+ "read_regions_one_based": [
+ [
+ 1,
+ 15
+ ],
+ [
+ 268,
+ 318
+ ],
+ [
+ 799,
+ 801
+ ]
+ ],
+ "read_scope": "Attribution and historical common-corruption/fixture counterexample; author report",
+ "license_observation": "Repository default Apache-2.0; no separate selected-file license marker; no source bytes redistributed",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "Makefile",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/Makefile",
+ "bytes": 13807,
+ "git_blob_sha1": "775d8aac970dead45390c491ca4a44645f96e8b5",
+ "sha256": "2604f1a95c317c2b9347358e92618fdd21a98ede84cd75791ba5cd63d8b0c428",
+ "read_regions_one_based": [
+ [
+ 1,
+ 170
+ ]
+ ],
+ "read_scope": "Build inputs, stale-binary dependency fix, fixture producers, selected test targets; definitions only",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "src/cache/dsv4_cache.c",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/cache/dsv4_cache.c",
+ "bytes": 30440,
+ "git_blob_sha1": "c9397a3b4e5562fddbbf9de1f67e41d47031151f",
+ "sha256": "416220036f925511be98f9b202dc6cd7523e6b11caf1ded5b09829bc641941e1",
+ "read_regions_one_based": [
+ [
+ 1,
+ 721
+ ]
+ ],
+ "read_scope": "Complete byte geometry, placement, resolve/read failure distinction, reader pool, begin/end and shutdown implementation",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "src/cache/dsv4_cache.h",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/cache/dsv4_cache.h",
+ "bytes": 13817,
+ "git_blob_sha1": "f7bd20e497e483013f08fa0f4b1e671d2ad2fd00",
+ "sha256": "3b26e722a62855e8d8f9524e145368cff4fd2c6a84c1aeb2f031ec2835940ba4",
+ "read_regions_one_based": [
+ [
+ 1,
+ 275
+ ]
+ ],
+ "read_scope": "Complete cache data structures and single-batch/lifetime contracts; performance commentary is excluded",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "src/io/dsv4_st.c",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/io/dsv4_st.c",
+ "bytes": 24526,
+ "git_blob_sha1": "aa5aca702e80407e38e73d0e978b62f4ca1549d5",
+ "sha256": "e33d9ef4d30fc56bec29e0e3e1293c119c84cd17bad8b4713f4fdc55c5e4f4eb",
+ "read_regions_one_based": [
+ [
+ 455,
+ 508
+ ]
+ ],
+ "read_scope": "Lookup and aligned/buffered byte-read consumer only; parser implementation not qualified",
+ "license_observation": "Repository default Apache-2.0; no separate selected-file license marker; no source bytes redistributed",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "src/io/dsv4_st.h",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/io/dsv4_st.h",
+ "bytes": 7249,
+ "git_blob_sha1": "ddc87a56a1d3d2cae3dd9481b709c52e74bfda5c",
+ "sha256": "71e67e89f58c12b2293d968119dfefafb800f4798adab44d4beb750facdc0092",
+ "read_regions_one_based": [
+ [
+ 1,
+ 150
+ ]
+ ],
+ "read_scope": "Complete tensor metadata and aligned-read contract, including port attribution",
+ "license_observation": "No SPDX marker; header attributes kimi-k3-in-c port as Apache-2.0; repository default Apache-2.0",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "src/model/dsv4_layer.h",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/model/dsv4_layer.h",
+ "bytes": 10549,
+ "git_blob_sha1": "42a7ae2fceefefbdbc78a73f84f85d6cdcc17c8e",
+ "sha256": "73b35193caec04fe5127aa4d392ea8431b4b90c268accad3ff58c728909ccaaa",
+ "read_regions_one_based": [
+ [
+ 1,
+ 80
+ ]
+ ],
+ "read_scope": "Scratch aliasing and source callback contract",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "src/model/dsv4_layer.c",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/model/dsv4_layer.c",
+ "bytes": 41201,
+ "git_blob_sha1": "c2ad079a852ad45551050f65e88a346105939e68",
+ "sha256": "8eecb6dbc5db356018a299f9cc103fd7115dbd6c17bf83f978b9f32781cea22b",
+ "read_regions_one_based": [
+ [
+ 369,
+ 487
+ ],
+ [
+ 714,
+ 835
+ ]
+ ],
+ "read_scope": "Per-token and batched MoE consumers, per-k storage, ordered accumulation and failure behavior",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "src/cli/dsv4_run.c",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/src/cli/dsv4_run.c",
+ "bytes": 23806,
+ "git_blob_sha1": "0b001241e5a87c6c2b57e3abf4d42ba60d982130",
+ "sha256": "710ad8a4edfadb112b3e3845e8ee6eefd6e1d8638c59774df2d71cbd3164a18f",
+ "read_regions_one_based": [
+ [
+ 50,
+ 80
+ ],
+ [
+ 298,
+ 328
+ ]
+ ],
+ "read_scope": "Actual cache callbacks, one in-flight handle and source wiring",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "tests/unit/test_cache.c",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tests/unit/test_cache.c",
+ "bytes": 14466,
+ "git_blob_sha1": "559fd71edaa602da1c5a6e50fff715e573ccce3d",
+ "sha256": "635d63a60ef9f1170234a2810bb78ea8549b5f993ea91e53b17461ff96019bf5",
+ "read_regions_one_based": [
+ [
+ 1,
+ 305
+ ]
+ ],
+ "read_scope": "Complete test definition, including independent byte Gate 6, serial/concurrent Gate 7 and w1-only resident snapshot Gate 8",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "tests/unit/test_model_oracle.c",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tests/unit/test_model_oracle.c",
+ "bytes": 17599,
+ "git_blob_sha1": "8ebbdf5a0817e8d95085a75e03db0806e533a6d3",
+ "sha256": "fecc5880efb5ff5fb3f62e5273651baf2d0376a4aae73969a2b1167d653f2487",
+ "read_regions_one_based": [
+ [
+ 1,
+ 391
+ ]
+ ],
+ "read_scope": "Complete test definition; overlap wiring, exit-zero SKIP, tolerant PyTorch/argmax and batched bit-equality boundaries",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "tools/make_st_fixture.py",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tools/make_st_fixture.py",
+ "bytes": 13639,
+ "git_blob_sha1": "c0b383fb6618ae2fe14f224554fa4a54b844b94c",
+ "sha256": "bd157bfcf68d8a2a11ebfbb5b872848420a776d2bdcba63d38496cc5850bc889",
+ "read_regions_one_based": [
+ [
+ 1,
+ 249
+ ]
+ ],
+ "read_scope": "Complete synthetic separated-run fixture definition; multiple experts and nonconstant data",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "tools/make_tiny_checkpoint.py",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tools/make_tiny_checkpoint.py",
+ "bytes": 14454,
+ "git_blob_sha1": "6a8f3f7fdec9a46f70c59c7e8d82ac9cbf775bce",
+ "sha256": "2f8d5e238e46de18bc4e409d4c173f1d35ef18602b3589d1312e82d460cdbc7a",
+ "read_regions_one_based": [
+ [
+ 1,
+ 312
+ ]
+ ],
+ "read_scope": "Complete synthetic checkpoint/expected-logit producer; tiny geometry, exact-representable values and repo-owned reference calls",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ },
+ {
+ "path": "tools/dsv4_ref.py",
+ "url": "https://github.com/ronak-create/deepseek-v4-in-c/blob/d0402056d0da4bf7edfd2d4977de479a0db9d201/tools/dsv4_ref.py",
+ "bytes": 16532,
+ "git_blob_sha1": "801bfeaf66e323d7c19ed73a7580f9b360b98c1c",
+ "sha256": "d32cee0a657cc636783661279ce35c4cf205293e352e1e647cec48438c8c18c1",
+ "read_regions_one_based": [
+ [
+ 1,
+ 65
+ ]
+ ],
+ "read_scope": "Reference provenance and omitted activation quantization/Hadamard scope; full mathematics not qualified",
+ "license_observation": "Apache-2.0 SPDX marker in file and repository root declaration",
+ "execution_status": "not_run"
+ }
+ ],
+ "extracted_scope": [
+ "Independent buffered tensor-byte comparison versus shared aligned/coalesced placement; separated-run, different-residue and nonconstant-payload fixture geometry",
+ "Resident and pending expert-slot lifetimes across begin/end, single-batch ownership, metadata versus partial-read failure and join-before-free cleanup",
+ "Actual I/O and reader-pool dispatch, per-token/batched callback reachability, per-k nonaliasing output storage and ordered accumulation",
+ "Tiny model reference, tolerance/argmax, batched bit-equality and exit-zero missing-fixture SKIP boundaries"
+ ],
+ "evidence_levels": {
+ "code_observation": "Pinned file bodies/regions, test definitions, callback dispatch predicates, read/publish/cleanup ownership",
+ "deduction": "A test using get alone cannot select the begin/end overlap branch; equal outputs from shared placement logic do not independently establish correct bytes",
+ "author_reports": "Historical common-corruption bug, gate repair, actual checkpoint layout and numerical/performance results remain attributed upstream reports",
+ "executed_cpu_tests": "none",
+ "device_model_execution": "none",
+ "agent_effects": "not_evaluated"
+ },
+ "claim_boundaries": [
+ "Buffered pread byte oracle shares the safetensors metadata/index; its independence is from aligned/coalesced placement, not from parsing.",
+ "Gate 6 byte-compares six tensors for fixture experts it obtains; do not describe its skip-on-missing behavior as exhaustive coverage.",
+ "Gate 8 snapshots resident w1 weights only; no all-tensor/all-eviction/all-timing proof follows.",
+ "Missing/invalid tensor metadata is rejected before victim selection; a partial I/O failure invalidates the chosen slot and is fatal to the model consumer.",
+ "Only one batch per cache is contracted. Returned residents and claimed pending slots have different lifetimes; end waits and publishes successful misses, and shutdown joins readers before freeing storage.",
+ "Expert computation uses shared scratch sequentially; reads overlap it. Per-k outputs do not alias the MoE input and are summed in k order.",
+ "Model gate wires begin/end, checks tolerant PyTorch logits plus argmax, then checks batched C logits bitwise against its validated per-token C path.",
+ "Tiny checkpoint/reference omit production-scale quality, full model domain, activation quantization and Hadamard behavior. Missing resources print SKIP and return 0."
+ ],
+ "rights": {
+ "repository_default": "Apache-2.0",
+ "attribution": "Ronak Parmar / ronak-create/deepseek-v4-in-c; cache/reader design port attributed upstream to Fareed Khan's kimi-k3-in-c (Apache-2.0)",
+ "redistribution": "Original attributed analysis and immutable links only; no upstream source code, prose, checkpoint/config/fixture payload, reference implementation or third-party header is redistributed",
+ "third_party": [
+ {
+ "path": "third_party/json.h",
+ "inspected_regions": [
+ [
+ 1,
+ 50
+ ]
+ ],
+ "observation": "No separate notice in inspected header; parser is a dependency, not extracted source",
+ "disposition": "exclude; no independent redistribution grant asserted"
+ },
+ {
+ "path": "third_party/dsv4_unicode.h",
+ "inspected_regions": [
+ [
+ 1,
+ 50
+ ]
+ ],
+ "observation": "Generated Unicode 15.0.0/Python 3.12.3 category data",
+ "disposition": "exclude; no data or tokenizer extraction"
+ },
+ {
+ "material": "DeepSeek checkpoints/config payloads and external inference/model.py or inference/kernel.py",
+ "disposition": "exclude; external model/source rights and official numerical equivalence not qualified"
+ }
+ ],
+ "notice_inventory": "No NOTICE or additional license file appears in the complete pinned tree inventory",
+ "scope": "Selected-file observations preserve declared provenance; the root license is not blanket clearance of third-party, checkpoint or external model/reference rights"
+ },
+ "source_script_disposition": [
+ {
+ "path": "tools/make_st_fixture.py",
+ "disposition": "reference-only",
+ "reason": "Explains fixture coverage; no script or generated payload copied"
+ },
+ {
+ "path": "tools/make_tiny_checkpoint.py",
+ "disposition": "reference-only",
+ "reason": "Explains the synthetic model-oracle producer; not executed or bundled"
+ },
+ {
+ "path": "tools/dsv4_ref.py",
+ "disposition": "reference-only",
+ "reason": "Identifies the repo-owned PyTorch comparator and its omissions; not copied or qualified as the official model reference"
+ }
+ ],
+ "existing_owner_boundary": "The weight-loading owner retains byte/readiness/lifetime responsibility. Generic benchmark acceptance, caller/oracle authority and baseline-only agreement limits remain existing inference-benchmark-design responsibilities; this increment creates no second owner or benchmark controller.",
+ "exclusions": [
+ "thermal/order drift",
+ "history/prefetch/useful-miss estimation",
+ "cache policy simulation",
+ "reader pool width performance",
+ "general benchmark controller",
+ "whole model mathematics or acceptance",
+ "new schema/runtime/backend/skill"
+ ],
+ "status": "candidate",
+ "empirical_validation": "not_run",
+ "efficacy": "not_evaluated"
+}