Fix grouped KV commit hash in repro bundle

README.md

@@ -35,7 +35,7 @@ The actual vLLM implementation lives here:
 ```text
 https://github.com/bluecopa/vllm-spectral.git
 branch: spectral-codebook-docker
-commit: 008dd7f87446710f1649e4581f0783fd7ead361c
+commit: 008dd7f87fb9de185e536ad30b4d524024ed9b9f
 ```
 
 ## Requirements
@@ -82,7 +82,7 @@ chmod +x /tmp/gemmacut-spectral-bootstrap/scripts/setup_repro_from_hf.sh
 The setup script:
 
 - clones the tested vLLM branch over HTTPS,
-- checks out `008dd7f87446710f1649e4581f0783fd7ead361c`,
+- checks out `008dd7f87fb9de185e536ad30b4d524024ed9b9f`,
 - downloads this repo's sidecar and helper scripts,
 - verifies the sidecar SHA256,
 - writes everything under `$HOST_ROOT`.
@@ -165,7 +165,7 @@ DISABLE_HYBRID_KV_CACHE_MANAGER=0
 kv_cache_dtype=fp8_e4m3
 ```
 
-`DISABLE_HYBRID_KV_CACHE_MANAGER=0` uses the default vLLM hybrid KV cache manager. Commit `008dd7f87446710f1649e4581f0783fd7ead361c` teaches that path to account for Spectral's nonuniform per-layer page sizes with group-local block pools. Set `DISABLE_HYBRID_KV_CACHE_MANAGER=1` only as a fallback/bisect mode.
+`DISABLE_HYBRID_KV_CACHE_MANAGER=0` uses the default vLLM hybrid KV cache manager. Commit `008dd7f87fb9de185e536ad30b4d524024ed9b9f` teaches that path to account for Spectral's nonuniform per-layer page sizes with group-local block pools. Set `DISABLE_HYBRID_KV_CACHE_MANAGER=1` only as a fallback/bisect mode.
 
 Set `HF_HUB_OFFLINE=1` only after the base model and drafter are already cached under `$HOST_ROOT/.cache/huggingface`.
 

manifest.json

@@ -11,7 +11,7 @@
   },
   "vllm_repo": "https://github.com/bluecopa/vllm-spectral.git",
   "vllm_branch": "spectral-codebook-docker",
-  "vllm_commit": "008dd7f87446710f1649e4581f0783fd7ead361c",
+  "vllm_commit": "008dd7f87fb9de185e536ad30b4d524024ed9b9f",
   "previous_github_branch_commit": "85430a3a1d4d9769f75c5e4b73bbbf73bd460caa",
   "base_model": "Intel/gemma-4-31B-it-int4-AutoRound",
   "eagle3_drafter": "RedHatAI/gemma-4-31B-it-speculator.eagle3",

results/candidate_grouped_kv_4k_niah_single_1_500_20260413_073312/result.json

@@ -12,7 +12,7 @@
   "max_seq_length": 4096,
   "max_tokens": 128,
   "model": "gemmacut-spectral",
-  "vllm_commit": "008dd7f87446710f1649e4581f0783fd7ead361c",
+  "vllm_commit": "008dd7f87fb9de185e536ad30b4d524024ed9b9f",
   "disable_hybrid_kv_cache_manager": 0,
   "spectral_cuda_graph": 1,
   "data_path": "/teamspace/studios/this_studio/gemmacut-h100-pilot/ruler_pilot/candidate_grouped_kv_data_4k_500/niah_single_1/validation.jsonl",

scripts/setup_repro_from_hf.sh

@@ -7,7 +7,7 @@ REPO_ID="${REPO_ID:-satya007/gemmacut-spectral}"
 HOST_ROOT="${HOST_ROOT:-$PWD/gemmacut-repro}"
 VLLM_REPO="${VLLM_REPO:-https://github.com/bluecopa/vllm-spectral.git}"
 VLLM_BRANCH="${VLLM_BRANCH:-spectral-codebook-docker}"
-VLLM_COMMIT="${VLLM_COMMIT:-008dd7f87446710f1649e4581f0783fd7ead361c}"
+VLLM_COMMIT="${VLLM_COMMIT:-008dd7f87fb9de185e536ad30b4d524024ed9b9f}"
 SIDECAR_SHA256="${SIDECAR_SHA256:-e47a36c13467cbedf720e7f782b976df3dcda2d989c727113a8315008661a3e4}"
 
 export HF_HUB_DISABLE_XET="${HF_HUB_DISABLE_XET:-1}"