From 1654d1f8d44d05b5d7014d56abe24ada3dee7255 Mon Sep 17 00:00:00 2001 From: Daniel Han <23090290+danielhanchen@users.noreply.github.com> Date: Thu, 8 Oct 2026 04:37:57 -0700 Subject: [PATCH 1/2] unsloth: pin #253 in place of #241 to drop the CUDA graph cache cap #253 is #241's pinned commit plus one commit that removes the 64 entry cap on the CUDA graph cache, which evicts graphs still in use under --split-mode tensor and makes decode as slow as running without CUDA graphs (unslothai/unsloth#12468). Upstream has no count cap. It replaces #241 rather than following it: a later pin that removes lines an earlier pin adds fails pin_contract.py by design, and #253 already carries everything #241's pin does. --- scripts/unsloth/feature-checks.json | 2 +- scripts/unsloth/pr-set.json | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/scripts/unsloth/feature-checks.json b/scripts/unsloth/feature-checks.json index 364e42c419b3..dead9849c901 100644 --- a/scripts/unsloth/feature-checks.json +++ b/scripts/unsloth/feature-checks.json @@ -99,7 +99,7 @@ "unslothai#152": "per-run mmap of a context's tensors; a memory-layout change with no observable output difference", "unslothai#157": "cudaMemcpyDefault in the ggml_cuda_cpy 2D fast path; needs a CUDA host", "unslothai#158": "ROCm_Host compute buffer type on HIP integrated GPUs; needs a ROCm host", - "unslothai#241": "CUDA graph cache key, macOS row prefetch, F16 glm5-next indexer cache under a quantized -ctk and the glm5-next NextN draft graph; the first needs a CUDA host, the second has no observable output difference, the last two need a real GLM-5-Next NextN head with --spec-type draft-mtp", + "unslothai#253": "CUDA graph cache key without a count cap (#241 plus the cap removal), macOS row prefetch, F16 glm5-next indexer cache under a quantized -ctk and the glm5-next NextN draft graph; the first needs a CUDA host (the cap only shows with --split-mode tensor on two GPUs), the second has no observable output difference, the last two need a real GLM-5-Next NextN head with --spec-type draft-mtp", "unslothai#243": "glm5-next compute buffer size; a memory-layout change with no observable output difference, and the size only shows on a CUDA host with a GLM-5-Next model" } } diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index 471205e5c7ae..f72eca0b7d4f 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -30,7 +30,7 @@ "https://github.com/unslothai/llama.cpp/pull/176/commits/e2519ebc09468b477847a7529834a8da998ed2da", "https://github.com/unslothai/llama.cpp/pull/239/commits/bec216444552a20bc39bf2156fa5396154970608", "https://github.com/unslothai/llama.cpp/pull/240/commits/36175a8325e24303704cfbab1b7d8a0bd30500a5", - "https://github.com/unslothai/llama.cpp/pull/241/commits/b96a713a48313a32c92fd6f77b367ea9d803417b", + "https://github.com/unslothai/llama.cpp/pull/253/commits/0e8ffdc72fce157ef5dbd72134b6171367bee6ce", "https://github.com/unslothai/llama.cpp/pull/243/commits/aba4a4c1db96d56846e86d1169dd35241bafd55d", "https://github.com/unslothai/llama.cpp/pull/247/commits/73c2f733e8beef7dd104de747442eb9e21f2bb96" ] From 948d5a042a147c85298464039918dd32fde43404 Mon Sep 17 00:00:00 2001 From: Daniel Han <23090290+danielhanchen@users.noreply.github.com> Date: Thu, 8 Oct 2026 09:26:15 -0700 Subject: [PATCH 2/2] unsloth: pin #253 at its pushed head 20dd977ea (same tree as the replayed merge) --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index 74f4ae979ee3..2d8e6cc86588 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -30,7 +30,7 @@ "https://github.com/unslothai/llama.cpp/pull/176/commits/e2519ebc09468b477847a7529834a8da998ed2da", "https://github.com/unslothai/llama.cpp/pull/239/commits/bec216444552a20bc39bf2156fa5396154970608", "https://github.com/unslothai/llama.cpp/pull/240/commits/36175a8325e24303704cfbab1b7d8a0bd30500a5", - "https://github.com/unslothai/llama.cpp/pull/253/commits/0e50f701f89c09aa821b82b4e5ed4d0cf5fdc488", + "https://github.com/unslothai/llama.cpp/pull/253/commits/20dd977eae5e1a3b0633bba85f03d8b09692379f", "https://github.com/unslothai/llama.cpp/pull/243/commits/f67cec2bd6806d5294e2abc5d62a16405dd3eb74", "https://github.com/unslothai/llama.cpp/pull/251/commits/13f4c714dff3bf7aa1a83be7490102b7706244ee" ]