Skip to content

unsloth: pin #253 in place of #241 to drop the CUDA graph cache cap (stacked on #252) #254

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
danielhanchen wants to merge 3 commits into from
Closed

unsloth: pin #253 in place of #241 to drop the CUDA graph cache cap (stacked on #252) #254

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1654d1f
unsloth: pin #253 in place of #241 to drop the CUDA graph cache cap
danielhanchen Oct 8, 2026
9cdf5f3
Merge pin-moe-cache-auto-repin (#252) into unsloth/pin-253-cuda-graph…
danielhanchen Oct 8, 2026
948d5a0
unsloth: pin #253 at its pushed head 20dd977ea (same tree as the repl…
danielhanchen Oct 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion scripts/unsloth/feature-checks.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -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 and the F16 glm5-next indexer cache under a quantized -ctk (the NextN draft graph is upstream since ggml-org#29928); the first needs a CUDA host, the second has no observable output difference, the last needs a real GLM-5-Next model",
"unslothai#253": "CUDA graph cache key without a count cap (#241 plus the cap removal), macOS row prefetch and the F16 glm5-next indexer cache under a quantized -ctk (the NextN draft graph is upstream since ggml-org#29928); 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 needs a real GLM-5-Next model",
"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",
"unslothai#251": "--moe-cache-mib auto sizes the ggml-org#29887 MoE expert cache from the fit; it only acts when a MoE model spills on a single GPU, the flag is off by default, and no runner in the pipeline has a GPU"
}
Expand Down
2 changes: 1 addition & 1 deletion scripts/unsloth/pr-set.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -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/9dd797259031034c547c0d0dc426776b003731cc",
"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"
]
Expand Down