software-engineering/spark-vllm-docker
2
0
Fork 0
Code Issues Actions Packages Projects Releases Wiki Activity

Qwen3-coder-next mod - use a new PR instead of reverting previous one

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-02-13 12:03:44 -08:00
parent 701147b1eb
commit c0524608c2
1 changed files with 9 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
mods/fix-qwen3-coder-next/run.sh
View File

@@ -4,8 +4,15 @@ set -e
echo "Patching Qwen3-Coder-Next crashing on start" echo "Patching Qwen3-Coder-Next crashing on start"
patch -p1 -d /usr/local/lib/python3.12/dist-packages < fix_crash.diff || echo "Patch is not applicable, skipping" patch -p1 -d /usr/local/lib/python3.12/dist-packages < fix_crash.diff || echo "Patch is not applicable, skipping"
echo "Reverting PR #34279 that causes slowness" # echo "Reverting PR #34279 that causes slowness"
patch -p1 -R -d /usr/local/lib/python3.12/dist-packages < fix_slowness.diff || echo "Reversing PR #34279 failed, skipping" # patch -p1 -R -d /usr/local/lib/python3.12/dist-packages < fix_slowness.diff || echo "Reversing PR #34279 failed, skipping"
if grep -q "Cast to int64 to prevent overflow in stride" /usr/local/lib/python3.12/dist-packages/vllm/model_executor/layers/fused_moe/fused_moe.py; then
echo "PR #34507 already applied, skipping."
else
echo "Applying PR #34507 for slowness fix..."
curl -L https://patch-diff.githubusercontent.com/raw/vllm-project/vllm/pull/34507.diff | patch -p1 -d /usr/local/lib/python3.12/dist-packages
fi
echo "Fixing Triton allocator bug" echo "Fixing Triton allocator bug"
cp _triton* /usr/local/lib/python3.12/dist-packages/ cp _triton* /usr/local/lib/python3.12/dist-packages/