Another fix for the Qwen mod as the slow PR was reversed in main

This commit is contained in:
Eugene Rakhmatulin
2026-02-13 13:46:00 -08:00
parent c0524608c2
commit 3470345624

View File

@@ -4,15 +4,16 @@ 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" # Restoring this one because the PR has been reverted in main
# patch -p1 -R -d /usr/local/lib/python3.12/dist-packages < fix_slowness.diff || echo "Reversing PR #34279 failed, skipping" echo "Reverting PR #34279 that causes slowness"
patch -p1 -R -d /usr/local/lib/python3.12/dist-packages < fix_slowness.diff || echo "Can't revert PR #34279, skipping as it was reverted in recent commits"
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 # 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." # echo "PR #34507 already applied, skipping."
else # else
echo "Applying PR #34507 for slowness fix..." # 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 # 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 # 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/