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

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

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-02-13 13:46:00 -08:00
parent c0524608c2
commit 3470345624
1 changed files with 9 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
mods/fix-qwen3-coder-next/run.sh
View File

@@ -4,15 +4,16 @@ set -e
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"
# 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"
# Restoring this one because the PR has been reverted in main
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
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
# 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"
cp _triton* /usr/local/lib/python3.12/dist-packages/