Files
spark-vllm-docker/mods/fix-qwen3-coder-next/run.sh

19 lines
856 B
Bash

#!/bin/bash
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"
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/