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

Now using an opened PR for glm-4.7-flash crash fix in the mod

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-02-17 12:45:17 -08:00
parent 1e7f2d5640
commit ef07046d51
1 changed files with 12 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
mods/fix-glm-4.7-flash-AWQ/run.sh
View File

@@ -3,6 +3,16 @@ set -e
echo "--- Applying GLM 4.7 AWQ speed patch..."
patch -p1 -d / < glm47_flash.patch
echo "=== OK"
echo "--- Applying vLLM crash patch..."
patch -p1 -d /usr/local/lib/python3.12/dist-packages < glm47_vllm_bug.patch || echo "=== Patch is not applicable, skipping"
echo "--- Applying vLLM crash patch (34695)..."
# Check if PR 34695 is already applied by looking for the changed file
if [ -f /usr/local/lib/python3.12/dist-packages/vllm/model_executor/layers/attention/mla_attention.py ]; then
# Check if the specific line from PR 34695 is already present
if grep -q "and hasattr(self.kv_b_proj, \"weight\")" /usr/local/lib/python3.12/dist-packages/vllm/model_executor/layers/attention/mla_attention.py 2>/dev/null; then
echo "=== PR 34695 is already applied, skipping"
else
curl -L https://patch-diff.githubusercontent.com/raw/vllm-project/vllm/pull/34695.diff | patch -p1 -d /usr/local/lib/python3.12/dist-packages || echo "=== Warning: Failed to apply PR 34695, continuing..."
fi
else
curl -L https://patch-diff.githubusercontent.com/raw/vllm-project/vllm/pull/34695.diff | patch -p1 -d /usr/local/lib/python3.12/dist-packages || echo "=== Warning: Failed to apply PR 34695, continuing..."
fi
echo "=== OK"