ci: Switch vllm config to qwen3

It preforms better in tool calling and structured tests

Signed-off-by: Derek Higgins <[email protected]>

Author: derekhiggins

.github/actions/setup-vllm/action.yml

@@ -11,13 +11,14 @@ runs:
           --name vllm \
           -p 8000:8000 \
           --privileged=true \
-          quay.io/higginsd/vllm-cpu:65393ee064 \
+          quay.io/higginsd/vllm-cpu:65393ee064-qwen3 \
           --host 0.0.0.0 \
           --port 8000 \
           --enable-auto-tool-choice \
-          --tool-call-parser llama3_json \
-          --model /root/.cache/Llama-3.2-1B-Instruct \
-          --served-model-name meta-llama/Llama-3.2-1B-Instruct
+          --tool-call-parser hermes \
+          --model /root/.cache/Qwen3-0.6B \
+          --served-model-name Qwen/Qwen3-0.6B \
+          --max-model-len 8192
 
           # Wait for vllm to be ready
           echo "Waiting for vllm to be ready..."

scripts/integration-tests.sh

@@ -212,11 +212,6 @@ fi
 echo "=== Running Integration Tests ==="
 EXCLUDE_TESTS="builtin_tool or safety_with_image or code_interpreter or test_rag"
 
-# Additional exclusions for vllm setup
-if [[ "$TEST_SETUP" == "vllm" ]]; then
-    EXCLUDE_TESTS="${EXCLUDE_TESTS} or test_inference_store_tool_calls"
-fi
-
 PYTEST_PATTERN="not( $EXCLUDE_TESTS )"
 if [[ -n "$TEST_PATTERN" ]]; then
     PYTEST_PATTERN="${PYTEST_PATTERN} and $TEST_PATTERN"

tests/integration/suites.py

@@ -78,7 +78,7 @@ class Setup(BaseModel):
             "VLLM_URL": "http://localhost:8000/v1",
         },
         defaults={
-            "text_model": "vllm/meta-llama/Llama-3.2-1B-Instruct",
+            "text_model": "vllm/Qwen/Qwen3-0.6B",
             "embedding_model": "sentence-transformers/all-MiniLM-L6-v2",
         },
     ),