fix: pin models to version, except for gpt-oss

.github/workflows/serverless-model-tests.yml

@@ -48,7 +48,7 @@ jobs:
 
       - name: Run serverless e2e test
         id: e2e_test
-        run: python scripts/serverless_e2e_test.py --config configs/qwen/qwen3_8b.yaml --build
+        run: python scripts/serverless_e2e_test.py --config configs/gpt-oss/gpt_oss_120b.yaml --build
         env:
           RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY_2 }}
           DOCKERHUB_REPO: ${{ vars.DOCKERHUB_REPO || 'runpod' }}

configs/gemma/gemma4_31b_it.yaml

@@ -8,4 +8,5 @@ kv-cache-dtype: fp8
 enforce-eager: false
 enable-prefix-caching: true
 enable-chunked-prefill: true
+vllm-release: v2.22.5 
 speculative-config: '{"model":"RedHatAI/gemma-4-31B-it-speculator.eagle3","method":"eagle3","num_speculative_tokens":3}'

configs/llama/llama3.1_8b_instruct.yaml

@@ -6,5 +6,6 @@ trust-remote-code: true
 quantization: fp8
 kv-cache-dtype: fp8
 enforce-eager: false
+vllm-release: v2.22.5 
 enable-prefix-caching: true
 speculative-config: '{"model":"RedHatAI/Llama-3.1-8B-Instruct-speculator.eagle3","method":"eagle3","num_speculative_tokens":3}'