[CI] add vllm 0.3.1 into deps build (#1549)

.github/workflows/lmi-dist-deps-build.yml

@@ -79,6 +79,15 @@ jobs:
           cd lmi_vllm
           pip wheel . --no-deps
           cp lmi_vllm-*.whl ../build_artifacts
+      - name: Build vllm 0.3.1 speculative decoding
+        run: |
+          . ./venv/bin/activate
+          git clone https://github.com/ymwangg/vllm -b specdec_v0.3.1
+          cd vllm
+          export TORCH_CUDA_ARCH_LIST="7.5 8.0 8.6 8.9 9.0+PTX"
+          export VLLM_INSTALL_PUNICA_KERNELS=1
+          pip wheel . --no-deps
+          cp vllm-*.whl ../build_artifacts
       - name: Build awq kernels
         run: |
           . ./venv/bin/activate
@@ -114,6 +123,7 @@ jobs:
           aws s3 cp dropout_layer_norm*.whl s3://djl-ai-staging/publish/flash_attn/cu121-pt211/
           aws s3 cp rotary_emb*.whl s3://djl-ai-staging/publish/flash_attn/cu121-pt211/
           aws s3 cp lmi_vllm*.whl s3://djl-ai-staging/publish/lmi_vllm/cu121-pt211/
+          aws s3 cp vllm*.whl s3://djl-ai-staging/publish/vllm/cu121-pt211/
           aws s3 cp awq*.whl s3://djl-ai-staging/publish/awq/cu121-pt211/
 
   stop-runners-p4d: