ci(test_api_server): update workflow

Signed-off-by: Xin Liu <[email protected]>
LlamaEdge · Jan 9, 2025 · 4e741b0 · 4e741b0
1 parent 3a2142e
commit 4e741b0
Showing 1 changed file with 22 additions and 0 deletions.
diff --git a/.github/workflows/test_api_server.yml b/.github/workflows/test_api_server.yml
@@ -111,11 +111,33 @@ jobs:
         run: |
           hurl --test --jobs 1 --max-time 30s ./tests/test_chat.hurl
 
+      - name: Stop rag-api-server for testing chat completions
+        run: |
+          pkill -f wasmedge
+
+      - name: Start rag-api-server for testing chat completions
+        run: |
+          nohup $HOME/.wasmedge/bin/wasmedge --dir .:. --nn-preload default:GGML:AUTO:Qwen2-1.5B-Instruct-Q3_K_M.gguf --nn-preload embedding:GGML:AUTO:nomic-embed-text-v1.5-f16.gguf rag-api-server.wasm --model-name Qwen2-1.5B-Instruct,nomic-embed-text-v1.5 --ctx-size 4096,512 --batch-size 16,512 --prompt-template chatml,embedding --rag-policy last-user-message --socket-addr 0.0.0.0:8080 > ./start-llamaedge.log 2>&1 &
+          sleep 20
+          cat start-llamaedge.log
+          lsof -i :8080
+
       - name: Run test_embeddings.hurl
         run: |
           hurl --test --jobs 1 ./tests/test_embeddings.hurl
           cat start-llamaedge.log
 
+      - name: Stop rag-api-server for testing chat completions
+        run: |
+          pkill -f wasmedge
+
+      - name: Start rag-api-server for testing chat completions
+        run: |
+          nohup $HOME/.wasmedge/bin/wasmedge --dir .:. --nn-preload default:GGML:AUTO:Qwen2-1.5B-Instruct-Q3_K_M.gguf --nn-preload embedding:GGML:AUTO:nomic-embed-text-v1.5-f16.gguf rag-api-server.wasm --model-name Qwen2-1.5B-Instruct,nomic-embed-text-v1.5 --ctx-size 4096,512 --batch-size 16,512 --prompt-template chatml,embedding --rag-policy last-user-message --socket-addr 0.0.0.0:8080 > ./start-llamaedge.log 2>&1 &
+          sleep 20
+          cat start-llamaedge.log
+          lsof -i :8080
+
       - name: Run test_rag.hurl
         run: |
           hurl --test --jobs 1 --connect-timeout 30s --max-time 60s ./tests/test_rag.hurl