test l20-0 and torchrun

.github/workflows/vllm.yml

@@ -18,7 +18,7 @@ on:
 
 jobs:
   vllm:
-    runs-on: [self-hosted, l20-1]
+    runs-on: [self-hosted, l20-0]
     env:
       HTTP_PROXY: ${{ secrets.PROXY_HTTP }}
       HTTPS_PROXY: ${{ secrets.PROXY_HTTPS }}

tests/rollout/test_vllm_hf_loader.py

@@ -121,6 +121,7 @@ def test_vllm_with_hf():
     print(f'hf response: {tokenizer.batch_decode(response)}')
     print(f'vllm response: {tokenizer.batch_decode(vllm_output)}')
     assert torch.allclose(response, vllm_output), f'hf_response:{response} | vllm_response:{vllm_output}'
+    assert False
     print('Check Pass')