Add dummy grok test to amd CI. (#5115)

.github/workflows/pr-test-amd.yml

@@ -7,12 +7,14 @@ on:
       - "python/sglang/**"
       - "test/**"
       - "sgl-kernel/**"
+      - ".github/workflows/pr-test-amd.yml"
   pull_request:
     branches: [ main ]
     paths:
       - "python/sglang/**"
       - "test/**"
       - "sgl-kernel/**"
+      - ".github/workflows/pr-test-amd.yml"
   workflow_dispatch:
 
 concurrency:
@@ -36,12 +38,12 @@ jobs:
           else
             DEVICE_FLAG="--device /dev/dri"
           fi
-          docker pull lmsysorg/sglang:v0.4.3.post4-rocm630
+          docker pull lmsysorg/sglang:v0.4.5-rocm630
           docker run -dt --user root --device=/dev/kfd $DEVICE_FLAG \
             -v ${{ github.workspace }}:/sglang-checkout --ipc=host --group-add video \
             --cap-add=SYS_PTRACE -e HF_TOKEN=${HF_TOKEN} --security-opt seccomp=unconfined \
             -w /sglang-checkout --name ci_sglang \
-            lmsysorg/sglang:v0.4.3.post4-rocm630
+            lmsysorg/sglang:v0.4.5-rocm630
 
       - name: Install dependencies
         run: |
@@ -53,12 +55,17 @@ jobs:
           docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
           docker exec -w /human-eval ci_sglang pip install -e .
 
+          docker exec -w / ci_sglang mkdir -p /dummy-grok
+          mkdir -p dummy-grok && wget https://sharkpublic.blob.core.windows.net/sharkpublic/sglang/dummy_grok.json -P dummy-grok
+          docker cp ./dummy-grok ci_sglang:/dummy-grok/
+
       - name: Evaluate Accuracy
         timeout-minutes: 20
         run: |
           docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 test_eval_accuracy_large.py
           docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 test_eval_fp8_accuracy.py
           docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 models/test_qwen_models.py
+          docker exec -w /sglang-checkout -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m sglang.bench_one_batch --batch-size 32 --input 1024 --output 8 --model /dummy-grok --tokenizer-path Xenova/grok-1-tokenizer --load-format dummy --tp 8 --quantization fp8
 
   mla-test-1-gpu-amd:
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
@@ -76,12 +83,12 @@ jobs:
           else
             DEVICE_FLAG="--device /dev/dri"
           fi
-          docker pull lmsysorg/sglang:v0.4.3.post4-rocm630
+          docker pull lmsysorg/sglang:v0.4.5-rocm630
           docker run -dt --user root --device=/dev/kfd $DEVICE_FLAG \
             -v ${{ github.workspace }}:/sglang-checkout --ipc=host --group-add video \
             --cap-add=SYS_PTRACE -e HF_TOKEN=${{ secrets.AMD_HF_TOKEN }} --security-opt seccomp=unconfined \
             -w /sglang-checkout --name ci_sglang \
-            lmsysorg/sglang:v0.4.3.post4-rocm630
+            lmsysorg/sglang:v0.4.5-rocm630
 
       - name: Install dependencies
         run: |