pi314ever
diff --git a/‎.devcontainer/devcontainer.json
Lines changed: 7 additions & 1 deletion b/‎.devcontainer/devcontainer.json
Lines changed: 7 additions & 1 deletion
diff --git a/‎.github/CODEOWNERS
Lines changed: 4 additions & 4 deletions b/‎.github/CODEOWNERS
Lines changed: 4 additions & 4 deletions
diff --git a/‎.github/workflows/execute-notebook.yml
Lines changed: 2 additions & 5 deletions b/‎.github/workflows/execute-notebook.yml
Lines changed: 2 additions & 5 deletions
diff --git a/‎.github/workflows/nightly-test-amd.yml
Lines changed: 58 additions & 0 deletions b/‎.github/workflows/nightly-test-amd.yml
Lines changed: 58 additions & 0 deletions
diff --git a/‎.github/workflows/pr-test-amd.yml
Lines changed: 195 additions & 11 deletions b/‎.github/workflows/pr-test-amd.yml
Lines changed: 195 additions & 11 deletions
@@ -20,5 +20,11 @@
     "runArgs": [
         "--gpus",
         "all"
-    ]
+    ],
+    // The two lines below ensures that your local changes in the sglang
+    // repo is automatically synced to the sglang pip package installed
+    // in the dev docker container. You can remove / comment out these
+    // two lines if you prefer to sync code changes manually.
+    "workspaceMount": "source=${localWorkspaceFolder},target=/sgl-workspace/sglang,type=bind",
+    "workspaceFolder": "/sgl-workspace/sglang"
 }
@@ -4,16 +4,16 @@
 /python/sglang/lang @merrymercy @Ying1123 @hnyls2002 @ByronHsu
 /python/sglang/srt @merrymercy @Ying1123 @hnyls2002 @zhyncs @ispobock @ByronHsu
 /python/sglang/srt/constrained @hnyls2002
-/python/sglang/srt/layers @merrymercy @Ying1123 @zhyncs @ispobock @HaiShaw
-/python/sglang/srt/lora @Ying1123
+/python/sglang/srt/layers @merrymercy @Ying1123 @zhyncs @ispobock @HaiShaw @ch-wan
+/python/sglang/srt/lora @Ying1123 @Fridge003
 /python/sglang/srt/managers @merrymercy @Ying1123 @hnyls2002 @xiezhq-hermann
 /python/sglang/srt/mem_cache @merrymercy @Ying1123 @hnyls2002 @xiezhq-hermann
 /python/sglang/srt/model_executor @merrymercy @Ying1123 @hnyls2002 @zhyncs @ispobock
 /python/sglang/srt/models @merrymercy @Ying1123 @hnyls2002 @zhyncs @ispobock @ByronHsu
-/python/sglang/srt/openai_api @merrymercy @Ying1123 @hnyls2002 @zhyncs @ispobock @ByronHsu
+/python/sglang/srt/openai_api @merrymercy @Ying1123 @hnyls2002 @zhyncs @ispobock @ByronHsu @CatherineSue
 /python/sglang/srt/sampling @merrymercy @hnyls2002
 /python/sglang/srt/speculative @Ying1123 @merrymercy @rkooo567 @kssteven418
 /test/lang @merrymercy @Ying1123 @ByronHsu
 /test/srt @merrymercy @Ying1123 @zhyncs
-/sgl-router @ByronHsu @Ying1123
+/sgl-router @ByronHsu @Ying1123 @slin1237
 /sgl-kernel @zhyncs @ispobock @HandH1998 @BBuf @yizhang2077 @merrymercy @yinfan98
@@ -22,11 +22,6 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
-      - name: Set up Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: '3.9'
-
       - name: Install dependencies
         run: |
           bash scripts/ci_install_dependency.sh
@@ -35,6 +30,8 @@ jobs:
           apt-get install -y pandoc
           apt-get update && apt-get install -y parallel retry
 
+          ln -sf "$(which python3)" /usr/bin/python
+
       - name: Setup Jupyter Kernel
         run: |
           python -m ipykernel install --user --name python3 --display-name "Python 3"
 
@@ -0,0 +1,58 @@
+name: Nightly Test (AMD)
+
+on:
+  schedule:
+    - cron: '0 0 * * *'
+  push:
+    branches:
+      - main
+    paths:
+      - "python/sglang/version.py"
+  workflow_dispatch:
+
+concurrency:
+  group: nightly-test-amd-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  nightly-test:
+    if: github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request'
+    strategy:
+      matrix:
+        runner: [linux-mi300-gpu-2, linux-mi325-gpu-2-nightly]
+    runs-on: ${{matrix.runner}}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Setup docker
+        run: |
+          # Ensure GPU isolation if pod is part of kubernetes setup with DEVICE_FLAG.
+          if [ -f "/etc/podinfo/gha-render-devices" ]; then
+            DEVICE_FLAG=$(cat /etc/podinfo/gha-render-devices)
+          else
+            DEVICE_FLAG="--device /dev/dri"
+          fi
+          touch github_summary.md
+          docker pull lmsysorg/sglang:v0.4.6.post3-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:v0.4.6.post3-rocm630
+
+      - name: Install dependencies
+        run: |
+          docker exec ci_sglang pip install --upgrade pip
+          docker exec ci_sglang pip uninstall sgl-kernel -y || true
+          docker exec -w /sglang-checkout/sgl-kernel ci_sglang bash -c "rm -f pyproject.toml && mv pyproject_rocm.toml pyproject.toml && python3 setup_rocm.py install"
+          docker exec ci_sglang pip install -e "python[dev_hip]"
+
+          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 ci_sglang pip install huggingface_hub[hf_xet]
+
+      - name: Nightly Test
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" ci_sglang python3 run_suite.py --suite nightly-amd --timeout-per-file 7200
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY
@@ -25,7 +25,10 @@ jobs:
   accuracy-test-1-gpu-amd:
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
       github.event.pull_request.draft == false
-    runs-on: linux-mi300-gpu-1
+    strategy:
+      matrix:
+        runner: [linux-mi300-gpu-1, linux-mi325-gpu-1]
+    runs-on: ${{matrix.runner}}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -38,12 +41,12 @@ jobs:
           else
             DEVICE_FLAG="--device /dev/dri"
           fi
-          docker pull ghcr.io/saienduri/sglang-aiter-v0.1.1:428
+          docker pull lmsysorg/sglang:v0.4.6.post3-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 \
-            ghcr.io/saienduri/sglang-aiter-v0.1.1:428
+            lmsysorg/sglang:v0.4.6.post3-rocm630
 
       - name: Install dependencies
         run: |
@@ -66,10 +69,54 @@ jobs:
           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
 
+  accuracy-test-2-gpu-amd:
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
+        github.event.pull_request.draft == false
+    strategy:
+      matrix:
+        runner: [linux-mi300-gpu-2, linux-mi325-gpu-2]
+    runs-on: ${{matrix.runner}}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Setup docker
+        run: |
+          # Ensure GPU isolation if pod is part of kubernetes setup with DEVICE_FLAG.
+          if [ -f "/etc/podinfo/gha-render-devices" ]; then
+            DEVICE_FLAG=$(cat /etc/podinfo/gha-render-devices)
+          else
+            DEVICE_FLAG="--device /dev/dri"
+          fi
+          docker pull lmsysorg/sglang:v0.4.6.post3-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:v0.4.6.post3-rocm630
+
+      - name: Install dependencies
+        run: |
+          docker exec ci_sglang pip install --upgrade pip
+          docker exec ci_sglang pip uninstall sgl-kernel -y || true
+          docker exec -w /sglang-checkout/sgl-kernel ci_sglang bash -c "rm -f pyproject.toml && mv pyproject_rocm.toml pyproject.toml && python3 setup_rocm.py install"
+          docker exec ci_sglang pip install -e "python[dev_hip]"
+
+          docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
+          docker exec -w /human-eval ci_sglang pip install -e .
+
+      - name: Evaluate accuracy (TP=2)
+        timeout-minutes: 20
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 test_moe_eval_accuracy_large.py
+
   mla-test-1-gpu-amd:
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
       github.event.pull_request.draft == false
-    runs-on: linux-mi300-gpu-1
+    strategy:
+      matrix:
+        runner: [linux-mi300-gpu-1, linux-mi325-gpu-1]
+    runs-on: ${{matrix.runner}}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -82,12 +129,12 @@ jobs:
           else
             DEVICE_FLAG="--device /dev/dri"
           fi
-          docker pull ghcr.io/saienduri/sglang-aiter-v0.1.1:428
+          docker pull lmsysorg/sglang:v0.4.6.post3-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 \
-            ghcr.io/saienduri/sglang-aiter-v0.1.1:428
+            lmsysorg/sglang:v0.4.6.post3-rocm630
 
       - name: Install dependencies
         run: |
@@ -104,10 +151,126 @@ jobs:
         run: |
           docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 test_mla.py
 
+  performance-test-1-gpu-part-1-amd:
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
+        github.event.pull_request.draft == false
+    strategy:
+      matrix:
+        runner: [linux-mi300-gpu-1, linux-mi325-gpu-1]
+    runs-on: ${{matrix.runner}}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Setup docker
+        run: |
+          # Ensure GPU isolation if pod is part of kubernetes setup with DEVICE_FLAG.
+          if [ -f "/etc/podinfo/gha-render-devices" ]; then
+            DEVICE_FLAG=$(cat /etc/podinfo/gha-render-devices)
+          else
+            DEVICE_FLAG="--device /dev/dri"
+          fi
+          docker pull lmsysorg/sglang:v0.4.6.post3-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:v0.4.6.post3-rocm630
+
+      - name: Install dependencies
+        run: |
+          docker exec ci_sglang pip install --upgrade pip
+          docker exec ci_sglang pip uninstall sgl-kernel -y || true
+          docker exec -w /sglang-checkout/sgl-kernel ci_sglang bash -c "rm -f pyproject.toml && mv pyproject_rocm.toml pyproject.toml && python3 setup_rocm.py install"
+          docker exec ci_sglang pip install -e "python[dev_hip]"
+
+          docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
+          docker exec -w /human-eval ci_sglang pip install -e .
+
+      - name: Benchmark single latency
+        timeout-minutes: 10
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_AMD_CI=1 -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_bs1_small
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_AMD_CI=1 -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_bs1_default
+
+      - name: Benchmark online latency
+        timeout-minutes: 10
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_AMD_CI=1 -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_serving.TestBenchServing.test_online_latency_default
+
+      - name: Benchmark offline throughput
+        timeout-minutes: 10
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_AMD_CI=1 -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_default
+
+      - name: Benchmark offline throughput (Non-streaming, small batch size)
+        timeout-minutes: 10
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_AMD_CI=1 -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_non_stream_small_batch_size
+
+      - name: Benchmark online latency (EAGLE)
+        timeout-minutes: 10
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_AMD_CI=1 -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_serving.TestBenchServing.test_online_latency_eagle
+
+  performance-test-1-gpu-part-2-amd:
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
+        github.event.pull_request.draft == false
+    strategy:
+      matrix:
+        runner: [linux-mi300-gpu-1, linux-mi325-gpu-1]
+    runs-on: ${{matrix.runner}}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Setup docker
+        run: |
+          # Ensure GPU isolation if pod is part of kubernetes setup with DEVICE_FLAG.
+          if [ -f "/etc/podinfo/gha-render-devices" ]; then
+            DEVICE_FLAG=$(cat /etc/podinfo/gha-render-devices)
+          else
+            DEVICE_FLAG="--device /dev/dri"
+          fi
+          docker pull lmsysorg/sglang:v0.4.6.post3-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:v0.4.6.post3-rocm630
+
+      - name: Install dependencies
+        run: |
+          docker exec ci_sglang pip install --upgrade pip
+          docker exec ci_sglang pip uninstall sgl-kernel -y || true
+          docker exec -w /sglang-checkout/sgl-kernel ci_sglang bash -c "rm -f pyproject.toml && mv pyproject_rocm.toml pyproject.toml && python3 setup_rocm.py install"
+          docker exec ci_sglang pip install -e "python[dev_hip]"
+
+          docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
+          docker exec -w /human-eval ci_sglang pip install -e .
+
+      - name: Benchmark offline throughput (w/o RadixAttention)
+        timeout-minutes: 10
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_AMD_CI=1 -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_without_radix_cache
+
+      - name: Benchmark offline throughput (w/ Triton)
+        timeout-minutes: 10
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_AMD_CI=1 -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_with_triton_attention_backend
+
+      - name: Benchmark offline throughput (w/ FP8)
+        timeout-minutes: 10
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_AMD_CI=1 -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_default_fp8
+
   bench-test-2-gpu-amd:
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
       github.event.pull_request.draft == false
-    runs-on: linux-mi300-gpu-2
+    strategy:
+      matrix:
+        runner: [linux-mi300-gpu-2, linux-mi325-gpu-2]
+    runs-on: ${{matrix.runner}}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -120,12 +283,12 @@ jobs:
           else
             DEVICE_FLAG="--device /dev/dri"
           fi
-          docker pull ghcr.io/saienduri/sglang-aiter-v0.1.1:428
+          docker pull lmsysorg/sglang:v0.4.6.post3-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 \
-            ghcr.io/saienduri/sglang-aiter-v0.1.1:428
+            lmsysorg/sglang:v0.4.6.post3-rocm630
 
       - name: Install dependencies
         run: |
@@ -141,15 +304,36 @@ jobs:
           mkdir -p dummy-grok && wget https://sharkpublic.blob.core.windows.net/sharkpublic/sglang/dummy_grok.json -O dummy-grok/config.json
           docker cp ./dummy-grok ci_sglang:/
 
-      - name: Evaluate Benchmark
+      - name: Benchmark dummy grok (TP=2)
         timeout-minutes: 20
         run: |
           docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 models/test_dummy_grok_models.py
 
+      - name: Benchmark single latency (TP=2)
+        timeout-minutes: 20
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 -e SGLANG_AMD_CI=1 ci_sglang python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_moe_tp2_bs1
+
+      - name: Benchmark single latency + torch.compile (TP=2)
+        timeout-minutes: 20
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_torch_compile_tp2_bs1
+
+      - name: Benchmark offline throughput (TP=2)
+        timeout-minutes: 20
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 -e SGLANG_AMD_CI=1 ci_sglang python3 -m unittest test_bench_serving.TestBenchServing.test_moe_offline_throughput_default
+
+      - name: Benchmark offline throughput (w/o RadixAttention) (TP=2)
+        timeout-minutes: 20
+        run: |
+          docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 -e SGLANG_AMD_CI=1 ci_sglang python3 -m unittest test_bench_serving.TestBenchServing.test_moe_offline_throughput_without_radix_cache
+
   finish:
     if: always()
     needs: [
-      accuracy-test-1-gpu-amd, mla-test-1-gpu-amd, bench-test-2-gpu-amd
+      accuracy-test-1-gpu-amd, mla-test-1-gpu-amd, bench-test-2-gpu-amd,
+      accuracy-test-2-gpu-amd, performance-test-1-gpu-part-1-amd, performance-test-1-gpu-part-2-amd
     ]
     runs-on: ubuntu-latest
     steps: