Merge branch 'develop' into feature/llguidance

Author: ST-XX

.github/workflows/_accuracy_test.yml

@@ -143,7 +143,7 @@ jobs:
           -v "${CACHE_DIR}/ConfigDir:/root/.config" \
           -e TZ="Asia/Shanghai" \
           --gpus '"device='"${DEVICES}"'"' ${docker_image} /bin/bash -xc '
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
 
           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 

.github/workflows/_base_test.yml

@@ -143,7 +143,7 @@ jobs:
           -v "${CACHE_DIR}/ConfigDir:/root/.config" \
           -e TZ="Asia/Shanghai" \
           --gpus '"device='"${DEVICES}"'"' ${docker_image} /bin/bash -xc '
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
 
           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 

.github/workflows/_build_linux.yml

@@ -155,7 +155,7 @@ jobs:
             elif [[ "${PADDLEVERSION}" != "" ]];then
               python -m pip install paddlepaddle-gpu==${PADDLEVERSION} -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
             else
-              python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+              python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
             fi
 
             pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
@@ -164,7 +164,7 @@ jobs:
             python -m pip install -r requirements.txt
             python -m pip install wheel
             # 编译RDMA
-            export ENABLE_FD_RDMA=1
+            export FD_ENABLE_RDMA_COMPILE=1
             bash build.sh 1 python false [${COMPILE_ARCH}]
             ls ./dist/*.whl
             '

.github/workflows/_logprob_test_linux.yml

@@ -134,7 +134,7 @@ jobs:
           -v "${CACHE_DIR}/ConfigDir:/root/.config" \
           -e TZ="Asia/Shanghai" \
           --gpus '"device='"${DEVICES}"'"' ${docker_image} /bin/bash -xc '
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
 
           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 

.github/workflows/_pre_ce_test.yml

@@ -154,7 +154,7 @@ jobs:
           --gpus "\"device=${DEVICES}\"" ${docker_image} /bin/bash -c '
           git config --global --add safe.directory /workspace/FastDeploy
           cd FastDeploy
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
           python -m pip install ${fd_wheel_url}
           bash scripts/run_pre_ce.sh
           '

.github/workflows/_stable_test.yml

@@ -146,7 +146,7 @@ jobs:
           -v "${CACHE_DIR}/ConfigDir:/root/.config" \
           -e TZ="Asia/Shanghai" \
           --gpus '"device='"${DEVICES}"'"' ${docker_image} /bin/bash -xc '
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
 
           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 

.github/workflows/_unit_test_coverage.yml

@@ -45,7 +45,7 @@ jobs:
     needs: check_cov_skip
     if: needs.check_cov_skip.outputs.can-skip != 'true'
     outputs:
-      diff_cov_file_url: ${{ steps.cov_upload.outputs.diff_cov_file_url }}
+      all_cov_file_url: ${{ steps.cov_upload.outputs.all_cov_file_url }}
       unittest_failed_url: ${{ steps.cov_upload.outputs.unittest_failed_url }}
       diff_cov_result_json_url: ${{ steps.cov_upload.outputs.diff_cov_result_json_url }}
     steps:
@@ -174,7 +174,7 @@ jobs:
           git config --global --add safe.directory /workspace/FastDeploy
           cd FastDeploy
           git diff origin/${BASE_REF}..HEAD --unified=0 > diff.txt
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
           pip config set global.extra-index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 
           python -m pip install -r scripts/unittest_requirement.txt
@@ -202,7 +202,7 @@ jobs:
           if [[ "$IS_PR" == "true" ]]; then
             echo "Running diff coverage for PR..."
             diff-cover python_coverage_all.xml --diff-file=diff.txt --fail-under=80 --json-report diff_coverage.json || COVERAGE_EXIT_CODE=9
-            python scripts/generate_diff_coverage_xml.py diff.txt python_coverage_all.xml
+            # python scripts/generate_diff_coverage_xml.py diff.txt python_coverage_all.xml
           else
             echo "Running full coverage"
             coverage report -m > full_coverage_report.txt
@@ -250,15 +250,15 @@ jobs:
 
           target_path_stripped="${target_path#paddle-github-action/}"
 
-          if [[ "$IS_PR" == "true" ]]; then
-            diff_cov_file="diff_coverage.xml"
-            if [ -f ${diff_cov_file} ]; then
-              python ${push_file} ${diff_cov_file} ${target_path}/CoverageData
-              DIFF_COV_FILE_URL=https://paddle-github-action.bj.bcebos.com/${target_path_stripped}/CoverageData/${diff_cov_file}
-              echo "diff_cov_file_url=${DIFF_COV_FILE_URL}" >> $GITHUB_OUTPUT
-              echo "diff_cov_file_url=${DIFF_COV_FILE_URL}" >> $GITHUB_ENV
-            fi
+          all_coverage_file="python_coverage_all.xml"
+          if [ -f ${all_coverage_file} ]; then
+            python ${push_file} ${all_coverage_file} ${target_path}/CoverageData
+            ALL_COV_FILE_URL=https://paddle-github-action.bj.bcebos.com/${target_path_stripped}/CoverageData/${all_coverage_file}
+            echo "all_cov_file_url=${ALL_COV_FILE_URL}" >> $GITHUB_OUTPUT
+            echo "all_cov_file_url=${ALL_COV_FILE_URL}" >> $GITHUB_ENV
+          fi
 
+          if [[ "$IS_PR" == "true" ]]; then
             diff_cov_result_json="diff_coverage.json"
             if [ -f ${diff_cov_result_json} ]; then
               python ${push_file} ${diff_cov_result_json} ${target_path}/CoverageData
@@ -352,28 +352,33 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 15
     env:
-      fd_archive_url: ${{ inputs.FASTDEPLOY_ARCHIVE_URL }}
+      all_cov_file_url: ${{ needs.run_tests_with_coverage.outputs.all_cov_file_url }}
     steps:
-      - name: coverage diff file download
+      - name: Clone FastDeploy
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.10'
+      - name: Download diff coverage file
         shell: bash
-        env:
-          diff_cov_file_url: ${{ needs.run_tests_with_coverage.outputs.diff_cov_file_url }}
         run: |
-          wget -q --no-proxy ${fd_archive_url}
-          tar -xf FastDeploy.tar.gz
-          cd FastDeploy
-          if [ -z "${diff_cov_file_url}" ]; then
-            echo "No diff coverage file URL provided."
+          echo "Downloading all coverage file..."
+          if ! wget --no-proxy "${all_cov_file_url}" -O python_coverage_all.xml; then
+            echo "Download failed, skipping upload."
             exit 0
           fi
-          wget "${diff_cov_file_url}" -O ./diff_coverage.xml || echo "Download cov file failed, but continuing..."
+
+          sed -i 's|<source>/workspace/FastDeploy/fastdeploy</source>|<source>fastdeploy</source>|' python_coverage_all.xml
+
       - name: Upload diff coverage report
-        if: ${{ needs.run_tests_with_coverage.outputs.diff_cov_file_url != null && needs.run_tests_with_coverage.outputs.diff_cov_file_url != '' }}
-        uses: codecov/codecov-action@v5
+        if: always() && hashFiles('python_coverage_all.xml') != ''
+        uses: codecov/codecov-action@v4
         with:
-          files: ./FastDeploy/diff_coverage.xml
+          files: ./python_coverage_all.xml
+          flags: GPU
           name: python diff coverage
+          fail_ci_if_error: false
           verbose: true
           disable_search: true
-          commit_parent: false
-          flags: diff

.github/workflows/ci_xpu.yml

@@ -16,6 +16,7 @@ concurrency:
 
 jobs:
   CI_XPU:
+    timeout-minutes: 60
     runs-on: [self-hosted, XPU-P800-8Card]
     steps:
       - name: Print current runner name

benchmarks/yaml/eb45-32k-wint4-mtp-h800-tp4.yaml

@@ -1,6 +1,6 @@
 max_model_len: 32768
 max_num_seqs: 96
-gpu_memory_utilization: 0.85
+gpu_memory_utilization: 0.8
 kv_cache_ratio: 0.71
 tensor_parallel_size: 4
 quantization: wint4

benchmarks/yaml/qwen3-235b-32k-fp8-tp1-dp4_decode.yaml

@@ -0,0 +1,13 @@
+max_model_len: 32768
+max_num_seqs: 32
+data_parallel_size: 4
+tensor_parallel_size: 1
+enable_expert_parallel: True
+enable_prefix_caching: False
+splitwise_role: decode
+cache_transfer_protocol: "rdma"
+rdma_comm_ports: "7671,7672,7673,7674"
+pd_comm_port: "2335"
+engine_worker_queue_port: "4582,4583,4584,4585"
+graph_optimization_config:
+  use_cudagraph: False