[AMD] Speed up DSV4 MoE weight loading from mmap views (#32315)
This commit is contained in:
Bingxu Chen
@@ -493,6 +493,7 @@ jobs:
|
||||
timeout-minutes: 180
|
||||
run: |
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-gpt-oss --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
@@ -1317,6 +1318,7 @@ jobs:
|
||||
echo "## SGLANG_HACK_FLASHMLA_BACKEND=unified_kv_triton" >> github_summary.md
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_HACK_FLASHMLA_BACKEND=unified_kv_triton \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-v4-flash --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
@@ -1329,6 +1331,7 @@ jobs:
|
||||
echo "## SGLANG_HACK_FLASHMLA_BACKEND=triton" >> github_summary.md
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_HACK_FLASHMLA_BACKEND=triton \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-v4-flash --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
@@ -1366,6 +1369,7 @@ jobs:
|
||||
echo "## SGLANG_HACK_FLASHMLA_BACKEND=unified_kv_triton" >> github_summary.md
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_HACK_FLASHMLA_BACKEND=unified_kv_triton \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-v4-pro --nightly --timeout-per-file 14400 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
@@ -1378,6 +1382,7 @@ jobs:
|
||||
echo "## SGLANG_HACK_FLASHMLA_BACKEND=triton" >> github_summary.md
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_HACK_FLASHMLA_BACKEND=triton \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-v4-pro --nightly --timeout-per-file 14400 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
@@ -1415,6 +1420,7 @@ jobs:
|
||||
echo "## SGLANG_HACK_FLASHMLA_BACKEND=unified_kv_triton" >> github_summary.md
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_HACK_FLASHMLA_BACKEND=unified_kv_triton \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-v4-pro-mtp --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
@@ -1427,6 +1433,7 @@ jobs:
|
||||
echo "## SGLANG_HACK_FLASHMLA_BACKEND=triton" >> github_summary.md
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_HACK_FLASHMLA_BACKEND=triton \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-v4-pro-mtp --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
@@ -1536,6 +1543,7 @@ jobs:
|
||||
run: |
|
||||
> github_summary.md # Clear summary file
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-8-gpu-qwen3-235b --nightly --timeout-per-file 3600 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
@@ -1610,6 +1618,7 @@ jobs:
|
||||
run: |
|
||||
> github_summary.md # Clear summary file
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-qwen35 --nightly --timeout-per-file 3600 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
|
||||
@@ -1464,6 +1464,7 @@ jobs:
|
||||
run: |
|
||||
> github_summary.md # Clear summary file
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-qwen35 --nightly --timeout-per-file 3600 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
|
||||
@@ -1279,6 +1279,7 @@ jobs:
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
-e SGLANG_DSV4_ACCURACY_ONLY=1 \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-v4-flash --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
exit ${TEST_EXIT_CODE:-0}
|
||||
@@ -1330,6 +1331,7 @@ jobs:
|
||||
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
|
||||
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
|
||||
-e SGLANG_DSV4_ACCURACY_ONLY=1 \
|
||||
-e SGLANG_MOE_COPY_WEIGHT_VIEWS_BEFORE_H2D=1 \
|
||||
python3 registered/amd/test_deepseek_v4_pro_fp4.py || TEST_EXIT_CODE=$?
|
||||
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
|
||||
if [[ "${{ github.event_name == 'schedule' || inputs.continue_on_error }}" == "true" ]]; then
|
||||
|
||||
Reference in New Issue
Block a user