Add TP2 MOE benchmarks for AMD. (#5909)
This commit is contained in:
22
.github/workflows/pr-test-amd.yml
vendored
22
.github/workflows/pr-test-amd.yml
vendored
@@ -141,11 +141,31 @@ 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: [
|
||||
|
||||
Reference in New Issue
Block a user