[CI] Fix B200 CI (#13387)

This commit is contained in:
Baizhou Zhang
2025-11-16 15:13:57 -08:00
committed by GitHub
parent 254f62d879
commit 3ccd7fa669
4 changed files with 87 additions and 62 deletions
+7 -4
View File
@@ -178,15 +178,16 @@ suites = {
TestFile("test_disaggregation_dp_attention.py", 155),
],
"per-commit-4-gpu-b200": [
TestFile("test_deepseek_v3_fp4_4gpu.py", 3600),
TestFile("test_deepseek_v3_fp4_4gpu.py", 1800),
TestFile("test_flash_attention_4.py", 300),
TestFile("test_gpt_oss_4gpu.py", 600),
TestFile("test_llama31_fp4.py", 300),
TestFile("test_eagle_infer_beta_dp_attention.py", 200),
# TODO: Add it back after the bug is fixed
# TestFile("test_eagle_infer_beta_dp_attention.py", 200),
],
"per-commit-4-gpu-gb200": [
TestFile("test_cutedsl_moe.py", 300),
TestFile("test_deepseek_v3_fp4_4gpu.py", 3600),
TestFile("test_deepseek_v3_fp4_4gpu.py", 1800),
# Disabled temporarily, see https://github.com/sgl-project/sglang/issues/12533
# TestFile("test_deepseek_v3_cutedsl_4gpu.py", 3600),
],
@@ -219,9 +220,10 @@ suites = {
],
"nightly-4-gpu-b200": [
TestFile("nightly/test_flashinfer_trtllm_gen_moe_backend.py", 300),
TestFile("test_fp4_moe.py", 300),
TestFile("nightly/test_gpt_oss_4gpu_perf.py", 600),
TestFile("nightly/test_flashinfer_trtllm_gen_attn_backend.py", 300),
TestFile("test_deepseek_v3_fp4_cutlass_moe.py", 900),
TestFile("test_fp4_moe.py", 300),
],
"nightly-8-gpu-b200": [],
"nightly-4-gpu": [
@@ -299,6 +301,7 @@ suites = {
TestFile("test_deepseek_chat_templates.py"),
TestFile("test_disaggregation.py"),
TestFile("test_double_sparsity.py"),
TestFile("test_eagle_infer_beta_dp_attention.py"),
TestFile("test_embedding_openai_server.py"),
TestFile("test_enable_thinking.py"),
TestFile("test_eplb.py"),