Fix retract for page size > 1 (#4914)
This commit is contained in:
@@ -16,7 +16,7 @@ suites = {
|
||||
TestFile("models/lora/test_lora.py", 76),
|
||||
TestFile("models/lora/test_lora_backend.py", 420),
|
||||
TestFile("models/lora/test_multi_lora_backend.py", 144),
|
||||
TestFile("models/test_embedding_models.py", 119),
|
||||
TestFile("models/test_embedding_models.py", 35),
|
||||
TestFile("models/test_generation_models.py", 103),
|
||||
TestFile("models/test_grok_models.py", 60),
|
||||
TestFile("models/test_qwen_models.py", 82),
|
||||
@@ -38,7 +38,7 @@ suites = {
|
||||
TestFile("test_metrics.py", 32),
|
||||
TestFile("test_mla.py", 92),
|
||||
TestFile("test_mla_deepseek_v3.py", 221),
|
||||
TestFile("test_mla_int8_deepseek_v3.py", 421),
|
||||
TestFile("test_mla_int8_deepseek_v3.py", 522),
|
||||
TestFile("test_mla_flashinfer.py", 395),
|
||||
TestFile("test_mla_fp8.py", 93),
|
||||
TestFile("test_no_chunked_prefill.py", 126),
|
||||
@@ -59,7 +59,7 @@ suites = {
|
||||
TestFile("test_srt_endpoint.py", 94),
|
||||
TestFile("test_torch_compile.py", 76),
|
||||
TestFile("test_torch_compile_moe.py", 85),
|
||||
TestFile("test_torch_native_attention_backend.py", 149),
|
||||
TestFile("test_torch_native_attention_backend.py", 123),
|
||||
TestFile("test_torchao.py", 70),
|
||||
TestFile("test_triton_attention_kernels.py", 4),
|
||||
TestFile("test_triton_attention_backend.py", 134),
|
||||
@@ -76,6 +76,16 @@ suites = {
|
||||
TestFile("test_hicache.py", 60),
|
||||
TestFile("test_hicache_mla.py", 90),
|
||||
],
|
||||
"per-commit-2-gpu": [
|
||||
TestFile("test_data_parallelism.py", 90),
|
||||
TestFile("test_dp_attention.py", 90),
|
||||
TestFile("test_update_weights_from_distributed.py", 100),
|
||||
TestFile("test_verl_engine.py", 100),
|
||||
TestFile("test_patch_torch.py", 30),
|
||||
TestFile("test_moe_ep.py", 220),
|
||||
TestFile("test_mla_tp.py", 420),
|
||||
TestFile("test_lora_tp.py", 300),
|
||||
],
|
||||
"nightly": [
|
||||
TestFile("test_nightly_gsm8k_eval.py"),
|
||||
],
|
||||
|
||||
Reference in New Issue
Block a user