@@ -100,21 +100,22 @@ def build_compiler_toolkit_test_list() -> list[OverrideDefinitions]:
100100 "llama3_fsdp_tp_flexattn_autobucketing_regional_inductor" ,
101101 ngpu = 4 ,
102102 ),
103- OverrideDefinitions (
104- [
105- [
106- "--model.name compiler_toolkit.llama3" ,
107- "--parallelism.data_parallel_shard_degree 2" ,
108- "--parallelism.tensor_parallel_degree 2" ,
109- "--model.flavor debugmodel_flex_attn" ,
110- "--job.custom_config_module=torchtitan.experiments.compiler_toolkit.job_config" ,
111- "--compile.passes autobucketing_reordering,regional_inductor,cudagraph" ,
112- ],
113- ],
114- "llama3 FSDP+TP+FlexAttn autobucketing regional_inductor+cudagraph" ,
115- "llama3_fsdp_tp_flexattn_autobucketing_regional_inductor_cudagraph" ,
116- ngpu = 4 ,
117- ),
103+ # TODO: enable this when cudagraph is fixed
104+ # OverrideDefinitions(
105+ # [
106+ # [
107+ # "--model.name compiler_toolkit.llama3",
108+ # "--parallelism.data_parallel_shard_degree 2",
109+ # "--parallelism.tensor_parallel_degree 2",
110+ # "--model.flavor debugmodel_flex_attn",
111+ # "--job.custom_config_module=torchtitan.experiments.compiler_toolkit.job_config",
112+ # "--compile.passes autobucketing_reordering,regional_inductor,cudagraph",
113+ # ],
114+ # ],
115+ # "llama3 FSDP+TP+FlexAttn autobucketing regional_inductor+cudagraph",
116+ # "llama3_fsdp_tp_flexattn_autobucketing_regional_inductor_cudagraph",
117+ # ngpu=4,
118+ # ),
118119 OverrideDefinitions (
119120 [
120121 [
0 commit comments