Support Deepseek MoE Model (#689)

This commit is contained in:
Liangsheng Yin
2024-07-21 03:09:29 -07:00
committed by GitHub
parent 5a4ef2b5c8
commit eedc12e12e
5 changed files with 519 additions and 23 deletions

View File

@@ -167,7 +167,7 @@ def _set_torch_compile_config():
torch._inductor.config.fx_graph_cache = True # Experimental feature to reduce compilation times, will be on by default in future
# FIXME: tmp workaround
torch._dynamo.config.accumulated_cache_size_limit = 128
torch._dynamo.config.accumulated_cache_size_limit = 256
def launch_server(