Support Deepseek MoE Model (#689)
This commit is contained in:
@@ -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(
|
||||
|
||||
Reference in New Issue
Block a user