Drop ModelRunner's duplicated parallel-degree fields and read them via self.ps (#31165)
This commit is contained in:
@@ -530,7 +530,7 @@ def _maybe_prepare_mlp_sync_batch(batch: ScheduleBatch, model_runner):
|
||||
batch,
|
||||
dp_size=model_runner.server_args.dp_size,
|
||||
attn_tp_size=get_parallel().attn_tp_size,
|
||||
attn_cp_size=model_runner.attn_cp_size,
|
||||
attn_cp_size=model_runner.ps.attn_cp_size,
|
||||
tp_group=model_runner.tp_group,
|
||||
get_idle_batch=None,
|
||||
disable_cuda_graph=model_runner.server_args.disable_cuda_graph,
|
||||
|
||||
Reference in New Issue
Block a user