Merge branch 'megatron_v0.11.0' of...

Merge branch 'megatron_v0.11.0' of http://developer.sourcefind.cn/codes/OpenDAS/dcu_megatron into megatron_v0.11.0

dcu_megatron/adaptor/megatron_adaptor.py

@@ -99,7 +99,7 @@ class CoreAdaptation(MegatronAdaptationABC):
         )
         from ..core.models.gpt.gpt_model import (
             gpt_model_forward,
-            gpt_model_init,
+            gpt_model_init_wrapper,
             shared_embedding_or_mtp_embedding_weight
         )
         from ..training.utils import get_batch_on_this_tp_rank

pretrain_gpt.py

@@ -136,8 +136,8 @@ def model_provider(pre_process=True, post_process=True) -> Union[GPTModel, megat
         else:
             mtp_transformer_layer_spec = transformer_layer_spec
 
-        mtp_spec = get_mtp_spec(mtp_transformer_layer_spec, use_te=use_te)
         with build_model_context(**build_model_context_args):
+            config.mtp_spec = get_mtp_spec(mtp_transformer_layer_spec, use_te=use_te)
             model = GPTModel(
                 config=config,
                 transformer_layer_spec=transformer_layer_spec,
@@ -151,8 +151,7 @@ def model_provider(pre_process=True, post_process=True) -> Union[GPTModel, megat
                 position_embedding_type=args.position_embedding_type,
                 rotary_percent=args.rotary_percent,
                 rotary_base=args.rotary_base,
-                rope_scaling=args.use_rope_scaling,
-                mtp_spec=mtp_spec
+                rope_scaling=args.use_rope_scaling
             )
     # model = torch.compile(model,mode='max-autotune-no-cudagraphs')
     print_rank_0(model)