fix init weights issue for critic/reward model (#983)

* Add file extension (#980)

Signed-off-by: Hongwei Chen <hongweichen@microsoft.com>
Signed-off-by: jouw <jouw@foxmail.com>

* fix init weights issue for critic/reward model

Signed-off-by: jouw <jouw@foxmail.com>

* Update submodule link to reflect https style (#981)

Signed-off-by: raviguptaamd <ravi.gupta@amd.com>
Signed-off-by: jouw <jouw@foxmail.com>

* fix formatting issue

Signed-off-by: jouw <jouw@foxmail.com>

---------

Signed-off-by: Hongwei Chen <hongweichen@microsoft.com>
Signed-off-by: jouw <jouw@foxmail.com>
Signed-off-by: raviguptaamd <ravi.gupta@amd.com>
Co-authored-by: Hongwei Chen <33092912+hwchen2017@users.noreply.github.com>
Co-authored-by: raviguptaamd <ravi.gupta@amd.com>

Author: 3 people

applications/DeepSpeed-Chat/dschat/utils/model/model_utils.py

@@ -11,6 +11,7 @@
 )
 from huggingface_hub import snapshot_download
 from transformers.integrations.deepspeed import HfDeepSpeedConfig
+from transformers.modeling_utils import no_init_weights
 
 from dschat.utils.model.reward_model import RewardModel
 from dschat.utils.utils import load_state_dict_into_model, print_rank_0
@@ -99,7 +100,8 @@ def create_hf_model(model_class,
         dschf = None
     if rlhf_training:
         # the weight loading is handled by create critic model
-        model = model_class.from_config(model_config)
+        with no_init_weights():
+            model = model_class.from_config(model_config)
     else:
         model = model_class.from_pretrained(
             model_name_or_path,