Add position_ids arg to OPTEmbedding forward function (#6939)

This PR updates the DeepSpeed `OPTEmbedding` forward function to include
a new `positions_ids` argument.

---------

Co-authored-by: Logan Adams <loadams@microsoft.com>

.github/workflows/nv-ds-chat.yml

@@ -43,7 +43,7 @@ jobs:
 
       - name: Install deepspeed
         run: |
-          pip install transformers==4.45.2
+          pip install transformers
           pip install .[dev]
           ds_report
 

deepspeed/module_inject/layers.py

@@ -191,7 +191,7 @@ class OPTEmbedding(EmbeddingLayer):
         self.offset = 2
         super().__init__(weight_shape, weight=weight)
 
-    def forward(self, attention_mask: torch.LongTensor, past_key_values_length: int = 0):
+    def forward(self, attention_mask: torch.LongTensor, past_key_values_length: int = 0, position_ids: int = 0):
         """`input_ids_shape` is expected to be [bsz x seqlen]."""
         attention_mask = attention_mask.long()