stanrom
/

ShareCaptioner

Feature Extraction

InternLMXComposer

Model card Files Files and versions Community

stanrom commited on May 15, 2024

Commit

e749409

·

verified ·

1 Parent(s): 2c8fb29

'meta' device instead of cuda

Files changed (1) hide show

modeling_InternLM_XComposer.py +1 -1

modeling_InternLM_XComposer.py CHANGED Viewed

@@ -67,7 +67,7 @@ class InternLMXComposerForCausalLM(PreTrainedModel):
             # speed up init llm
             with torch.device('meta'):
                 self.internlm_model = InternLMForCausalLM._from_config(config)
-            self.internlm_model.to_empty(device=config.device).to(torch.float16)
         self.internlm_proj = nn.Linear(4096,
                                     self.internlm_model.config.hidden_size)

             # speed up init llm
             with torch.device('meta'):
                 self.internlm_model = InternLMForCausalLM._from_config(config)
+                self.internlm_model.to_empty(device=config.device).to(torch.float16)
         self.internlm_proj = nn.Linear(4096,
                                     self.internlm_model.config.hidden_size)