openbmb
/

MiniCPM3-4B

Text Generation

Model card Files Files and versions Community

FrankC0st1e commited on 14 days ago

Commit

fc345c8

•

1 Parent(s): 65d6b09

fix loss in modeling_minicpm.py

Files changed (1) hide show

modeling_minicpm.py +1 -1

modeling_minicpm.py CHANGED Viewed

@@ -1343,7 +1343,7 @@ class MiniCPM3ForCausalLM(MiniCPM3PreTrainedModel):
             shift_logits = logits[..., :-1, :].contiguous()
             shift_labels = labels[..., 1:].contiguous()
             # Flatten the tokens
-            loss_fct = CrossEntropyLoss(reduction="none")
             shift_logits = shift_logits.view(-1, self.config.vocab_size)
             shift_labels = shift_labels.view(-1)
             # Enable model parallelism

             shift_logits = logits[..., :-1, :].contiguous()
             shift_labels = labels[..., 1:].contiguous()
             # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
             shift_logits = shift_logits.view(-1, self.config.vocab_size)
             shift_labels = shift_labels.view(-1)
             # Enable model parallelism