Update llama.py

This commit is contained in:
Daniel Han 2025-03-21 16:50:29 -07:00
commit 2c6f0fb24d

View file

@ -1169,7 +1169,8 @@ def CausalLM_fast_forward(fast_forward_inference):
if not return_dict:
output = (logits,) + outputs[1:]
return (loss,) + output if loss is not None else output
print(outputs.past_key_values)
print(outputs.past_key_values, outputs.past_key_values[0][0].shape)
raise
return CausalLMOutputWithPast(
loss = loss,
logits = logits,