MPTForCausalLM.forward() got an unexpected keyword argument 'inputs_embeds'

[leehanchung]

MPTForCausalLM.forward() got an unexpected keyword argument 'inputs_embeds'

[bhupendrasalesken]

i was able to run it with editing the : envs/lora-instruct/lib/python3.10/site-packages/accelerate/hooks.py, but not very sure about the repercussions.

def new_forward(*args, **kwargs):
160         print(f"old args ={args} kwargs={kwargs}")
161         args, kwargs = module._hf_hook.pre_forward(module, *args, **kwargs)
             # edited next line
165         kwargs.pop("inputs_embeds",None)
166         if module._hf_hook.no_grad:
167             with torch.no_grad():
168                 output = old_forward(*args, **kwargs)
169         else:
170             output = old_forward(*args, **kwargs)
171         return module._hf_hook.post_forward(module, output)
172
173     module.forward = new_forward
174     return module