Setting num_return_sequences results in shape mismatch error.
#28
by
Watarungurunnn
- opened
hf_args:
do_sample: true
temperature: 0.8
top_k: 50
top_p: 0.95
num_return_sequences: 30
model = AutoModelForCausalLM.from_pretrained(
model_path,
trust_remote_code=True,
torch_dtype="auto",
device_map="auto",
**model_args,
)
generated_tokens = model.generate(
inputs=input_ids,
pad_token_id=tokenizer.pad_token_id,
**hf_args,
)
Error:
File "/home/user_2942/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
return forward_call(*args, **kwargs)
File "/home/user_2942/.local/lib/python3.10/site-packages/transformers/models/gemma2/modeling_gemma2.py", line 1068, in forward
outputs = self.model(
File "/home/user_2942/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/home/user_2942/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
return forward_call(*args, **kwargs)
File "/home/user_2942/.local/lib/python3.10/site-packages/transformers/models/gemma2/modeling_gemma2.py", line 908, in forward
layer_outputs = decoder_layer(
File "/home/user_2942/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/home/user_2942/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
return forward_call(*args, **kwargs)
File "/home/user_2942/.local/lib/python3.10/site-packages/transformers/models/gemma2/modeling_gemma2.py", line 650, in forward
hidden_states, self_attn_weights, present_key_value = self.self_attn(
File "/home/user_2942/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/home/user_2942/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
return forward_call(*args, **kwargs)
File "/home/user_2942/.local/lib/python3.10/site-packages/transformers/models/gemma2/modeling_gemma2.py", line 252, in forward
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
File "/home/user_2942/.local/lib/python3.10/site-packages/transformers/cache_utils.py", line 1227, in update
return update_fn(
File "/home/user_2942/.local/lib/python3.10/site-packages/transformers/cache_utils.py", line 1202, in _static_update
k_out[:, :, cache_position] = key_states
RuntimeError: shape mismatch: value tensor of shape [30, 16, 942, 128] cannot be broadcast to indexing result of shape [1, 16, 942, 128]
time="2024-07-09T04:45:52 UTC" level=info msg="sub-process exited" argo=true error="<nil>"
Error: exit status 1
Hi,
Changing line 1767 in generation/utils.py to getattr(generation_config, "num_beams", 1) * getattr(generation_config, "num_return_sequences", 1) * batch_size , fixed the problem for me. Hope you find that helpful :)
Any update on this issue? I cannot use a fork of transformers on my project.
It was fixed and released already, just make sure to update transformers 😄
Hi @Watarungurunnn , Could you please confirm if you are facing still issue after updated the transformers, let us know will assist you or else we can close this issue.
Thank you.
Seems fixed! Thank you
Watarungurunnn
changed discussion status to
closed