Add support for Gemma 3 models within Fastchat #3705
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mortanius-1
reviewed
Jul 6, 2025
| if device_map == "sequential": | ||
| device_map = "auto" | ||
| # print("From pretrained kwargs", from_pretrained_kwargs) | ||
| tokenizer = AutoTokenizer.from_pretrained(model_path, revision=revision) |
There was a problem hiding this comment.
Hi,
I have a small suggestion:
Suggested change
| tokenizer = AutoTokenizer.from_pretrained(model_path, revision=revision) | |
| tokenizer = AutoTokenizer.from_pretrained(model_path, revision=revision, pad_to_multiple_of=8) |
See this similar issue in huggingface/transformers: huggingface/transformers#36815
Some prompts may trigger an error similar to the following:
ERROR | stderr | Exception in thread Thread-5 (<lambda>):
ERROR | stderr | Traceback (most recent call last):
ERROR | stderr | File "/usr/lib/python3.10/threading.py", line 1016, in _bootstrap_inner
ERROR | stderr | self.run()
ERROR | stderr | File "/usr/lib/python3.10/threading.py", line 953, in run
ERROR | stderr | self._target(*self._args, **self._kwargs)
ERROR | stderr | File "/home/example/projects/FastChat/fastchat/model/model_gemma3.py", line 81, in <lambda>
ERROR | stderr | target=lambda: model.generate(input_ids=input_ids, **generate_kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
ERROR | stderr | return func(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/generation/utils.py", line 2465, in generate
ERROR | stderr | result = self._sample(
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/generation/utils.py", line 3434, in _sample
ERROR | stderr | outputs = model_forward(**model_inputs, return_dict=True)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl
ERROR | stderr | return self._call_impl(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl
ERROR | stderr | return forward_call(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/utils/generic.py", line 965, in wrapper
ERROR | stderr | output = func(self, *args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/utils/deprecation.py", line 172, in wrapped_func
ERROR | stderr | return func(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/models/gemma3/modeling_gemma3.py", line 942, in forward
ERROR | stderr | outputs: BaseModelOutputWithPast = self.model(
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl
ERROR | stderr | return self._call_impl(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl
ERROR | stderr | return forward_call(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/utils/generic.py", line 965, in wrapper
ERROR | stderr | output = func(self, *args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/models/gemma3/modeling_gemma3.py", line 722, in forward
ERROR | stderr | layer_outputs = decoder_layer(
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl
ERROR | stderr | return self._call_impl(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl
ERROR | stderr | return forward_call(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/models/gemma3/modeling_gemma3.py", line 420, in forward
ERROR | stderr | hidden_states, self_attn_weights = self.self_attn(
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl
ERROR | stderr | return self._call_impl(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl
ERROR | stderr | return forward_call(*args, **kwargs)
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/models/gemma3/modeling_gemma3.py", line 342, in forward
ERROR | stderr | attn_output, attn_weights = attention_interface(
ERROR | stderr | File "/home/example/projects/fastchat-venv/lib/python3.10/site-packages/transformers/integrations/sdpa_attention.py", line 54, in sdpa_attention_forward
ERROR | stderr | attn_output = torch.nn.functional.scaled_dot_product_attention(
ERROR | stderr | RuntimeError: p.attn_bias_ptr is not correctly aligned
There was a problem hiding this comment.
Hi,
Thanks for this, we actually ended up creating: https://www.github.com/transformerlab/transformerlab-inference.
We use that instead since fastchat hasn't been merging and stopped new developments.
This model is added on there and works without flash attention which was causing your original issue, please let me know if it also occuses without flash attention too?
|
您好，我最近正在休假中，无法亲自回复您的邮件。我将在假期结束后，尽快给您回复。
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Why are these changes needed?
Added support for Gemma 3 text version. It adds support for inference of all Gemma 3 models (base as well as instruct) for text only mode inference.
Related issue number (if applicable)
Closes #3697