[CI] Set max transformers version for Ultravox model test

[ywang96]

We need to upgrade the transformers version to 4.51.0 so that Llama-4 can work in our nightly after #16113 is merged. However, Ultravox impl on huggingface is currently breaking on 4.51.0 from the standard model test, so this PR sets the max transformers version for this model to unblock other PRs.

I have verified the same test works on 4.50.3

cc @farzadab - could you maybe take a look and see if the model is still working on the latest transformers version?

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[ywang96]

Full trace:

tests/models/decoder_only/audio_language/test_ultravox.py:126: in run_test
    hf_model.generate_greedy_logprobs_limit(          <----- huggingface impl is broken
tests/conftest.py:585: in generate_greedy_logprobs_limit
    output = self.model.generate(
/tmp-nvme/vllm/lib/python3.12/site-packages/torch/utils/_contextlib.py:116: in decorate_context
    return func(*args, **kwargs)
/tmp-nvme/vllm/lib/python3.12/site-packages/transformers/generation/utils.py:2460: in generate
    result = self._sample(
/tmp-nvme/vllm/lib/python3.12/site-packages/transformers/generation/utils.py:3426: in _sample
    outputs = self(**model_inputs, return_dict=True)
/tmp-nvme/vllm/lib/python3.12/site-packages/torch/nn/modules/module.py:1739: in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
/tmp-nvme/vllm/lib/python3.12/site-packages/torch/nn/modules/module.py:1750: in _call_impl
    return forward_call(*args, **kwargs)
../.cache/huggingface/modules/transformers_modules/fixie-ai/ultravox-v0_5-llama-3_2-1b/06c7f4eb509f60ce5f03563c0514756ebf357d39/ultravox_model.py:226: in forward
    audio_tower_output = self.audio_tower.forward(
../.cache/huggingface/modules/transformers_modules/fixie-ai/ultravox-v0_5-llama-3_2-1b/06c7f4eb509f60ce5f03563c0514756ebf357d39/ultravox_model.py:674: in forward
    inputs_embeds = nn.functional.gelu(self.conv1(input_features))
/tmp-nvme/vllm/lib/python3.12/site-packages/torch/nn/modules/module.py:1739: in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
/tmp-nvme/vllm/lib/python3.12/site-packages/torch/nn/modules/module.py:1750: in _call_impl
    return forward_call(*args, **kwargs)
/tmp-nvme/vllm/lib/python3.12/site-packages/torch/nn/modules/conv.py:375: in forward
    return self._conv_forward(input, self.weight, self.bias)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = Conv1d(128, 1280, kernel_size=(3,), stride=(1,), padding=(1,))
input = tensor([[[-0.5781, -0.5781, -0.5781,  ..., -0.5781, -0.5781, -0.5781],
         [-0.5781, -0.5781, -0.5781,  ..., -0.5...
         [-0.5781, -0.5781, -0.5781,  ..., -0.5781, -0.5781, -0.5781]]],
       device='cuda:0', dtype=torch.bfloat16)
weight = Parameter containing:
tensor([[[-1.9646e-04, -4.3297e-04,  1.1902e-03],
         [ 1.3062e-02,  1.4893e-02,  1.4404e-0...     [-2.1973e-03, -4.4861e-03, -1.1673e-03],
         [ 7.8583e-04, -4.2114e-03,  1.0834e-03]]], dtype=torch.bfloat16)
bias = Parameter containing:
tensor([ 0.0654,  0.0085,  0.0076,  ..., -0.0023,  0.0535,  0.0216],
       dtype=torch.bfloat16)

    def _conv_forward(self, input: Tensor, weight: Tensor, bias: Optional[Tensor]):
        if self.padding_mode != "zeros":
            return F.conv1d(
                F.pad(
                    input, self._reversed_padding_repeated_twice, mode=self.padding_mode
                ),
                weight,
                bias,
                self.stride,
                _single(0),
                self.dilation,
                self.groups,
            )
>       return F.conv1d(
            input, weight, bias, self.stride, self.padding, self.dilation, self.groups
        )
E       RuntimeError: Input type (CUDABFloat16Type) and weight type (CPUBFloat16Type) should be the same

[farzadab]

Thanks for reporting. This should be fixed now.

[DarkLight1337]

Thanks, I have opened #16381 to renable the test