deps: pull in transformers latest version

[anhuong]

Description of the change

• set lower limit of transformers to 4.41.0 which uses TrainingArgumenteval_strategy instead of evaluation_strategy
• set upper limit of transformers to 5.0, thus pulling in new versions of transformers

Related issue number

Previously we had seen degraded performance in v4.41 of transformers as described in issue #201. After further evaluation described in the issue, the throughput degradation on llama3-8B was only 17%, only on 2 A100, and not at all on 4 A100, and that memory utilization was significantly improved by moving to 4.41.

How to verify the PR

Was the PR tested

• I have added >=1 unit test(s) for every new method I have added.
• I have ensured all unit tests pass

[anhuong]

Note that transformers starting with v4.42.0 unit tests fail when running inference with error KeyError: 'Cache only has 0 layers, attempted to access layer with index 0'. Tuning succeeded and only failed on test cases that ran Prompt Tuning and ran inference after failed, the LoRA and fine tuning tests succeeded.
See unit tests for list of dep versions. Full error:

            # Run inference on the text
>           output_inference = loaded_model.run(
                "### Text: @NortonSupport Thanks much.\n\n### Label:", max_new_tokens=50
            )

tests/test_sft_trainer.py:248: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
scripts/run_inference.py:246: in run
    peft_outputs = self.peft_model.generate(
.tox/py/lib/python3.10/site-packages/peft/peft_model.py:1493: in generate
    outputs = self.base_model.generate(**kwargs)
.tox/py/lib/python3.10/site-packages/torch/utils/_contextlib.py:115: in decorate_context
    return func(*args, **kwargs)
.tox/py/lib/python3.10/site-packages/transformers/generation/utils.py:1914: in generate
    result = self._sample(
.tox/py/lib/python3.10/site-packages/transformers/generation/utils.py:2648: in _sample
    model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
.tox/py/lib/python3.10/site-packages/peft/peft_model.py:1522: in prepare_inputs_for_generation
    if model_kwargs["past_key_values"][0][0].shape[-2] >= model_kwargs["input_ids"].shape[1]:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = DynamicCache(), layer_idx = 0

    def __getitem__(self, layer_idx: int) -> List[Tuple[torch.Tensor]]:
        """
        Support for backwards-compatible `past_key_value` indexing, e.g. `past_key_value[0][0].shape[2]` to get the
        sequence length.
        """
        if layer_idx < len(self):
            return (self.key_cache[layer_idx], self.value_cache[layer_idx])
        else:
>           raise KeyError(f"Cache only has {len(self)} layers, attempted to access layer with index {layer_idx}")
E           KeyError: 'Cache only has 0 layers, attempted to access layer with index 0'

.tox/py/lib/python3.10/site-packages/transformers/cache_utils.py:314: KeyError

Whereas transformers v4.41.2 unit tests succeeds.

Also found related issue from transformers repo which is closed but has new comments with people hitting the same issue but is talking about an older version of transformers v4.38

[anhuong]

closing in favor of #246