feat: enhance model checks for transformers pipelines

[davidberenstein1957]

Description

• Added functions to identify transformers pipelines for causal language models, sequence-to-sequence models, and speech recognition.
• Updated various quantization and compilation algorithms to utilize the new model checks, allowing for better integration with transformers pipelines.
• Introduced new test fixtures for transformers pipelines in the test suite to ensure comprehensive coverage.

from transformers import pipeline
from pruna import smash, PrunaModel, SmashConfig

for algo in [("compiler", "torch_compile"), ("quantizer", "hqq")]:
    model_name = "HuggingFaceTB/SmolLM2-360M-Instruct"
    pipe = pipeline(
        task="text-generation",
        model=model_name,
    )
    print(pipe("Hello, how are you?"))
    smash_config = SmashConfig(device="cuda")
    smash_config[algo[0]] = algo[1]
    smashed_pipe = smash(pipe, smash_config=smash_config)
    smashed_pipe.save_pretrained("test_model")
    smashed_pipe = PrunaModel.from_pretrained("test_model")
    print(smashed_pipe("Hello, how are you?", max_new_tokens=100))

Related Issue

Fixes #276

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

How Has This Been Tested?

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

Additional Notes

from transformers import pipeline

from pruna import SmashConfig, smash

pipe = pipeline("text-generation", model="HuggingFaceTB/SmolLM2-360M-Instruct", device="cpu")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)
config = SmashConfig(device="cpu")
config["quantizer"] = "torchao"
smashed_pipe = smash(pipe, config)
print(smashed_pipe(messages, max_new_tokens=100))

[davidberenstein1957]

@begumcig do we need to do anything special with the inference handler here? Not as far as I could see but I would love to hear your thoughts :)

[johannaSommer]

Love it!!! Just a few questions 🙌

[llcnt]

So nice to support LM pipelines :)
I only have one question regarding save/load functionality: did you re-ran the gpu tests? Or try to save and re-load a pipeline smashed with hqq ? (I wonder if we need to add the save_pipeline_info function inside hqq custom saved function)

[johannaSommer]

Wonderful, looks good from my side! 🙌

[davidberenstein1957]

@llcnt you were right. I fixed it now.

[llcnt]

Thx for the hqq fix :) LGTM!