PyTorch 2.0 implementation on Hugging Face

nlp
1

Is there a way to easily pass the torch.compile directly to Hugging Face’s pipeline? Was thinking of something like this.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM,pipeline

tokenizer = AutoTokenizer.from_pretrained(“distilgpt2”)
model = torch.compile(AutoModelForCausalLM.from_pretrained(“distilgpt2”))

generator = pipeline(task=“text-generation”, model=model, tokenizer=tokenizer)

But was met with this error:

The model ‘OptimizedModule’ is not supported for text-generation. Supported models are [‘BartForCausalLM’, ‘BertLMHeadModel’, ‘BertGenerationDecoder’, ‘BigBirdForCausalLM’, ‘BigBirdPegasusForCausalLM’, ‘BlenderbotForCausalLM’, ‘BlenderbotSmallForCausalLM’, ‘BloomForCausalLM’, ‘CamembertForCausalLM’, ‘CodeGenForCausalLM’, ‘CTRLLMHeadModel’, ‘Data2VecTextForCausalLM’, ‘ElectraForCausalLM’, ‘ErnieForCausalLM’, ‘GPT2LMHeadModel’, ‘GPTNeoForCausalLM’, ‘GPTNeoXForCausalLM’, ‘GPTNeoXJapaneseForCausalLM’, ‘GPTJForCausalLM’, ‘MarianForCausalLM’, ‘MBartForCausalLM’, ‘MegatronBertForCausalLM’, ‘MvpForCausalLM’, ‘OpenAIGPTLMHeadModel’, ‘OPTForCausalLM’, ‘PegasusForCausalLM’, ‘PLBartForCausalLM’, ‘ProphetNetForCausalLM’, ‘QDQBertLMHeadModel’, ‘ReformerModelWithLMHead’, ‘RemBertForCausalLM’, ‘RobertaForCausalLM’, ‘RoCBertForCausalLM’, ‘RoFormerForCausalLM’, ‘Speech2Text2ForCausalLM’, ‘TransfoXLLMHeadModel’, ‘TrOCRForCausalLM’, ‘XGLMForCausalLM’, ‘XLMWithLMHeadModel’, ‘XLMProphetNetForCausalLM’, ‘XLMRobertaForCausalLM’, ‘XLMRobertaXLForCausalLM’, ‘XLNetLMHeadModel’].

I could pass the torch.compile function in Hugging Face’s codebase, but that seems like a lot of work for maintenance until Hugging Face implements PyTorch 2.0. Your suggestion is appreciated!

2

HuggingFace has an example of how this would work here Getting started with Pytorch 2.0 and Hugging Face Transformers