Full stack library to fine-tune and align large language models.
The trl library is a full stack tool to fine-tune and align transformer language and diffusion models using methods such as Supervised Fine-tuning step (SFT), Reward Modeling (RM) and the Proximal Policy Optimization (PPO) as well as Direct Preference Optimization (DPO).
The library is built on top of the transformers library and thus allows to use any model architecture available there.
Efficient and scalable:accelerateis the backbone oftrlwhich allows to scale model training from a single GPU to a large scale multi-node cluster with methods such as DDP and DeepSpeed.PEFTis fully integrated and allows to train even the largest models on modest hardware with quantisation and methods such as LoRA or QLoRA.unslothis also integrated and allows to significantly speed up training with dedicated kernels.
CLI: With the CLI you can fine-tune and chat with LLMs without writing any code using a single command and a flexible config system.Trainers: The Trainer classes are an abstraction to apply many fine-tuning methods with ease such as theSFTTrainer,DPOTrainer,RewardTrainer,PPOTrainer,CPOTrainer, andORPOTrainer.AutoModels: TheAutoModelForCausalLMWithValueHead&AutoModelForSeq2SeqLMWithValueHeadclasses add an additional value head to the model which allows to train them with RL algorithms such as PPO.Examples: Train GPT2 to generate positive movie reviews with a BERT sentiment classifier, full RLHF using adapters only, train GPT-j to be less toxic, StackLlama example, etc. following the examples.
Install the library with pip:
pip install trlIf you want to use the latest features before an official release you can install from source:
pip install git+https://github.com/huggingface/trl.gitIf you want to use the examples you can clone the repository with the following command:
git clone https://github.com/huggingface/trl.gitYou can use TRL Command Line Interface (CLI) to quickly get started with Supervised Fine-tuning (SFT), Direct Preference Optimization (DPO) and test your aligned model with the chat CLI:
SFT:
trl sft --model_name_or_path facebook/opt-125m --dataset_name imdb --output_dir opt-sft-imdbDPO:
trl dpo --model_name_or_path facebook/opt-125m --dataset_name trl-internal-testing/hh-rlhf-helpful-base-trl-style --output_dir opt-sft-hh-rlhf Chat:
trl chat --model_name_or_path Qwen/Qwen1.5-0.5B-ChatRead more about CLI in the relevant documentation section or use --help for more details.
For more flexibility and control over the training, you can use the dedicated trainer classes to fine-tune the model in Python.
This is a basic example of how to use the SFTTrainer from the library. The SFTTrainer is a light wrapper around the transformers Trainer to easily fine-tune language models or adapters on a custom dataset.
# importsfromdatasetsimportload_datasetfromtrlimportSFTTrainer# get datasetdataset=load_dataset("imdb", split="train")
# get trainertrainer=SFTTrainer(
"facebook/opt-350m",
train_dataset=dataset,
dataset_text_field="text",
max_seq_length=512,
)
# traintrainer.train()This is a basic example of how to use the RewardTrainer from the library. The RewardTrainer is a wrapper around the transformers Trainer to easily fine-tune reward models or adapters on a custom preference dataset.
# importsfromtransformersimportAutoModelForSequenceClassification, AutoTokenizerfromtrlimportRewardTrainer# load model and dataset - dataset needs to be in a specific formatmodel=AutoModelForSequenceClassification.from_pretrained("gpt2", num_labels=1)
tokenizer=AutoTokenizer.from_pretrained("gpt2")
...
# load trainertrainer=RewardTrainer(
model=model,
tokenizer=tokenizer,
train_dataset=dataset,
)
# traintrainer.train()This is a basic example of how to use the PPOTrainer from the library. Based on a query the language model creates a response which is then evaluated. The evaluation could be a human in the loop or another model's output.
# importsimporttorchfromtransformersimportAutoTokenizerfromtrlimportPPOTrainer, PPOConfig, AutoModelForCausalLMWithValueHead, create_reference_modelfromtrl.coreimportrespond_to_batch# get modelsmodel=AutoModelForCausalLMWithValueHead.from_pretrained('gpt2')
model_ref=create_reference_model(model)
tokenizer=AutoTokenizer.from_pretrained('gpt2')
tokenizer.pad_token=tokenizer.eos_token# initialize trainerppo_config=PPOConfig(batch_size=1, mini_batch_size=1)
# encode a queryquery_txt="This morning I went to the "query_tensor=tokenizer.encode(query_txt, return_tensors="pt")
# get model responseresponse_tensor=respond_to_batch(model, query_tensor)
# create a ppo trainerppo_trainer=PPOTrainer(ppo_config, model, model_ref, tokenizer)
# define a reward for response# (this could be any reward such as human feedback or output from another model)reward= [torch.tensor(1.0)]
# train model for one step with ppotrain_stats=ppo_trainer.step([query_tensor[0]], [response_tensor[0]], reward)DPOTrainer is a trainer that uses Direct Preference Optimization algorithm. This is a basic example of how to use the DPOTrainer from the library. The DPOTrainer is a wrapper around the transformers Trainer to easily fine-tune reward models or adapters on a custom preference dataset.
# importsfromtransformersimportAutoModelForCausalLM, AutoTokenizerfromtrlimportDPOTrainer# load model and dataset - dataset needs to be in a specific formatmodel=AutoModelForCausalLM.from_pretrained("gpt2")
tokenizer=AutoTokenizer.from_pretrained("gpt2")
...
# load trainertrainer=DPOTrainer(
model=model,
tokenizer=tokenizer,
train_dataset=dataset,
)
# traintrainer.train()If you want to contribute to trl or customizing it to your needs make sure to read the contribution guide and make sure you make a dev install:
git clone https://github.com/huggingface/trl.git
cd trl/
make devThe PPO implementation largely follows the structure introduced in the paper "Fine-Tuning Language Models from Human Preferences" by D. Ziegler et al. [paper, code].
DPO is based on the original implementation of "Direct Preference Optimization: Your Language Model is Secretly a Reward Model" by E. Mitchell et al. [paper, code]
@misc{vonwerra2022trl,
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang},
title = {TRL: Transformer Reinforcement Learning},
year = {2020},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/huggingface/trl}}
}