trl/pr_5607/en/dpo_trainer.md

DPO Trainer

Overview

TRL supports the Direct Preference Optimization (DPO) Trainer for training language models, as described in the paper Direct Preference Optimization: Your Language Model is Secretly a Reward Model by Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, Chelsea Finn.

The abstract from the paper is the following:

While large-scale unsupervised language models (LMs) learn broad world knowledge and some reasoning skills, achieving precise control of their behavior is difficult due to the completely unsupervised nature of their training. Existing methods for gaining such steerability collect human labels of the relative quality of model generations and fine-tune the unsupervised LM to align with these preferences, often with reinforcement learning from human feedback (RLHF). However, RLHF is a complex and often unstable procedure, first fitting a reward model that reflects the human preferences, and then fine-tuning the large unsupervised LM using reinforcement learning to maximize this estimated reward without drifting too far from the original model. In this paper we introduce a new parameterization of the reward model in RLHF that enables extraction of the corresponding optimal policy in closed form, allowing us to solve the standard RLHF problem with only a simple classification loss. The resulting algorithm, which we call Direct Preference Optimization (DPO), is stable, performant, and computationally lightweight, eliminating the need for sampling from the LM during fine-tuning or performing significant hyperparameter tuning. Our experiments show that DPO can fine-tune LMs to align with human preferences as well as or better than existing methods. Notably, fine-tuning with DPO exceeds PPO-based RLHF in ability to control sentiment of generations, and matches or improves response quality in summarization and single-turn dialogue while being substantially simpler to implement and train.

This post-training method was contributed by Kashif Rasul and later refactored by Quentin Gallouédec.

Quick start

This example demonstrates how to train a language model using the DPOTrainer from TRL. We train a Qwen 3 0.6B model on the UltraFeedback dataset.

from trl import DPOTrainer
from datasets import load_dataset

trainer = DPOTrainer(
    model="Qwen/Qwen3-0.6B",
    train_dataset=load_dataset("trl-lib/ultrafeedback_binarized", split="train"),
)
trainer.train()

Expected dataset type and format

DPO requires a preference dataset. The DPOTrainer is compatible with both standard and conversational dataset formats. When provided with a conversational dataset, the trainer will automatically apply the chat template to the dataset.

# Standard format
## Explicit prompt (recommended)
preference_example = {"prompt": "The sky is", "chosen": " blue.", "rejected": " green."}
# Implicit prompt
preference_example = {"chosen": "The sky is blue.", "rejected": "The sky is green."}

# Conversational format
## Explicit prompt (recommended)
preference_example = {"prompt": [{"role": "user", "content": "What color is the sky?"}],
                      "chosen": [{"role": "assistant", "content": "It is blue."}],
                      "rejected": [{"role": "assistant", "content": "It is green."}]}
## Implicit prompt
preference_example = {"chosen": [{"role": "user", "content": "What color is the sky?"},
                                 {"role": "assistant", "content": "It is blue."}],
                      "rejected": [{"role": "user", "content": "What color is the sky?"},
                                   {"role": "assistant", "content": "It is green."}]}

If your dataset is not in one of these formats, you can preprocess it to convert it into the expected format. Here is an example with the Vezora/Code-Preference-Pairs dataset:

from datasets import load_dataset

dataset = load_dataset("Vezora/Code-Preference-Pairs")

def preprocess_function(example):
    return {
        "prompt": [{"role": "user", "content": example["input"]}],
        "chosen": [{"role": "assistant", "content": example["accepted"]}],
        "rejected": [{"role": "assistant", "content": example["rejected"]}],
    }

dataset = dataset.map(preprocess_function, remove_columns=["instruction", "input", "accepted", "ID"])
print(next(iter(dataset["train"])))

{
    "prompt": [{"role": "user", "content": "Create a nested loop to print every combination of numbers [...]"}],
    "chosen": [{"role": "assistant", "content": "Here is an example of a nested loop in Python [...]"}],
    "rejected": [{"role": "assistant", "content": "Here is an example of a nested loop in Python [...]"}],
}

Looking deeper into the DPO method

Direct Preference Optimization (DPO) is a training method designed to align a language model with preference data. Instead of supervised input–output pairs, the model is trained on pairs of completions to the same prompt, where one completion is preferred over the other. The objective directly optimizes the model to widen the margin between the log-likelihoods of preferred and dispreferred completions, relative to a reference model, without requiring an explicit reward model. In practice, this is typically achieved by suppressing the likelihood of dispreferred completions rather than by increasing the likelihood of preferred ones.

This section breaks down how DPO works in practice, covering the key steps: preprocessing and loss computation.

Preprocessing and tokenization

During training, each example is expected to contain a prompt along with a preferred (chosen) and a dispreferred (rejected) completion. For more details on the expected formats, see Dataset formats. The DPOTrainer tokenizes each input using the model's tokenizer.

Computing the loss

The loss used in DPO is defined as follows: $${\mathcal{L}}_{\mathrm{D}\mathrm{P}\mathrm{O}}(\theta )=-{\mathbb{E}}_{(x,y^{+},y^{-})}\text{\hspace{0.17em}⁣}[\log \sigma \text{\hspace{0.17em}⁣}\left(\beta \right(\log \frac{{\pi }_{\theta }(y^{+}\text{\hspace{0.17em}⁣}\mid x)}{{\pi }_{\mathrm{r}\mathrm{e}\mathrm{f}}(y^{+}\text{\hspace{0.17em}⁣}\mid x)}-\log \frac{{\pi }_{\theta }(y^{-}\text{\hspace{0.17em}⁣}\mid x)}{{\pi }_{\mathrm{r}\mathrm{e}\mathrm{f}}(y^{-}\text{\hspace{0.17em}⁣}\mid x)}\left)\right)]\backslash mathcal\{L\}\_\{\backslash mathrm\{DPO\}\}(\backslash theta)\; =\; -\backslash mathbb\{E\}\_\{(x,y^\{+\},y^\{-\})\}\backslash !\backslash left[\backslash log\; \backslash sigma\backslash !\backslash left(\backslash beta\backslash Big(\backslash log\backslash frac\{\backslash pi\_\{\backslash theta\}(y^\{+\}\backslash !\backslash mid\; x)\}\{\backslash pi\_\{\backslash mathrm\{ref\}\}(y^\{+\}\backslash !\backslash mid\; x)\}-\backslash log\; \backslash frac\{\backslash pi\_\{\backslash theta\}(y^\{-\}\backslash !\backslash mid\; x)\}\{\backslash pi\_\{\backslash mathrm\{ref\}\}(y^\{-\}\backslash !\backslash mid\; x)\}\backslash Big)\backslash right)\backslash right]$$

where $xx$ is the prompt, $y^{+}y^+$ is the preferred completion and $y^{-}y^-$ is the dispreferred completion. ${\pi }_{\theta }\backslash pi\_\{\backslash theta\}$ is the policy model being trained, ${\pi }_{\mathrm{r}\mathrm{e}\mathrm{f}}\backslash pi\_\{\backslash mathrm\{ref\}\}$ is the reference model, $\sigma \backslash sigma$ is the sigmoid function, and $\beta >0\backslash beta\; >\; 0$ is a hyperparameter that controls the strength of the preference signal.

Loss Types

Several formulations of the objective have been proposed in the literature. Initially, the objective of DPO was defined as presented above.

loss_type=	Description
"sigmoid" (default)	Given the preference data, we can fit a binary classifier according to the Bradley-Terry model and in fact the DPO authors propose the sigmoid loss on the normalized likelihood via the logsigmoid to fit a logistic regression.
"hinge"	The RSO authors propose to use a hinge loss on the normalized likelihood from the SLiC paper. In this case, the beta is the reciprocal of the margin.
"ipo"	The IPO authors argue the logit transform can overfit and propose the identity transform to optimize preferences directly; TRL exposes this as loss_type="ipo".
"exo_pair"	The EXO authors propose reverse-KL preference optimization. label_smoothing must be strictly greater than 0.0; a recommended value is 1e-3 (see Eq. 16 for the simplified pairwise variant). The full method uses K>2 SFT completions and approaches PPO as K grows.
"nca_pair"	The NCA authors shows that NCA optimizes the absolute likelihood for each response rather than the relative likelihood.
"robust"	The Robust DPO authors propose an unbiased DPO loss under noisy preferences. Use label_smoothing in DPOConfig to model label-flip probability; valid values are in the range [0.0, 0.5).
"bco_pair"	The BCO authors train a binary classifier whose logit serves as a reward so that the classifier maps {prompt, chosen completion} pairs to 1 and {prompt, rejected completion} pairs to 0. For unpaired data, we recommend the dedicated experimental.bco.BCOTrainer.
"sppo_hard"	The SPPO authors claim that SPPO is capable of solving the Nash equilibrium iteratively by pushing the chosen rewards to be as large as 1/2 and the rejected rewards to be as small as -1/2 and can alleviate data sparsity issues. The implementation approximates this algorithm by employing hard label probabilities, assigning 1 to the winner and 0 to the loser.
"aot" or loss_type="aot_unpaired"	The AOT authors propose Distributional Preference Alignment via Optimal Transport. loss_type="aot" is for paired data; loss_type="aot_unpaired" is for unpaired data. Both enforce stochastic dominance via sorted quantiles; larger per-GPU batch sizes help.
"apo_zero" or loss_type="apo_down"	The APO method introduces an anchored objective. apo_zero boosts winners and downweights losers (useful when the model underperforms the winners). apo_down downweights both, with stronger pressure on losers (useful when the model already outperforms winners).
"discopop"	The DiscoPOP paper uses LLMs to discover more efficient offline preference optimization losses. In the paper the proposed DiscoPOP loss (which is a log-ratio modulated loss) outperformed other optimization losses on different tasks (IMDb positive text generation, Reddit TLDR summarization, and Alpaca Eval 2.0).
"sft"	SFT (Supervised Fine-Tuning) loss is the negative log likelihood loss, used to train the model to generate preferred responses.

Logged metrics

While training and evaluating we record the following reward metrics:

• global_step: The total number of optimizer steps taken so far.
• epoch: The current epoch number, based on dataset iteration.
• num_tokens: The total number of tokens processed so far.
• loss: The average cross-entropy loss computed over non-masked tokens in the current logging interval.
• entropy: The average entropy of the model's predicted token distribution over non-masked tokens.
• mean_token_accuracy: The proportion of non-masked tokens for which the model’s top-1 prediction matches the token from the chosen completion.
• learning_rate: The current learning rate, which may change dynamically if a scheduler is used.
• grad_norm: The L2 norm of the gradients, computed before gradient clipping.
• logits/chosen: The average logit values assigned by the model to the tokens in the chosen completion.
• logits/rejected: The average logit values assigned by the model to the tokens in the rejected completion.
• logps/chosen: The average log-probability assigned by the model to the tokens in the chosen completion.
• logps/rejected: The average log-probability assigned by the model to the tokens in the rejected completion.
• rewards/chosen: The average implicit reward computed for the chosen completion, computed as $\beta \log \frac{{\pi }_{\theta }(y^{+}\text{\hspace{0.17em}⁣}\mid x)}{{\pi }_{\mathrm{r}\mathrm{e}\mathrm{f}}(y^{+}\text{\hspace{0.17em}⁣}\mid x)}\backslash beta\; \backslash log\; \backslash frac\{\backslash pi\_\{\backslash theta\}(y^\{+\}\backslash !\backslash mid\; x)\}\{\backslash pi\_\{\backslash mathrm\{ref\}\}(y^\{+\}\backslash !\backslash mid\; x)\}$.
• rewards/rejected: The average implicit reward computed for the rejected completion, computed as $\beta \log \frac{{\pi }_{\theta }(y^{-}\text{\hspace{0.17em}⁣}\mid x)}{{\pi }_{\mathrm{r}\mathrm{e}\mathrm{f}}(y^{-}\text{\hspace{0.17em}⁣}\mid x)}\backslash beta\; \backslash log\; \backslash frac\{\backslash pi\_\{\backslash theta\}(y^\{-\}\backslash !\backslash mid\; x)\}\{\backslash pi\_\{\backslash mathrm\{ref\}\}(y^\{-\}\backslash !\backslash mid\; x)\}$.
• rewards/margins: The average implicit reward margin between the chosen and rejected completions.
• rewards/accuracies: The proportion of examples where the implicit reward for the chosen completion is higher than that for the rejected completion.

Customization

Compatibility and constraints

Some argument combinations are intentionally restricted in the current DPOTrainer implementation:

• use_weighting=True is not supported with loss_type="aot" or loss_type="aot_unpaired".
• With use_liger_kernel=True:
  • only a single loss_type is supported,
  • compute_metrics is not supported,
  • precompute_ref_log_probs=True is not supported.
• sync_ref_model=True is not supported when training with PEFT models that do not keep a standalone ref_model.
• sync_ref_model=True cannot be combined with precompute_ref_log_probs=True.
• precompute_ref_log_probs=True is not supported with IterableDataset (train or eval).

Multi-loss combinations

The DPO trainer supports combining multiple loss functions with different weights, enabling more sophisticated optimization strategies. This is particularly useful for implementing algorithms like MPO (Mixed Preference Optimization). MPO is a training approach that combines multiple optimization objectives, as described in the paper Enhancing the Reasoning Ability of Multimodal Large Language Models via Mixed Preference Optimization.

To combine multiple losses, specify the loss types and corresponding weights as lists:

# MPO: Combines DPO (sigmoid) for preference and BCO (bco_pair) for quality
training_args = DPOConfig(
    loss_type=["sigmoid", "bco_pair", "sft"],  # loss types to combine
    loss_weights=[0.8, 0.2, 1.0]  # corresponding weights, as used in the MPO paper
)

Model initialization

You can directly pass the kwargs of the from_pretrained() method to the DPOConfig. For example, if you want to load a model in a different precision, analogous to

model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-0.6B", dtype=torch.bfloat16)

you can do so by passing the model_init_kwargs={"dtype": torch.bfloat16} argument to the DPOConfig.

from trl import DPOConfig

training_args = DPOConfig(
    model_init_kwargs={"dtype": torch.bfloat16},
)

Note that all keyword arguments of from_pretrained() are supported.

Train adapters with PEFT

We support tight integration with 🤗 PEFT library, allowing any user to conveniently train adapters and share them on the Hub, rather than training the entire model.

from datasets import load_dataset
from trl import DPOTrainer
from peft import LoraConfig

dataset = load_dataset("trl-lib/ultrafeedback_binarized", split="train")

trainer = DPOTrainer(
    "Qwen/Qwen3-0.6B",
    train_dataset=dataset,
    peft_config=LoraConfig(),
)

trainer.train()

You can also continue training your PeftModel. For that, first load a PeftModel outside DPOTrainer and pass it directly to the trainer without the peft_config argument being passed.

from datasets import load_dataset
from trl import DPOTrainer
from peft import AutoPeftModelForCausalLM

model = AutoPeftModelForCausalLM.from_pretrained("trl-lib/Qwen3-4B-LoRA", is_trainable=True)
dataset = load_dataset("trl-lib/ultrafeedback_binarized", split="train")

trainer = DPOTrainer(
    model=model,
    train_dataset=dataset,
)

trainer.train()

When training adapters, you typically use a higher learning rate (≈1e‑5) than full fine-tuning since only new parameters are being learned.

DPOConfig(learning_rate=1e-5, ...)

Train with Liger Kernel

Liger Kernel is a collection of Triton kernels for LLM training that boosts multi-GPU throughput by 20%, cuts memory use by 60% (enabling up to 4× longer context), and works seamlessly with tools like FlashAttention, PyTorch FSDP, and DeepSpeed. For more information, see Liger Kernel Integration.

Rapid Experimentation for DPO

RapidFire AI is an open-source experimentation engine that sits on top of TRL and lets you launch multiple DPO configurations at once, even on a single GPU. Instead of trying configurations sequentially, RapidFire lets you see all their learning curves earlier, stop underperforming runs, and clone promising ones with new settings in flight without restarting. For more information, see RapidFire AI Integration.

Train with Unsloth

Unsloth is an open‑source framework for fine‑tuning and reinforcement learning that trains LLMs (like Llama, Mistral, Gemma, DeepSeek, and more) up to 2× faster with up to 70% less VRAM, while providing a streamlined, Hugging Face–compatible workflow for training, evaluation, and deployment. For more information, see Unsloth Integration.

Tool Calling with DPO

The DPOTrainer fully supports fine-tuning models with tool calling capabilities. In this case, each dataset example should include:

• The conversation messages (prompt, chosen and rejected), including any tool calls (tool_calls) and tool responses (tool role messages)
• The list of available tools in the tools column, typically provided as JSON schemas

For details on the expected dataset structure, see the Dataset Format — Tool Calling section.

Training Vision Language Models

DPOTrainer fully supports training Vision-Language Models (VLMs). To train a VLM, provide a dataset with either an image column (single image per sample) or an images column (list of images per sample). For more information on the expected dataset structure, see the Dataset Format — Vision Dataset section. An example of such a dataset is the RLAIF-V Dataset dataset.

from trl import DPOConfig, DPOTrainer
from datasets import load_dataset

trainer = DPOTrainer(
    model="Qwen/Qwen2.5-VL-3B-Instruct",
    args=DPOConfig(max_length=None),
    train_dataset=load_dataset("HuggingFaceH4/rlaif-v_formatted", split="train"),
)
trainer.train()

For VLMs, truncating may remove image tokens, leading to errors during training. To avoid this, set max_length=None in the DPOConfig. This allows the model to process the full sequence length without truncating image tokens.

DPOConfig(max_length=None, ...)

Only use max_length when you've verified that truncation won't remove image tokens for the entire dataset.

DPOTrainer[[trl.DPOTrainer]]

trl.DPOTrainer[[trl.DPOTrainer]]

Source

Trainer for Direct Preference Optimization (DPO) method. This algorithm was initially proposed in the paper Direct Preference Optimization: Your Language Model is Secretly a Reward Model. This class is a wrapper around the Trainer class and inherits all of its attributes and methods.

Example:

from trl import DPOTrainer
from datasets import load_dataset

dataset = load_dataset("trl-lib/ultrafeedback_binarized", split="train")

trainer = DPOTrainer(
    model="Qwen/Qwen2.5-0.5B-Instruct",
    train_dataset=dataset,
)
trainer.train()

traintrl.DPOTrainer.trainhttps://github.com/huggingface/trl/blob/vr_5607/transformers/trainer.py#L1323[{"name": "resume_from_checkpoint", "val": ": str | bool | None = None"}, {"name": "trial", "val": ": optuna.Trial | dict[str, Any] | None = None"}, {"name": "ignore_keys_for_eval", "val": ": list[str] | None = None"}]- resume_from_checkpoint (str or bool, optional) -- If a str, local path to a saved checkpoint as saved by a previous instance of Trainer. If a bool and equals True, load the last checkpoint in args.output_dir as saved by a previous instance of Trainer. If present, training will resume from the model/optimizer/scheduler states loaded here.

• trial (optuna.Trial or dict[str, Any], optional) -- The trial run or the hyperparameter dictionary for hyperparameter search.
• ignore_keys_for_eval (list[str], optional) -- A list of keys in the output of your model (if it is a dictionary) that should be ignored when gathering predictions for evaluation during the training.0~trainer_utils.TrainOutputObject containing the global step count, training loss, and metrics.

Main training entry point.

Parameters:

model (str or PreTrainedModel or PeftModel) : Model to be trained. Can be either: - A string, being the model id of a pretrained model hosted inside a model repo on huggingface.co, or a path to a directory containing model weights saved using save_pretrained, e.g., './my_model_directory/'. The model is loaded using .from_pretrained (where `` is derived from the model config) with the keyword arguments in args.model_init_kwargs. - A PreTrainedModel object. Only causal language models are supported. - A PeftModel object. Only causal language models are supported.

ref_model (PreTrainedModel, optional) : Reference model used to compute the reference log probabilities. - If provided, this model is used directly as the reference policy. - If None, the trainer will automatically use the initial policy corresponding to model, i.e. the model state before DPO training starts.

args (DPOConfig, optional) : Configuration for this trainer. If None, a default configuration is used.

data_collator (DataCollator, optional) : Function to use to form a batch from a list of elements of the processed train_dataset or eval_dataset. Will default to DataCollatorForPreference if the model is a language model and DataCollatorForVisionPreference if the model is a vision-language model. Custom collators must truncate sequences before padding; the trainer does not apply post-collation truncation.

train_dataset (Dataset or IterableDataset) : Dataset to use for training. This trainer supports both language modeling type and prompt-completion type. The format of the samples can be either: - Standard: Each sample contains plain text. - Conversational: Each sample contains structured messages (e.g., role and content).

eval_dataset (Dataset, IterableDataset or dict[str, Dataset | IterableDataset]) : Dataset to use for evaluation. It must meet the same requirements as train_dataset.

processing_class (PreTrainedTokenizerBase or ProcessorMixin, optional) : Processing class used to process the data. The padding side must be set to "left". If None, the processing class is loaded from the model's name with from_pretrained. A padding token, tokenizer.pad_token, must be set. If the processing class has not set a padding token, tokenizer.eos_token will be used as the default.

compute_metrics (Callable[[EvalPrediction], dict], optional) : The function that will be used to compute metrics at evaluation. Must take a EvalPrediction and return a dictionary string to metric values. When passing SFTConfig with batch_eval_metrics set to True, your compute_metrics function must take a boolean compute_result argument. This will be triggered after the last eval batch to signal that the function needs to calculate and return the global summary statistics rather than accumulating the batch-level statistics.

callbacks (list of TrainerCallback, optional) : List of callbacks to customize the training loop. Will add those to the list of default callbacks detailed in here. If you want to remove one of the default callbacks used, use the remove_callback method.

optimizers (tuple[torch.optim.Optimizer | None, torch.optim.lr_scheduler.LambdaLR | None], optional, defaults to (None, None)) : A tuple containing the optimizer and the scheduler to use. Will default to an instance of AdamW on your model and a scheduler given by get_linear_schedule_with_warmup controlled by args.

peft_config (PeftConfig, optional) : PEFT configuration used to wrap the model. If None, the model is not wrapped.

Returns:

~trainer_utils.TrainOutput

Object containing the global step count, training loss, and metrics.

save_model[[trl.DPOTrainer.save_model]]

Source

Will save the model, so you can reload it using from_pretrained().

Will only save from the main process.

push_to_hub[[trl.DPOTrainer.push_to_hub]]

Source

Upload self.model and self.processing_class to the 🤗 model hub on the repo self.args.hub_model_id.

Parameters:

commit_message (str, optional, defaults to "End of training") : Message to commit while pushing.

blocking (bool, optional, defaults to True) : Whether the function should return only when the git push has finished.

token (str, optional, defaults to None) : Token with write permission to overwrite Trainer's original args.

revision (str, optional) : The git revision to commit from. Defaults to the head of the "main" branch.

kwargs (dict[str, Any], optional) : Additional keyword arguments passed along to ~Trainer.create_model_card.

Returns:

The URL of the repository where the model was pushed if blocking=False, or a Future object tracking the progress of the commit if blocking=True.

DPOConfig[[trl.DPOConfig]]

trl.DPOConfig[[trl.DPOConfig]]

Source

Configuration class for the DPOTrainer.

This class includes only the parameters that are specific to DPO training. For a full list of training arguments, please refer to the TrainingArguments documentation. Note that default values in this class may differ from those in TrainingArguments.

Using HfArgumentParser we can turn this class into argparse arguments that can be specified on the command line.

These parameters have default values different from TrainingArguments:

• logging_steps: Defaults to 10 instead of 500.
• gradient_checkpointing: Defaults to True instead of False.
• bf16: Defaults to True if fp16 is not set, instead of False.
• learning_rate: Defaults to 1e-6 instead of 5e-5.