[ORPO] Move ORPOTrainer to experimental

docs/source/_toctree.yml

@@ -70,8 +70,6 @@
       title: KTO
     - local: nash_md_trainer
       title: Nash-MD
-    - local: orpo_trainer
-      title: ORPO
     - local: ppo_trainer
       title: PPO
     - local: prm_trainer
@@ -117,8 +115,10 @@
     title: GRPO With Replay Buffer
   - local: gspo_token
     title: GSPO-token
-  - local: papo_trainer
-    title: PAPO
   - local: openenv
     title: OpenEnv Integration
+  - local: orpo_trainer
+    title: ORPO
+  - local: papo_trainer
+    title: PAPO
   title: Experimental

docs/source/orpo_trainer.md

@@ -34,7 +34,7 @@ Below is the script to train the model:
 ```python
 # train_orpo.py
 from datasets import load_dataset
-from trl import ORPOConfig, ORPOTrainer
+from trl.experimental.orpo import ORPOConfig, ORPOTrainer
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2-0.5B-Instruct")

examples/scripts/orpo.py

@@ -63,7 +63,8 @@
 from datasets import load_dataset
 from transformers import AutoModelForCausalLM, AutoTokenizer, HfArgumentParser
 
-from trl import ModelConfig, ORPOConfig, ORPOTrainer, ScriptArguments, get_peft_config
+from trl import ModelConfig, ScriptArguments, get_peft_config
+from trl.experimental.orpo import ORPOConfig, ORPOTrainer
 
 
 # Enable logging in a Hugging Face Space

tests/test_orpo_trainer.py

@@ -17,7 +17,7 @@
 from datasets import load_dataset
 from transformers import AutoModelForCausalLM, AutoModelForSeq2SeqLM, AutoTokenizer
 
-from trl import ORPOConfig, ORPOTrainer
+from trl.experimental.orpo import ORPOConfig, ORPOTrainer
 
 from .testing_utils import TrlTestCase, require_peft
 

trl/experimental/orpo/__init__.py

@@ -0,0 +1,19 @@
+# Copyright 2020-2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .orpo_config import ORPOConfig
+from .orpo_trainer import ORPOTrainer
+
+
+__all__ = ["ORPOConfig", "ORPOTrainer"]

trl/experimental/orpo/orpo_config.py

@@ -0,0 +1,169 @@
+# Copyright 2020-2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from dataclasses import dataclass, field
+from typing import Any
+
+from transformers import TrainingArguments
+
+
+@dataclass
+class ORPOConfig(TrainingArguments):
+    r"""
+    Configuration class for the [`ORPOTrainer`].
+
+    This class includes only the parameters that are specific to ORPO training. For a full list of training arguments,
+    please refer to the [`~transformers.TrainingArguments`] documentation. Note that default values in this class may
+    differ from those in [`~transformers.TrainingArguments`].
+
+    Using [`~transformers.HfArgumentParser`] we can turn this class into
+    [argparse](https://docs.python.org/3/library/argparse#module-argparse) arguments that can be specified on the
+    command line.
+
+    Parameters:
+        max_length (`int` or `None`, *optional*, defaults to `1024`):
+            Maximum length of the sequences (prompt + completion) in the batch. This argument is required if you want
+            to use the default data collator.
+        max_prompt_length (`int` or `None`, *optional*, defaults to `512`):
+            Maximum length of the prompt. This argument is required if you want to use the default data collator.
+        max_completion_length (`int`, *optional*):
+            Maximum length of the completion. This argument is required if you want to use the default data collator
+            and your model is an encoder-decoder.
+        beta (`float`, *optional*, defaults to `0.1`):
+            Parameter controlling the relative ratio loss weight in the ORPO loss. In the
+            [paper](https://huggingface.co/papers/2403.07691), it is denoted by λ. In the
+            [code](https://github.com/xfactlab/orpo), it is denoted by `alpha`.
+        disable_dropout (`bool`, *optional*, defaults to `True`):
+            Whether to disable dropout in the model.
+        label_pad_token_id (`int`, *optional*, defaults to `-100`):
+            Label pad token id. This argument is required if you want to use the default data collator.
+        padding_value (`int`, *optional*):
+            Padding value to use. If `None`, the padding value of the tokenizer is used.
+        truncation_mode (`str`, *optional*, defaults to `"keep_end"`):
+            Truncation mode to use when the prompt is too long. Possible values are `"keep_end"` or `"keep_start"`.
+            This argument is required if you want to use the default data collator.
+        generate_during_eval (`bool`, *optional*, defaults to `False`):
+            If `True`, generates and logs completions from the model to W&B or Comet during evaluation.
+        is_encoder_decoder (`bool`, *optional*):
+            When using the `model_init` argument (callable) to instantiate the model instead of the `model` argument,
+            you need to specify if the model returned by the callable is an encoder-decoder model.
+        model_init_kwargs (`dict[str, Any]`, *optional*):
+            Keyword arguments to pass to `AutoModelForCausalLM.from_pretrained` when instantiating the model from a
+            string.
+        dataset_num_proc (`int`, *optional*):
+            Number of processes to use for processing the dataset.
+    """
+
+    _VALID_DICT_FIELDS = TrainingArguments._VALID_DICT_FIELDS + ["model_init_kwargs"]
+
+    # Parameters whose default values are overridden from TrainingArguments
+    learning_rate: float = field(
+        default=1e-6,
+        metadata={"help": "The initial learning rate for AdamW."},
+    )
+    logging_steps: float = field(
+        default=10,
+        metadata={
+            "help": "Log every X updates steps. Should be an integer or a float in range `[0,1)`. If smaller than 1, "
+            "will be interpreted as ratio of total training steps."
+        },
+    )
+    gradient_checkpointing: bool = field(
+        default=True,
+        metadata={
+            "help": "If True, use gradient checkpointing to save memory at the expense of slower backward pass."
+        },
+    )
+    bf16: bool | None = field(
+        default=None,
+        metadata={
+            "help": "Whether to use bf16 (mixed) precision instead of 32-bit. Requires Ampere or higher NVIDIA "
+            "architecture or Intel XPU or using CPU (use_cpu) or Ascend NPU. If not set, it defaults to `True` if "
+            "`fp16` is not set."
+        },
+    )
+
+    max_length: int | None = field(
+        default=1024,
+        metadata={"help": "Maximum length of the sequences (prompt + completion) in the batch."},
+    )
+    max_prompt_length: int | None = field(
+        default=512,
+        metadata={
+            "help": "Maximum length of the prompt. This argument is required if you want to use the default data "
+            "collator and your model is an encoder-decoder."
+        },
+    )
+    max_completion_length: int | None = field(
+        default=None,
+        metadata={
+            "help": "Maximum length of the completion. This argument is required if you want to use the default data "
+            "collator and your model is an encoder-decoder."
+        },
+    )
+    beta: float = field(
+        default=0.1,
+        metadata={
+            "help": "Parameter controlling the relative ratio loss weight in the ORPO loss. In the paper, it is "
+            "denoted by λ."
+        },
+    )
+    disable_dropout: bool = field(
+        default=True,
+        metadata={"help": "Whether to disable dropout in the model."},
+    )
+    label_pad_token_id: int = field(
+        default=-100,
+        metadata={
+            "help": "Label pad token id. This argument is required if you want to use the default data collator."
+        },
+    )
+    padding_value: int | None = field(
+        default=None,
+        metadata={"help": "Padding value to use. If `None`, the padding value of the tokenizer is used."},
+    )
+    truncation_mode: str = field(
+        default="keep_end",
+        metadata={
+            "help": "Truncation mode to use when the prompt is too long.",
+            "choices": ["keep_end", "keep_start"],
+        },
+    )
+    generate_during_eval: bool = field(
+        default=False,
+        metadata={"help": "If `True`, generates and logs completions from the model to W&B during evaluation."},
+    )
+    is_encoder_decoder: bool | None = field(
+        default=None,
+        metadata={
+            "help": "When using the `model_init` argument (callable) to instantiate the model instead of the `model` "
+            "argument, you need to specify if the model returned by the callable is an encoder-decoder model."
+        },
+    )
+    model_init_kwargs: dict[str, Any] | None = field(
+        default=None,
+        metadata={
+            "help": "Keyword arguments to pass to `AutoModelForCausalLM.from_pretrained` when instantiating the model "
+            "from a string."
+        },
+    )
+    dataset_num_proc: int | None = field(
+        default=None,
+        metadata={"help": "Number of processes to use for processing the dataset."},
+    )
+
+    def __post_init__(self):
+        self.bf16 = not (self.fp16) if self.bf16 is None else self.bf16
+
+        super().__post_init__()