奖励建模
TRL 支持自定义奖励建模,任何人都可以在其数据集和模型上执行奖励建模。
请查看 examples/scripts/reward_modeling.py
中的完整灵活示例。
预期数据集格式
由于模型将针对示例对进行训练以预测哪个示例更受欢迎,因此 RewardTrainer 对数据集的格式有非常具体的要求。我们提供以下来自 Anthropic/hh-rlhf
数据集的示例
因此,如果您使用默认的 RewardDataCollatorWithPadding
数据整理器,则最终数据集对象至少应包含两个 4 个条目。条目应命名为
input_ids_chosen
attention_mask_chosen
input_ids_rejected
attention_mask_rejected
使用 RewardTrainer
准备完数据集后,您可以像使用 🤗 Transformers 中的 Trainer
类一样使用 RewardTrainer。您应该将一个 AutoModelForSequenceClassification
模型传递给 RewardTrainer,以及一个 RewardConfig,它配置训练的超参数。
利用 🤗 PEFT 训练奖励模型
只需在 RewardTrainer 的关键字参数中传递一个 peft_config
,训练器应该会自动处理将模型转换为 PEFT 模型!
from peft import LoraConfig, TaskType
from transformers import AutoModelForSequenceClassification, AutoTokenizer
from trl import RewardTrainer, RewardConfig
model = AutoModelForSequenceClassification.from_pretrained("gpt2")
peft_config = LoraConfig(
task_type=TaskType.SEQ_CLS,
inference_mode=False,
r=8,
lora_alpha=32,
lora_dropout=0.1,
)
...
trainer = RewardTrainer(
model=model,
args=training_args,
tokenizer=tokenizer,
train_dataset=dataset,
peft_config=peft_config,
)
trainer.train()
向损失函数添加边际值
如同 Llama 2 论文 中所述,可以通过向数据集添加 margin
列来向损失函数添加边际值。奖励收集器会自动传递它,并相应地计算损失。
def add_margin(row):
# Assume you have a score_chosen and score_rejected columns that you want to use to compute the margin
return {'margin': row['score_chosen'] - row['score_rejected']}
dataset = dataset.map(add_margin)
奖励中心化
在许多场景中,最好确保奖励模型的输出均值为零。这通常通过首先计算模型的平均得分,然后减去它来实现。
[Eisenstein 等人,2023] 提出了一个旨在直接学习中心化奖励模型的辅助损失函数。该辅助损失函数最小化奖励的平方和,鼓励模型自然地产生均值为零的输出
该辅助损失函数与主损失函数结合,由 [RewardConfig]
中的参数 center_rewards_coefficient
加权。默认情况下,此功能处于禁用状态 (center_rewards_coefficient = None
)。
reward_config = RewardConfig(
center_rewards_coefficient=0.01,
...
)
有关参考结果,请参阅 PR #1932。
RewardTrainer
类 trl.RewardTrainer
< 源代码 >( model: Union = None args: Optional = None data_collator: Optional = None train_dataset: Optional = None eval_dataset: Union = None tokenizer: Optional = None model_init: Optional = None compute_metrics: Optional = None callbacks: Optional = None optimizers: Tuple = (None, None) preprocess_logits_for_metrics: Optional = None max_length: Optional = None peft_config: Optional = None )
RewardTrainer 可用于训练自定义奖励模型。它是 transformers.Trainer
类的子类,并继承其所有属性和方法。建议使用 AutoModelForSequenceClassification
作为奖励模型。奖励模型应在配对示例的数据集上进行训练,其中每个示例都是两个序列的元组。奖励模型应该被训练来预测配对中哪个示例与手头的任务更相关。
奖励训练器期望数据集具有非常特定的格式。如果未使用默认的 RewardDataCollatorWithPadding
数据收集器,则数据集应至少包含两个 4 个条目。条目应命名为
input_ids_chosen
attention_mask_chosen
input_ids_rejected
attention_mask_rejected
可选地,您还可以将 margin
条目传递给数据集。此条目应包含用于调节奖励模型损失的边际值,如 https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/ 中所述。如果您不传递边际值,则不会使用任何边际值。
可视化奖励模型 logits 预测
RewardConfig
类 trl.RewardConfig
< 源代码 >( output_dir: str overwrite_output_dir: bool = False do_train: bool = False do_eval: bool = False do_predict: bool = False eval_strategy: Union = 'no' prediction_loss_only: bool = False per_device_train_batch_size: int = 8 per_device_eval_batch_size: int = 8 per_gpu_train_batch_size: Optional = None per_gpu_eval_batch_size: Optional = None gradient_accumulation_steps: int = 1 eval_accumulation_steps: Optional = None eval_delay: Optional = 0 torch_empty_cache_steps: Optional = None learning_rate: float = 5e-05 weight_decay: float = 0.0 adam_beta1: float = 0.9 adam_beta2: float = 0.999 adam_epsilon: float = 1e-08 max_grad_norm: float = 1.0 num_train_epochs: float = 3.0 max_steps: int = -1 lr_scheduler_type: Union = 'linear' lr_scheduler_kwargs: Union = <factory> warmup_ratio: float = 0.0 warmup_steps: int = 0 log_level: Optional = 'passive' log_level_replica: Optional = 'warning' log_on_each_node: bool = True logging_dir: Optional = None logging_strategy: Union = 'steps' logging_first_step: bool = False logging_steps: float = 500 logging_nan_inf_filter: bool = True save_strategy: Union = 'steps' save_steps: float = 500 save_total_limit: Optional = None save_safetensors: Optional = True save_on_each_node: bool = False save_only_model: bool = False restore_callback_states_from_checkpoint: bool = False no_cuda: bool = False use_cpu: bool = False use_mps_device: bool = False seed: int = 42 data_seed: Optional = None jit_mode_eval: bool = False use_ipex: bool = False bf16: bool = False fp16: bool = False fp16_opt_level: str = 'O1' half_precision_backend: str = 'auto' bf16_full_eval: bool = False fp16_full_eval: bool = False tf32: Optional = None local_rank: int = -1 ddp_backend: Optional = None tpu_num_cores: Optional = None tpu_metrics_debug: bool = False debug: Union = '' dataloader_drop_last: bool = False eval_steps: Optional = None dataloader_num_workers: int = 0 dataloader_prefetch_factor: Optional = None past_index: int = -1 run_name: Optional = None disable_tqdm: Optional = None remove_unused_columns: Optional = True label_names: Optional = None load_best_model_at_end: Optional = False metric_for_best_model: Optional = None greater_is_better: Optional = None ignore_data_skip: bool = False fsdp: Union = '' fsdp_min_num_params: int = 0 fsdp_config: Union = None fsdp_transformer_layer_cls_to_wrap: Optional = None accelerator_config: Union = None deepspeed: Union = None label_smoothing_factor: float = 0.0 optim: Union = 'adamw_torch' optim_args: Optional = None adafactor: bool = False group_by_length: bool = False length_column_name: Optional = 'length' report_to: Union = None ddp_find_unused_parameters: Optional = None ddp_bucket_cap_mb: Optional = None ddp_broadcast_buffers: Optional = None dataloader_pin_memory: bool = True dataloader_persistent_workers: bool = False skip_memory_metrics: bool = True use_legacy_prediction_loop: bool = False push_to_hub: bool = False resume_from_checkpoint: Optional = None hub_model_id: Optional = None hub_strategy: Union = 'every_save' hub_token: Optional = None hub_private_repo: bool = False hub_always_push: bool = False gradient_checkpointing: bool = False gradient_checkpointing_kwargs: Union = None include_inputs_for_metrics: bool = False include_for_metrics: List = <factory> eval_do_concat_batches: bool = True fp16_backend: str = 'auto' evaluation_strategy: Union = None push_to_hub_model_id: Optional = None push_to_hub_organization: Optional = None push_to_hub_token: Optional = None mp_parameters: str = '' auto_find_batch_size: bool = False full_determinism: bool = False torchdynamo: Optional = None ray_scope: Optional = 'last' ddp_timeout: Optional = 1800 torch_compile: bool = False torch_compile_backend: Optional = None torch_compile_mode: Optional = None dispatch_batches: Optional = None split_batches: Optional = None include_tokens_per_second: Optional = False include_num_input_tokens_seen: Optional = False neftune_noise_alpha: Optional = None optim_target_modules: Union = None batch_eval_metrics: bool = False eval_on_start: bool = False use_liger_kernel: Optional = False eval_use_gather_object: Optional = False max_length: Optional = None dataset_num_proc: Optional = None center_rewards_coefficient: Optional = None )
参数
- max_length (
Optional[int]
, 可选, 默认为None
) — 批次中序列(提示 + 补全)的最大长度。如果您想使用默认的数据整理器,则需要此参数。 - dataset_num_proc (
int
, 可选, 默认为None
) — 用于处理数据集的进程数。 - center_rewards_coefficient (
float
, 可选, 默认为None
) — 鼓励奖励模型输出均值为零的奖励的系数(由 https://huggingface.co/papers/2312.09244,公式 2 提议)。推荐值:0.01
。
RewardTrainer 的配置类。
使用 HfArgumentParser,我们可以将此类转换为 argparse 参数,这些参数可以在命令行上指定。