trl-training

tarafından huggingface

TRL (Transformers Reinforcement Learning) kullanarak transformatör dil modellerini eğitin ve ince ayar yapın. SFT, DPO, GRPO, KTO, RLOO ve Ödül Modeli eğitimini destekler…

npx skills add https://github.com/huggingface/trl --skill trl-training

TRL

Each method pairs a *Trainer class with a *Config dataclass. Configs extend transformers.TrainingArguments, so all of its arguments work in any trainer config.

TrainerDataset type
SFTTrainerlanguage modeling or prompt-completion
DPOTrainerpreference (chosen/rejected pairs)
GRPOTrainerprompt-only + reward function(s)
DistillationTrainerprompt-only + a teacher model (on-policy distillation)
KTOTrainerunpaired preference (per-sample bool label)
RewardTrainerpreference (chosen/rejected pairs); trains a scalar reward model, not a policy

Many more trainers (OnlineDPO, ORPO, CPO, GKD, …) live in trl.experimental with unstable APIs: https://huggingface.co/docs/trl/experimental_overview

from datasets import load_dataset
from trl import SFTConfig, SFTTrainer

trainer = SFTTrainer(
    model="Qwen/Qwen2.5-0.5B",  # model ID or a PreTrainedModel instance
    args=SFTConfig(output_dir="Qwen2.5-0.5B-SFT"),
    train_dataset=load_dataset("trl-lib/Capybara", split="train"),
)
trainer.train()

Pass model as a string and route loading kwargs through model_init_kwargs (e.g. {"dtype": "bfloat16", "attn_implementation": "kernels-community/flash-attn2"}) instead of calling from_pretrained yourself. The tokenizer/processor is inferred from the model; pass processing_class only when it differs. For LoRA, pass peft_config=LoraConfig(...).

Dataset formats

Conversational: {"messages": [{"role": ..., "content": ...}]} (language modeling) or {"prompt": [...], "completion": [...]}. The chat template is applied automatically — never apply it yourself. Extra columns are allowed; GRPO forwards them to reward functions. Reference: https://huggingface.co/docs/trl/dataset_formats

SFT: the fields that matter

SFTConfig(
    max_length=1024,        # truncation length; None disables truncation
    packing=True,           # pack sequences into max_length blocks: fewer pad tokens, higher throughput
    padding_free=True,      # flatten batch, no padding; requires FlashAttention; implied by packing
    use_liger_kernel=True,  # fused Liger kernels, reduces peak memory
    assistant_only_loss=True,  # loss only on assistant turns (conversational datasets)
)

GRPO: online RL

def reward_len(completions, **kwargs):
    return [-abs(20 - len(c[0]["content"])) for c in completions]

trainer = GRPOTrainer(
    model="Qwen/Qwen2.5-0.5B-Instruct",
    reward_funcs=reward_len,  # or a list; rewards are summed
    args=GRPOConfig(output_dir="Qwen2.5-0.5B-GRPO", max_completion_length=512),
    train_dataset=load_dataset("trl-lib/DeepMath-103K", split="train"),
)

Reward functions are called with keyword arguments prompts, completions, completion_ids, trainer_state, plus every extra dataset column — accept **kwargs for the ones you ignore. Return list[float], one reward per completion. With conversational data, completions is a list of message lists, not strings.

The generation batch is per_device_train_batch_size × num_processes × steps_per_generation (or set generation_batch_size directly) and must be divisible by num_generations (default 8). Generation is the usual bottleneck — enable vLLM with use_vllm=True: vllm_mode="colocate" shares the training GPUs (size with vllm_gpu_memory_utilization); vllm_mode="server" uses a separate trl vllm-serve --model <model_id>.

AsyncGRPOTrainer (trl.experimental.async_grpo) implements the same algorithm with generation decoupled from training: a background worker streams completions from a vLLM server while the training loop consumes them, so the two overlap instead of alternating.

CLI

Flags mirror the config fields: trl sft --model_name_or_path Qwen/Qwen2.5-0.5B --dataset_name trl-lib/Capybara. YAML via --config; distributed presets via --accelerate_config zero3 (Python scripts: accelerate launch train.py).

huggingface tarafından daha fazla skill

sync-models
huggingface
Sohbet arayüzünün model yapılandırmasını HuggingFace yönlendiricisiyle senkronize edin — yeni modeller için açıklamalar ekleyin, akıl yürütme yapabilenleri işaretleyin, 32B+ modeller için yapıtları etkinleştirin…
custom-blocks
huggingface
Use when the user has written (or wants to write) a `ModularPipelineBlocks` subclass in a local Python file and needs to package it into a Hub-uploadable…
self-review
huggingface
Use before opening a PR, or whenever asked to self-review a diffusers contribution. Applies the same rubric as the `@claude` CI (checks the diff against…
hf-cloud-sagemaker-production-defaults
huggingface
Otomatik ölçeklendirme, CloudWatch alarmları ve etiketleme varsayılan olarak etkin olan bir SageMaker uç noktası (gerçek zamanlı veya async) oluşturun. Bir… oluşturulacaksa bu beceriyi kullanın.
hf-cloud-serving-image-selection
huggingface
SageMaker model dağıtımı için doğru sunucu konteynerini seçin ve mevcut görüntü URI'sini bulun. Bir modeli dağıtmayı planlarken bu beceriyi kullanın.
Hugging Face Cli
huggingface
Execute Hugging Face Hub operations using the `hf` CLI. Use when the user needs to download models/datasets/spaces, upload files to Hub repositories, create repos, manage local cache, or run compute jobs on HF infrastructure. Covers authentication, file transfers, repository creation, cache operations, and cloud compute.
Hugging Face Datasets
huggingface
Hugging Face Hub'da veri kümeleri oluşturun ve yönetin. Depo başlatma, yapılandırma/sistem istemleri tanımlama, satır güncellemelerini akışla aktarma ve SQL tabanlı veri kümesi sorgulama/dönüştürme işlemlerini destekler. Kapsamlı veri kümesi iş akışları için HF MCP sunucusuyla birlikte çalışacak şekilde tasarlanmıştır.
Hugging Face Evaluation
huggingface
Hugging Face model kartlarında değerlendirme sonuçları ekleyin ve yönetin. README içeriğinden değerlendirme tablolarını çıkarma, Artificial Analysis API'den puanları içe aktarma ve vLLM/lighteval ile özel model değerlendirmeleri çalıştırmayı destekler. Model-index meta veri formatıyla çalışır.