fine-tuning-with-trl

Teaches a model which answers people prefer — show it good examples, train a scorer on pairs of better-and-worse answers, then let the model practise against that scorer, alone or as the whole chain.

WHEN TO USE

SCOPE

USE FOR

Run supervised fine-tuning on prompt-completion pairs to teach a base model to follow instructions.
Train a reward model from chosen-versus-rejected preference pairs.
Align a model directly on preference data with DPO, skipping a separate reward model.
Run online reinforcement learning with a custom reward function using GRPO.
Run the full RLHF pipeline — SFT, reward model, then PPO — end to end.

NOT FOR

For basic fine-tuning without any reinforcement learning, use the HuggingFace Trainer instead.
For YAML-based training configuration, use Axolotl instead.
For fast LoRA training, use Unsloth instead.
For educational, minimal fine-tuning, use LitGPT instead.

This is a saved copy of the skill's SKILL.md instructions. It may differ from the upstream file or the version you install.

In this document (8)

TRL - Transformer Reinforcement Learning

Quick start

TRL provides post-training methods for aligning language models with human preferences.

Installation:

pip install trl transformers datasets peft accelerate

Supervised Fine-Tuning (instruction tuning):

from trl import SFTTrainer

trainer = SFTTrainer(
    model="Qwen/Qwen2.5-0.5B",
    train_dataset=dataset,  # Prompt-completion pairs
)
trainer.train()

DPO (align with preferences):

from trl import DPOTrainer, DPOConfig

config = DPOConfig(output_dir="model-dpo", beta=0.1)
trainer = DPOTrainer(
    model=model,
    args=config,
    train_dataset=preference_dataset,  # chosen/rejected pairs
    processing_class=tokenizer
)
trainer.train()

Common workflows

Workflow 1: Full RLHF pipeline (SFT → Reward Model → PPO)

Complete pipeline from base model to human-aligned model.

Copy this checklist:

RLHF Training:
- [ ] Step 1: Supervised fine-tuning (SFT)
- [ ] Step 2: Train reward model
- [ ] Step 3: PPO reinforcement learning
- [ ] Step 4: Evaluate aligned model

Step 1: Supervised fine-tuning

Train base model on instruction-following data:

from transformers import AutoModelForCausalLM, AutoTokenizer
from trl import SFTTrainer, SFTConfig
from datasets import load_dataset

# Load model
model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-0.5B")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-0.5B")

# Load instruction dataset
dataset = load_dataset("trl-lib/Capybara", split="train")

# Configure training
training_args = SFTConfig(
    output_dir="Qwen2.5-0.5B-SFT",
    per_device_train_batch_size=4,
    num_train_epochs=1,
    learning_rate=2e-5,
    logging_steps=10,
    save_strategy="epoch"
)

# Train
trainer = SFTTrainer(
    model=model,
    args=training_args,
    train_dataset=dataset,
    tokenizer=tokenizer
)
trainer.train()
trainer.save_model()

Step 2: Train reward model

Train model to predict human preferences:

from transformers import AutoModelForSequenceClassification
from trl import RewardTrainer, RewardConfig

# Load SFT model as base
model = AutoModelForSequenceClassification.from_pretrained(
    "Qwen2.5-0.5B-SFT",
    num_labels=1  # Single reward score
)
tokenizer = AutoTokenizer.from_pretrained("Qwen2.5-0.5B-SFT")

# Load preference data (chosen/rejected pairs)
dataset = load_dataset("trl-lib/ultrafeedback_binarized", split="train")

# Configure training
training_args = RewardConfig(
    output_dir="Qwen2.5-0.5B-Reward",
    per_device_train_batch_size=2,
    num_train_epochs=1,
    learning_rate=1e-5
)

# Train reward model
trainer = RewardTrainer(
    model=model,
    args=training_args,
    processing_class=tokenizer,
    train_dataset=dataset
)
trainer.train()
trainer.save_model()

Step 3: PPO reinforcement learning

Optimize policy using reward model:

python -m trl.scripts.ppo \
    --model_name_or_path Qwen2.5-0.5B-SFT \
    --reward_model_path Qwen2.5-0.5B-Reward \
    --dataset_name trl-internal-testing/descriptiveness-sentiment-trl-style \
    --output_dir Qwen2.5-0.5B-PPO \
    --learning_rate 3e-6 \
    --per_device_train_batch_size 64 \
    --total_episodes 10000

Step 4: Evaluate

from transformers import pipeline

# Load aligned model
generator = pipeline("text-generation", model="Qwen2.5-0.5B-PPO")

# Test
prompt = "Explain quantum computing to a 10-year-old"
output = generator(prompt, max_length=200)[0]["generated_text"]
print(output)

Workflow 2: Simple preference alignment with DPO

Align model with preferences without reward model.

Copy this checklist:

DPO Training:
- [ ] Step 1: Prepare preference dataset
- [ ] Step 2: Configure DPO
- [ ] Step 3: Train with DPOTrainer
- [ ] Step 4: Evaluate alignment

Step 1: Prepare preference dataset

Dataset format:

{
  "prompt": "What is the capital of France?",
  "chosen": "The capital of France is Paris.",
  "rejected": "I don't know."
}

Load dataset:

from datasets import load_dataset

dataset = load_dataset("trl-lib/ultrafeedback_binarized", split="train")
# Or load your own
# dataset = load_dataset("json", data_files="preferences.json")

Step 2: Configure DPO

from trl import DPOConfig

config = DPOConfig(
    output_dir="Qwen2.5-0.5B-DPO",
    per_device_train_batch_size=4,
    num_train_epochs=1,
    learning_rate=5e-7,
    beta=0.1,  # KL penalty strength
    max_prompt_length=512,
    max_length=1024,
    logging_steps=10
)

Step 3: Train with DPOTrainer

from transformers import AutoModelForCausalLM, AutoTokenizer
from trl import DPOTrainer

model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-0.5B-Instruct")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-0.5B-Instruct")

trainer = DPOTrainer(
    model=model,
    args=config,
    train_dataset=dataset,
    processing_class=tokenizer
)

trainer.train()
trainer.save_model()

CLI alternative:

trl dpo \
    --model_name_or_path Qwen/Qwen2.5-0.5B-Instruct \
    --dataset_name argilla/Capybara-Preferences \
    --output_dir Qwen2.5-0.5B-DPO \
    --per_device_train_batch_size 4 \
    --learning_rate 5e-7 \
    --beta 0.1

Workflow 3: Memory-efficient online RL with GRPO

Train with reinforcement learning using minimal memory.

Copy this checklist:

GRPO Training:
- [ ] Step 1: Define reward function
- [ ] Step 2: Configure GRPO
- [ ] Step 3: Train with GRPOTrainer

Step 1: Define reward function

def reward_function(completions, **kwargs):
    """
    Compute rewards for completions.

    Args:
        completions: List of generated texts

    Returns:
        List of reward scores (floats)
    """
    rewards = []
    for completion in completions:
        # Example: reward based on length and unique words
        score = len(completion.split())  # Favor longer responses
        score += len(set(completion.lower().split()))  # Reward unique words
        rewards.append(score)
    return rewards

Or use a reward model:

from transformers import pipeline

reward_model = pipeline("text-classification", model="reward-model-path")

def reward_from_model(completions, prompts, **kwargs):
    # Combine prompt + completion
    full_texts = [p + c for p, c in zip(prompts, completions)]
    # Get reward scores
    results = reward_model(full_texts)
    return [r["score"] for r in results]

Step 2: Configure GRPO

from trl import GRPOConfig

config = GRPOConfig(
    output_dir="Qwen2-GRPO",
    per_device_train_batch_size=4,
    num_train_epochs=1,
    learning_rate=1e-5,
    num_generations=4,  # Generate 4 completions per prompt
    max_new_tokens=128
)

Step 3: Train with GRPOTrainer

from datasets import load_dataset
from trl import GRPOTrainer

# Load prompt-only dataset
dataset = load_dataset("trl-lib/tldr", split="train")

trainer = GRPOTrainer(
    model="Qwen/Qwen2-0.5B-Instruct",
    reward_funcs=reward_function,  # Your reward function
    args=config,
    train_dataset=dataset
)

trainer.train()

CLI:

trl grpo \
    --model_name_or_path Qwen/Qwen2-0.5B-Instruct \
    --dataset_name trl-lib/tldr \
    --output_dir Qwen2-GRPO \
    --num_generations 4

When to use vs alternatives

Use TRL when:

Need to align model with human preferences
Have preference data (chosen/rejected pairs)
Want to use reinforcement learning (PPO, GRPO)
Need reward model training
Doing RLHF (full pipeline)

Method selection:

SFT: Have prompt-completion pairs, want basic instruction following
DPO: Have preferences, want simple alignment (no reward model needed)
PPO: Have reward model, need maximum control over RL
GRPO: Memory-constrained, want online RL
Reward Model: Building RLHF pipeline, need to score generations

Use alternatives instead:

HuggingFace Trainer: Basic fine-tuning without RL
Axolotl: YAML-based training configuration
LitGPT: Educational, minimal fine-tuning
Unsloth: Fast LoRA training

Common issues

Issue: OOM during DPO training

Reduce batch size and sequence length:

config = DPOConfig(
    per_device_train_batch_size=1,  # Reduce from 4
    max_length=512,  # Reduce from 1024
    gradient_accumulation_steps=8  # Maintain effective batch
)

Or use gradient checkpointing:

model.gradient_checkpointing_enable()

Issue: Poor alignment quality

Tune beta parameter:

# Higher beta = more conservative (stays closer to reference)
config = DPOConfig(beta=0.5)  # Default 0.1

# Lower beta = more aggressive alignment
config = DPOConfig(beta=0.01)

Issue: Reward model not learning

Check loss type and learning rate:

config = RewardConfig(
    learning_rate=1e-5,  # Try different LR
    num_train_epochs=3  # Train longer
)

Ensure preference dataset has clear winners:

# Verify dataset
print(dataset[0])
# Should have clear chosen > rejected

Issue: PPO training unstable

Adjust KL coefficient:

config = PPOConfig(
    kl_coef=0.1,  # Increase from 0.05
    cliprange=0.1  # Reduce from 0.2
)

Advanced topics

SFT training guide: See references/sft-training.md for dataset formats, chat templates, packing strategies, and multi-GPU training.

DPO variants: See references/dpo-variants.md for IPO, cDPO, RPO, and other DPO loss functions with recommended hyperparameters.

Reward modeling: See references/reward-modeling.md for outcome vs process rewards, Bradley-Terry loss, and reward model evaluation.

Online RL methods: See references/online-rl.md for PPO, GRPO, RLOO, and OnlineDPO with detailed configurations.

Hardware requirements

GPU: NVIDIA (CUDA required)
VRAM: Depends on model and method
- SFT 7B: 16GB (with LoRA)
- DPO 7B: 24GB (stores reference model)
- PPO 7B: 40GB (policy + reward model)
- GRPO 7B: 24GB (more memory efficient)
Multi-GPU: Supported via accelerate
Mixed precision: BF16 recommended (A100/H100)

Memory optimization:

Use LoRA/QLoRA for all methods
Enable gradient checkpointing
Use smaller batch sizes with gradient accumulation

Resources

Docs: https://huggingface.co/docs/trl/
GitHub: https://github.com/huggingface/trl
Papers:
- "Training language models to follow instructions with human feedback" (InstructGPT, 2022)
- "Direct Preference Optimization: Your Language Model is Secretly a Reward Model" (DPO, 2023)
- "Group Relative Policy Optimization" (GRPO, 2024)
Examples: https://github.com/huggingface/trl/tree/main/examples/scripts

This skill ships inside the AI Research Skills plugin.

Run this outside your AI tool, using Node.js 22.20.0 or newer. Skills CLI 1.7.2 skips skill selection; prompts cancel installation without an interactive terminal. If choosing Project when asked about installation scope, open your project's folder before running this.

Install checklist

Check selected tools if a picker appears; Claude Code may be unchecked under Additional agents. ↑/↓ moves, Space toggles selections, Enter confirms.
Without a tool picker, check Installing to: for your intended tool. Stop if it names a different tool.
If asked, choose Symlink (Recommended) for shared files or Copy to all agents for separate copies; review the summary before confirming installation.
Install the find-skills skill? defaults to Yes, adding another skill for all projects. Choose No if you don't want that extra installation.

Skills CLI 1.7.2 sends install data to skills.sh by default, including the source, skill names, target tools and file details; GitHub sources are reported only when GitHub confirms the repository is public, while other remote sources can also be reported. Confirmed-public GitHub sources also trigger a security-audit request. Before installing, set the environment variable DO_NOT_TRACK=1 or DISABLE_TELEMETRY=1 to disable both.

npx skills add orchestra-research/ai-research-skills --skill "fine-tuning-with-trl" --full-depth

After installing for Claude Code, open it in the installation project if you chose Project; Global makes the skill available across projects on this computer. Once Claude Code loads the skill, its description helps Claude decide when to use it.

If installation created a top-level skills folder that did not exist when your Claude Code session started, run /reload-skills in that session. Run it again after later changes in that folder.

What's inside

Files included with this skill.

View on GitHub

TRL - Transformer Reinforcement Learning

Quick start

Common workflows

Workflow 1: Full RLHF pipeline (SFT → Reward Model → PPO)

Workflow 2: Simple preference alignment with DPO

Workflow 3: Memory-efficient online RL with GRPO

When to use vs alternatives

Common issues

Advanced topics

Hardware requirements

Resources

Local tool required

What's inside

More skills

academic-plotting

ara-compiler

ara-research-manager

ara-rigor-reviewer

audiocraft-audio-generation

autogpt-agents