npx skills add ...
npx skills add firecrawl/ai-research-skills --skill fine-tuning-with-trl
npx skills add firecrawl/ai-research-skills --skill fine-tuning-with-trl
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.
TRL provides post-training methods for aligning language models with human preferences.
Installation:
Supervised Fine-Tuning (instruction tuning):
DPO (align with preferences):
Complete pipeline from base model to human-aligned model.
Copy this checklist:
Step 1: Supervised fine-tuning
Train base model on instruction-following data:
Step 2: Train reward model
Train model to predict human preferences:
Step 3: PPO reinforcement learning
Optimize policy using reward model:
Step 4: Evaluate
Align model with preferences without reward model.
Copy this checklist:
Step 1: Prepare preference dataset
Dataset format:
Load dataset:
Step 2: Configure DPO
Step 3: Train with DPOTrainer
CLI alternative:
Train with reinforcement learning using minimal memory.
Copy this checklist:
Step 1: Define reward function
Or use a reward model:
Step 2: Configure GRPO
Step 3: Train with GRPOTrainer
CLI:
Use TRL when:
Method selection:
Use alternatives instead:
Issue: OOM during DPO training
Reduce batch size and sequence length:
Or use gradient checkpointing:
Issue: Poor alignment quality
Tune beta parameter:
Issue: Reward model not learning
Check loss type and learning rate:
Ensure preference dataset has clear winners:
Issue: PPO training unstable
Adjust KL coefficient:
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.
accelerateMemory optimization:
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()RLHF Training:
- [ ] Step 1: Supervised fine-tuning (SFT)
- [ ] Step 2: Train reward model
- [ ] Step 3: PPO reinforcement learning
- [ ] Step 4: Evaluate aligned modelfrom 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()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()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 10000from 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)DPO Training:
- [ ] Step 1: Prepare preference dataset
- [ ] Step 2: Configure DPO
- [ ] Step 3: Train with DPOTrainer
- [ ] Step 4: Evaluate alignment{
"prompt": "What is the capital of France?",
"chosen": "The capital of France is Paris.",
"rejected": "I don't know."
}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")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
)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()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 0.1GRPO Training:
- [ ] Step 1: Define reward function
- [ ] Step 2: Configure GRPO
- [ ] Step 3: Train with GRPOTrainerdef 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 rewardsfrom 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]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
)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()trl grpo \
--model_name_or_path Qwen/Qwen2-0.5B-Instruct \
--dataset_name trl-lib/tldr \
--output_dir Qwen2-GRPO \
--num_generations 4config = DPOConfig(
per_device_train_batch_size=1, # Reduce from 4
max_length=512, # Reduce from 1024
gradient_accumulation_steps=8 # Maintain effective batch
)model.gradient_checkpointing_enable()# 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)config = RewardConfig(
learning_rate=1e-5, # Try different LR
num_train_epochs=3 # Train longer
)# Verify dataset
print(dataset[0])
# Should have clear chosen > rejectedconfig = PPOConfig(
kl_coef=0.1, # Increase from 0.05
cliprange=0.1 # Reduce from 0.2
)