Training with PPO

Reinforcement Learning from Human Feedback (RLHF)

Mina Parham

AI Engineer

Fine-Tuning with reinforcement learning

The initial LLM and reward model in the RLHF process.

Fine-Tuning with reinforcement learning

The full RLHF process.

Fine-Tuning a Language Model with PPO

A diagram of a query given to a LLM to generate a continuation based on it.

Fine-Tuning a Language Model with PPO

A diagram of a query given to a LLM, that completes the query: 'we're half way there, oh livin' on a prayer'.

Fine-Tuning a Language Model with PPO

A diagram of a query given to a LLM, that completes the query: 'we're half way there, oh livin' on a prayer', with another LLM evaluating the completion.

Fine-Tuning a Language Model with PPO

PPO: gradual adjustment for the model
Avoids overfitting to feedback

A robot and a snail representing slow improvement of the algorithm.

Implementing PPOTrainer with TRL

from trl import PPOConfig
config = PPOConfig(model_name="gpt2",learning_rate=1.4e-5)

from trl import AutoModelForCausalLMWithValueHead
model = AutoModelForCausalLMWithValueHead.from_pretrained(config.model_name)
tokenizer = AutoTokenizer.from_pretrained(config.model_name)

from trl import PPOTrainer
ppo_trainer = PPOTrainer(model=model,config=config,dataset=dataset,
                         tokenizer=tokenizer)

Starting the training loop

for epoch in tqdm(range(10), "epoch: "):


  for batch in tqdm(ppo_trainer.dataloader):

    # Get responses
    response_tensors = ppo_trainer.generate(batch["input_ids"])

    batch["response"] = [tokenizer.decode(r.squeeze()) for r in response_tensors]

    # Compute reward score
    texts = [q + r for q, r in zip(batch["query"], batch["response"])]

    rewards = reward_model(texts)

    stats = ppo_trainer.step(query_tensors, response_tensors, rewards)
    ppo_trainer.log_stats(stats, batch, rewards)

Let's practice!

Reinforcement Learning from Human Feedback (RLHF)