Retour sur plusieurs tours

import math

from datasets import Dataset
from transformers import AutoModelForCausalLM, AutoTokenizer
from comlrl.trainers.reinforce import MAGRPOConfig, MAGRPOTrainer

model_name = "Qwen/Qwen2.5-0.5B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
train_dataset = Dataset.from_dict({
    "prompt": [
        "Describe a city in the clouds:",
        "Invent a new holiday and explain it:",
        "Write a bedtime story for a dragon:",
        "Design a playground on Mars:",
    ]
})


def length_ratio_reward(completions1, completions2):
    rewards = []
    for first, second in zip(completions1, completions2):
        ratio = len(second) / max(len(first), 1)
        distance = 0.0 if 2 <= ratio <= 3 else min(abs(ratio - 2), abs(ratio - 3))
        rewards.append(math.exp(-distance))
    return rewards

def length_feedback(prompt, agent_completions, num_agents, **history):
    lengths = [len(text) for text in agent_completions]
    ratio = lengths[1] / max(lengths[0], 1)
    return [
        f"{prompt}\n\nYour last answer had {lengths[i]} characters. "
        f"The second answer was {ratio:.1f} times as long as the first; "
        "the goal is 2 to 3 times. Please write an improved answer."
        for i in range(num_agents)
    ]

config = MAGRPOConfig(
    num_turns=2,
    num_train_epochs=2,
    num_generations=4,
    max_new_tokens=64,
    logging_steps=1,
    early_termination_threshold=0.9,
)
trainer = MAGRPOTrainer(
    agents=[AutoModelForCausalLM.from_pretrained(model_name) for _ in range(2)],
    tokenizer=tokenizer,
    reward_func=length_ratio_reward,
    external_transition=length_feedback,
    train_dataset=train_dataset,
    args=config,
)
trainer.train()