train_memory_agent.py

# /// script
# requires-python = ">=3.10"
# dependencies = [
#     "trl>=0.12.0",
#     "peft>=0.7.0",
#     "transformers>=4.36.0",
#     "accelerate>=0.24.0",
#     "trackio",
#     "datasets",
# ]
# ///

import os
import huggingface_hub
import trackio
from datasets import load_dataset
from peft import LoraConfig
from trl import SFTTrainer, SFTConfig

# Explicit login with token from secrets
token = os.environ.get("HF_TOKEN")
if token:
    huggingface_hub.login(token=token)
    print("Logged in to HF Hub")
else:
    print("WARNING: No HF_TOKEN found!")

print("Loading dataset...")
dataset = load_dataset("erik1988/way2agi-memory-agent-sft", data_files="memory-agent-sft-v3-merged.jsonl", split="train")
print(f"Dataset loaded: {len(dataset)} examples")

dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
train_dataset = dataset_split["train"]
eval_dataset = dataset_split["test"]
print(f"Train: {len(train_dataset)}, Eval: {len(eval_dataset)}")

peft_config = LoraConfig(
    r=32,
    lora_alpha=64,
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM",
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"],
)

config = SFTConfig(
    output_dir="elias-memory-agent-v1",
    push_to_hub=True,
    hub_model_id="erik1988/elias-memory-agent-v1",
    hub_strategy="every_save",
    hub_token=token,
    max_length=None,
    num_train_epochs=5,
    per_device_train_batch_size=2,
    gradient_accumulation_steps=4,
    learning_rate=2e-4,
    logging_steps=5,
    save_strategy="steps",
    save_steps=50,
    save_total_limit=3,
    eval_strategy="steps",
    eval_steps=50,
    warmup_ratio=0.1,
    lr_scheduler_type="cosine",
    gradient_checkpointing=True,
    report_to="trackio",
    project="way2agi-memory-agent",
    run_name="memory-agent-sft-v3-qwen1.5b",
)

print("Initializing trainer...")
trainer = SFTTrainer(
    model="Qwen/Qwen2.5-1.5B-Instruct",
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    args=config,
    peft_config=peft_config,
)

print("Starting training...")
trainer.train()
print("Pushing to Hub...")
trainer.push_to_hub()
trackio.finish()
print("Done! Model at: https://huggingface.co/erik1988/elias-memory-agent-v1")