|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| from datasets import load_dataset
|
| from peft import LoraConfig
|
| from trl import SFTTrainer, SFTConfig
|
|
|
| print("Loading dataset...")
|
| dataset = load_dataset("trl-lib/Capybara", split="train")
|
| dataset = dataset.shuffle(seed=42).select(range(500))
|
| print(f"Using {len(dataset)} examples")
|
|
|
| dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
|
|
|
| config = SFTConfig(
|
| output_dir="qwen25-test",
|
| push_to_hub=True,
|
| hub_model_id="luiscosio/qwen25-test",
|
| num_train_epochs=1,
|
| per_device_train_batch_size=2,
|
| gradient_accumulation_steps=4,
|
| learning_rate=2e-4,
|
| logging_steps=10,
|
| save_strategy="epoch",
|
| bf16=True,
|
| report_to="none",
|
| )
|
|
|
| peft_config = LoraConfig(
|
| r=16,
|
| lora_alpha=32,
|
| bias="none",
|
| task_type="CAUSAL_LM",
|
| target_modules=["q_proj", "v_proj"],
|
| )
|
|
|
| print("Initializing trainer with Qwen2.5-0.5B...")
|
| trainer = SFTTrainer(
|
| model="Qwen/Qwen2.5-0.5B",
|
| train_dataset=dataset_split["train"],
|
| args=config,
|
| peft_config=peft_config,
|
| )
|
|
|
| print("Starting training...")
|
| trainer.train()
|
| trainer.push_to_hub()
|
| print("Done!")
|
|
|
