rpe-ai-v4

trainer = SFTTrainer(
    model = model,
    tokenizer = tokenizer,
    train_dataset = dataset,
    dataset_text_field = "text",
    max_seq_length = 4096, # Increased to be safe for long receipts
    dataset_num_proc = 2,
    packing = True, # Much faster training
    args = TrainingArguments(
        per_device_train_batch_size = 1,
        gradient_accumulation_steps = 16, # smoother gradients (Effective Batch Size = 16)
        warmup_steps = 10, # Slight increase for stability
        num_train_epochs = 50, 
        learning_rate = 2e-4,
        fp16 = not is_bfloat16_supported(),
        bf16 = is_bfloat16_supported(),
        logging_steps = 1,
        optim = "adamw_8bit",
        weight_decay = 0.01,
        lr_scheduler_type = "linear",
        seed = 3407,
        output_dir = "outputs",
        report_to = "none",
        save_total_limit = 5, # <--- ONLY KEEP THE LAST 5 CHECKPOINTS
        save_steps = 350,     # <--- SAVE ONCE PER EPOCH (5600 / 16 = 350 steps)

    ),
)