gargvinayakk/dpo-mistralai-7b-mental-health

peft_config = LoraConfig(
    r=8,
    lora_alpha=16,
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM",
    target_modules=['k_proj', 'gate_proj', 'v_proj', 'up_proj', 'q_proj', 'o_proj', 'down_proj']
)

training_args = TrainingArguments(
    per_device_train_batch_size=4,
    per_device_eval_batch_size=1,
    gradient_accumulation_steps=4,
    save_steps=10,
    gradient_checkpointing=True,
    learning_rate=5e-5,
    lr_scheduler_type="cosine",
    max_steps=100,
    save_strategy="no",
    logging_steps=1,
    output_dir="new_model/",
    optim="paged_adamw_32bit",
    warmup_steps=10,
    fp16=True,
    report_to="wandb",
    evaluation_strategy="steps",
    eval_steps=20,
)