Fine-Tuning LLMs with Hugging Face

Two approaches: no-code AutoTrain or code-based LoRA with PEFT.

Option 1: AutoTrain (No Code)

1. Go to [huggingface.co/autotrain](https://huggingface.co/autotrain)

2. Select a base model (e.g., Llama 3.2 3B)

3. Upload your training data (CSV or JSONL)

4. Configure: learning rate, epochs, batch size

5. Click Train — results in 1-4 hours

Option 2: LoRA with PEFT (Code)

```python

from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments

from peft import LoraConfig, get_peft_model

from trl import SFTTrainer

Load base model

model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.2-3B-Instruct")

tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.2-3B-Instruct")

Configure LoRA

lora_config = LoraConfig(

r=16,

lora_alpha=32,

target_modules=["q_proj", "v_proj"],

lora_dropout=0.05,

task_type="CAUSAL_LM"

)

model = get_peft_model(model, lora_config)

Train

trainer = SFTTrainer(

model=model,

train_dataset=dataset,

args=TrainingArguments(

output_dir="./results",

num_train_epochs=3,

per_device_train_batch_size=4,

learning_rate=2e-4,

),

)

trainer.train()

Push to Hub

model.push_to_hub("your-username/my-fine-tuned-model")

```

When to Use What

| Method | Best For | GPU Needed |

|--------|----------|------------|

| AutoTrain | Quick experiments, non-technical users | Cloud (included) |

| LoRA/QLoRA | Custom training, research, production | 16GB+ VRAM |

| Full fine-tuning | Maximum quality, large datasets | 80GB+ VRAM |