Unsloth Fine-Tuning Guide: Train a Custom LLM with Your Own Data

Want AI to understand your business — customer service reply style, code standards, product knowledge. But fine-tuning is slow and expensive, and I've heard a single model can cost tens of thousands. Unsloth says: a few hundred bucks and a few images, done.

💡 What You Will Learn

Want AI to understand your business — customer service reply style, code standards, product knowledge. But fine-tuning is slow and expensive, and I've heard a single model can cost tens of thousands.

📜 Table of Contents

• UnslothWhat Is
• Common PitfallsSummary

UnslothWhat Is

pip install unsloth
# conda
conda create -n unsloth python=3.11
conda activate unsloth
pip install unsloth

{"instruction": "", "input": "", "output": ""}
{"instruction": "", "input": "", "output": ""}

from unsloth import FastLanguageModel
import torch

# Auto/Automatic
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="unsloth/Qwen2.5-7B-bnb-4bit",
    max_seq_length=2048,
    dtype=None,
    load_in_4bit=True,
)

# LoRAParameter
model = FastLanguageModel.get_peft_model(
    model,
    r=16,
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj"],
    lora_alpha=16,
    lora_dropout=0,
    bias="none",
    use_gradient_checkpointing="unsloth",
)

# 
from transformers import TrainingArguments
from trl import SFTTrainer

trainer = SFTTrainer(
    model=model,
    tokenizer=tokenizer,
    train_dataset=dataset,
    args=TrainingArguments(
        per_device_train_batch_size=2,
        gradient_accumulation_steps=4,
        num_train_epochs=3,
        learning_rate=2e-4,
        output_dir="./outputs",
    ),
)
trainer.train()

# 
model.save_pretrained("./my-custom-model")
tokenizer.save_pretrained("./my-custom-model")

model, tokenizer = FastLanguageModel.from_pretrained("./my-custom-model")

prompt = ""
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0]))

Common PitfallsSummary