langtune 0.1.11

Efficient LoRA Fine-Tuning for Large Language Models - Train smarter, not harder.

Langtune: LoRA Fine-Tuning for Text LLMs

Fine-tune your first LLM in under 5 minutes

---

What You'll Need

# Quick system check
python --version 

# Check GPU support (Optional but recommended)
python -c "import torch; print('GPU ready!' if torch.cuda.is_available() else 'CPU mode - still works!')"

Install LangTrain

# Step 1: Create a clean environment (recommended)
python -m venv langtrain-env
source langtrain-env/bin/activate  # Windows: langtrain-env\Scripts\activate

# Step 2: Install LangTune
pip install langtune

# Step 3: Verify it worked
python -c "import langtune; print('✅ LangTune installed!')"

Train Your First Model

from langtune import LoRATrainer

# Step 1: Define your training data
training_data = [
    {"user": "Hello!", "assistant": "Hi there! How can I help you today?"},
    {"user": "What can you do?", "assistant": "I can answer questions, have conversations, and help with various tasks!"},
    {"user": "Thanks!", "assistant": "You're welcome! Feel free to ask anything else."}
]

# Step 2: Create the trainer
# This sets up everything for you automatically
trainer = LoRATrainer(
    model_name="microsoft/DialoGPT-medium",
    output_dir="./my_first_chatbot",
)

# Step 3: Train!
trainer.train(training_data)

# Step 4: Test your model
response = trainer.chat("Hello!")
print(f"Your AI says: {response}")

Use Your Trained Model

from langtune import ChatModel

# Load your trained model
model = ChatModel.load("./my_first_chatbot")

# Have a conversation
print(model.chat("Hello!"))
print(model.chat("What can you do?"))

Using Your Own Data

from langtune import LoRATrainer

trainer = LoRATrainer(
    model_name="microsoft/DialoGPT-medium",
    output_dir="./custom_chatbot",
)

# Method 1: Load from a JSONL file
# File should contain: {"user": "...", "assistant": "..."}
trainer.train_from_file("my_conversations.jsonl")

# Method 2: Load from Hugging Face datasets
trainer.train_from_hub("your_username/your_dataset")

Next Steps

1. Train a larger model: Use QLoRATrainer for 4-bit quantization (runs Llama-3-8B on 6GB VRAM!).
2. Deploy as API: Use langtune.deploy("./my_model", port=8000).
3. Read the Docs: Check out langtrain.xyz/docs.

---

Architecture Overview

Langtune uses a modular transformer backbone with LoRA adapters injected into attention and MLP layers.

flowchart TD
 subgraph LoRA_Adapters["LoRA Adapters"]
        LA1(["LoRA Adapter 1"])
        LA2(["LoRA Adapter 2"])
  end
    A(["Input Tokens"]) --> B(["Embedding Layer"])
    B --> D1(["Encoder Layer 1"])
    D1 --> D2(["Encoder Layer 2"])
    LA1 -.-> D1
    LA2 -.-> D2
    D2 --> F(["Output Logits"])
     LA1:::loraStyle
     LA2:::loraStyle
    classDef loraStyle fill:#e1f5fe,stroke:#0277bd,stroke-width:2px

Contributing

Contributions are welcome! See CONTRIBUTING.md.

License

MIT License. See LICENSE.

Citation

@software{langtune2025,
  author = {Pritesh Raj},
  title = {langtune: LLMs with Efficient LoRA Fine-Tuning},
  url = {https://github.com/langtrain-ai/langtune},
  year = {2025}
}