https://github.com/promptslab/llmtuner

FineTune LLMs in few lines of code (Text2Text, Text2Speech, Speech2Text)
https://github.com/promptslab/llmtuner

fine-tuning fine-tuning-llm finetune finetune-gpt finetune-llama finetune-llm finetune-llms finetune-whisper finetunechatgpt finetuning finetuning-large-language-models finetuning-rl llm llm-framework llm-inference llm-training llmops llmtuner whisper whisper-finetune

Last synced: 4 days ago
JSON representation

FineTune LLMs in few lines of code (Text2Text, Text2Speech, Speech2Text)

• Host: GitHub
• URL: https://github.com/promptslab/llmtuner
• Owner: promptslab
• License: apache-2.0
• Created: 2023-04-28T12:44:42.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2024-01-13T10:32:58.000Z (11 months ago)
• Last Synced: 2024-12-11T15:49:19.018Z (11 days ago)
• Topics: fine-tuning, fine-tuning-llm, finetune, finetune-gpt, finetune-llama, finetune-llm, finetune-llms, finetune-whisper, finetunechatgpt, finetuning, finetuning-large-language-models, finetuning-rl, llm, llm-framework, llm-inference, llm-training, llmops, llmtuner, whisper, whisper-finetune
• Language: Python
• Homepage:
• Size: 591 KB
• Stars: 232
• Watchers: 4
• Forks: 15
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        




LLMTuner





  
LLMTuner: Fine-Tune Llama, Whisper, and other LLMs with best practices like LoRA, QLoRA, through a sleek, scikit-learn-inspired interface.



 


  

    

  

  

    

  

  

    

  

  

    

  



## Installation

### With pip

This repository is tested on Python 3.7+

You should install Promptify using Pip command

```bash

pip3 install git+https://github.com/promptslab/LLMTuner.git

```

## Quick tour

To finetune Large models, we provide the `Tuner` API.

```python

from llmtuner import Tuner, Dataset, Model, Deployment

# Initialize the Whisper model with parameter-efficient fine-tuning

model = Model("openai/whisper-small", use_peft=True)

# Create a dataset instance for the audio files

dataset = Dataset('/path/to/audio_folder')

# Set up the tuner with the model and dataset for fine-tuning

tuner = Tuner(model, dataset)

# Fine-tune the model

trained_model = tuner.fit()

# Inference with Fine-tuned model

tuner.inference('sample.wav')

# Launch an interactive UI for the fine-tuned model

tuner.launch_ui('Model Demo UI')

# Set up deployment for the fine-tuned model

deploy = Deployment('aws')  # Options: 'fastapi', 'aws', 'gcp', etc.

# Launch the model deployment

deploy.launch()

```

Features 🤖 



  


🏋️‍♂️ Effortless Fine-Tuning: Finetune state-of-the-art LLMs like Whisper, Llama with minimal code

  


⚡️ Built-in utilities for techniques like LoRA and QLoRA 

  


⚡️ Interactive UI: Launch webapp demos for your finetuned models with one click

  


🏎️ Simplified Inference: Fast inference without separate code

  


🌐 Deployment Readiness: (Coming Soon) Deploy your models with minimal effort to aws, gcp etc, ready to share with the world.



### Supported Models :

| Task Name | Colab Notebook | Status |

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

| Fine-Tune Whisper | [Fine-Tune Whisper](https://colab.research.google.com/drive/1j_1AcPRk4s1uivVRSwrsOfodjPv55Jpc?usp=sharing) | ✅  |

| Fine-Tune Whisper Quantized  | [LoRA](https://colab.research.google.com/drive/1ia9KvqEGOxARtJScPBY6ccF8l41-w_l5?usp=sharing) | ✅    |

| Fine-Tune Llama | [Coming soon..](#) | ✅    |

## Community 



If you are interested in Fine-tuning Open source LLMs, Building scalable Large models, Prompt-Engineering, and other latest research discussions, please consider joining PromptsLab









```

@misc{LLMtuner2023,

  title = {LLMTuner: Fine-Tune Large Models with best practices through a sleek, scikit-learn-inspired interface.},

  author = {Pal, Ankit},

  year = {2023},

  howpublished = {\url{https://github.com/promptslab/LLMtuner}}

}

```

## 💁 Contributing

We welcome any contributions to our open source project, including new features, improvements to infrastructure, and more comprehensive documentation. 

Please see the [contributing guidelines](#)