LightGPT-instruct-6B is a language model based on GPT-J 6B, fine-tuned on a high-quality instruction dataset. It provides a transformer-based language model that can generate responses to specific instructions.
- Text generation for completing specific tasks based on given instructions
- 🤖 Transformer-based Language Model
- 🔒 Apache 2.0 License
- 📚 Fine-tuned from EleutherAI/gpt-j-6b
- 📧 Contact: GitHub issues
To deploy LightGPT-instruct, you can use Amazon SageMaker. The example code below demonstrates the deployment process using DJLModel from sagemaker.djl_inference.model.
- LAMBADA PPL: 4.33
- LAMBADA Acc: 65.0%
- Winogrande: 64.6%
- Hellaswag: 63.9%
- PIQA: 75.5%
- The model may struggle with following long instructions or summarizing lengthy texts.
- It might provide incorrect answers to math and reasoning questions.
- Caution should be exercised with regards to potential factual errors or misleading outputs.
- The model is designed for English conversations only.