GPT-NeoX

This is a simple implementation of Eleuther GPT-NeoX for inference and fine-tuning.

• Model definition
• Tokenizer
• Checkpoint downloading and loading helpers
• Utilities
• LLM.int8() quantization

Samples

• Generating text
• Fine-tuning the biases with pipeline-parallel
• Generating text with LLM.int8()

Evaluation

• Evaluating half precision model on a single GPU
• Evaluating LLM.int8() model

Official Eleuther GPT-NoeX is source code is available at eleutherai/gpt-neox.