owLLM is a framework for training multiple LLMs model with different techniques. We aim to offer support for RLHF/RLAIF/DPO in the future.

Features:

• Train various models such as llama, mistral, phi, stablelm
• Supports fullfinetune, lora, qlora
• Integrated with flash attention, rope scaling, and multipacking
• Works with single GPU or multiple GPUs via FSDP or Deepspeed or DDP
• And more!

owLLM provides a unified repository for fine-tuning
a variety of AI models with ease