I'm a Machine Learning Engineer at Predibase, working on large language models and distributed training. I’m passionate about open-source projects, especially in the field of AI and ML. 🚀

I love working on Ludwig, an open-source toolbox for low-code/no-code deep learning.

Some of my recent work includes:

1. Co-creating Turbo LoRA, a parameter efficient training method to jointly fine-tune for quality and speculative decoding, improving throughput by upto 3.5x for fine-tuned LoRA adapters. Read More
2. Developing a mixture of agents synthetic data generation algorithm for supervised instruction tuning that can beat K-shot GPT-4o using just 10 rows of data. Read More
3. Co-authoring LoRA Land: 310 Fine-tuned LLMs that Rival GPT-4, A Technical Report. Read the paper.
4. Speeding up LLM training by 15x. Read More

You can connect with me on LinkedIn to discuss all things AI 🤖.