Sourab Mangrulkar

Machine Learning Engineer

Sourab Mangrulkar has a Bachelor’s degree from Nation Institute of Technology Goa (NIT Goa) with Specialization in ML and Deep Learning.
He has more than 5 years of experience with 2 years in Microsoft, 2 at Amazon and over 1 at Hugging Face. He is currently ML Engineer at Hugging Face and previously was Applied Scientist at Amazon and Data & Applied Scientist at Microsoft. The problems he worked are diverse ranging from clock through rate prediction in Ads Systems, relevance modeling involving multilingual and multimodal models in Sponsored products and recommendations. At Hugging Face, he is core contributor to Accelerate and PEFT open source projects which democratise large language model training and inference at scale to be simple, extensible and efficient. He has integrated DeepSpeed, PyTorch FSDP, Megatron-LM, SageMaker Data Parallel into Accelerate and started the project PEFT which contain performant methods that enable finetuning very large models on consumer hardware while preserving the performance. Ping him for best practices with respect to large model training, inference and parameter-efficient fine-tuning methods.

Generative AI Parameter-Efficient Fine-Tuning: Doing more with less
2:30 pm - 3:30 pm-Audi 1

Day 2

Download Full Agenda