Vaibhav Srivastav

In this talk, VB from Hugging Face will share the latest trends shaping open large language models in 2025 — from new model releases and adoption patterns to the challenges of scaling and regulation. The session will highlight where open-source is thriving, what hurdles remain, and what engineers should expect next.

Hugging Face is an open-source company that creates libraries such as transformers, tokenizers, safetensors, llama.cpp, ratchet, and much more. In this talk, we'll delve into the Hugging Face stack and how different local AI frameworks leverage it.

Late last year, we sped up Whisper (https://github.com/Vaibhavs10/insanely-fast-whisper) by 10x. It involved a lot of hacking (like flash attention, batching, torch.compile, etc) and tricks that can be implemented in other models, too. Join me as I walk through those optimisations and tricks.

Have you been following the OpenAI Whisper hype? Want to learn the inner workings of it and learn how to fin-tune it in your own language? Follow along in a code-first workshop on building, deploying and inferring state of the art speech recognition models.