sanchit-gandhi/whisper-jax
Optimized JAX/Flax implementation of OpenAI's Whisper speech-to-text model achieving up to 70x speedup over the original PyTorch version.
Not currently ranked — collecting fresh signals.
star history
This repository provides a high-performance implementation of OpenAI’s Whisper model using JAX and Flax. It leverages the Hugging Face Transformers implementation as a foundation while achieving significant inference speedups, particularly on TPU hardware where it can transcribe 30 minutes of audio in approximately 30 seconds. The project offers a FlaxWhisperPipeline abstraction for easy usage and supports CPU, GPU, and TPU execution.
Frequently asked
- What is sanchit-gandhi/whisper-jax?
- Optimized JAX/Flax implementation of OpenAI's Whisper speech-to-text model achieving up to 70x speedup over the original PyTorch version.
- Is whisper-jax open source?
- Yes — sanchit-gandhi/whisper-jax is open source, released under the Apache-2.0 license.
- What language is whisper-jax written in?
- sanchit-gandhi/whisper-jax is primarily written in Jupyter Notebook.
- How popular is whisper-jax?
- sanchit-gandhi/whisper-jax has 4.7k stars on GitHub.
- Where can I find whisper-jax?
- sanchit-gandhi/whisper-jax is on GitHub at https://github.com/sanchit-gandhi/whisper-jax.