High-quality datasets, tools, and concepts for LLM fine-tuning.

Schedule-Free Optimization in PyTorch

Train GEMMA on TPU/GPU! (Codebase for training Gemma-Ko Series)

Transformer with Mu-Parameterization, implemented in Jax/Flax. Supports FSDP on TPU pods.

Code for SaGe subword tokenizer (EACL 2023)

Whisper realtime streaming for long speech-to-text transcription and translation

Repo for "Monarch Mixer: A Simple Sub-Quadratic GEMM-Based Architecture"

A minimal PyTorch re-implementation of the OpenAI GPT (Generative Pretrained Transformer) training

Large language models (LLMs) made easy, EasyLM is a one stop solution for pre-training, finetuning, evaluating and serving LLMs in JAX/Flax.

Converts text to speech in realtime

Code and documents of LongLoRA and LongAlpaca (ICLR 2024 Oral)

Structured Text Generation

[NeurIPS'23 Oral] Visual Instruction Tuning (LLaVA) built towards GPT-4V level capabilities and beyond.

Extend existing LLMs way beyond the original training length with constant memory usage, without retraining

Official codes of DiffBIR: Towards Blind Image Restoration with Generative Diffusion Prior

Multipack distributed sampler for fast padding-free training of LLMs

Making large AI models cheaper, faster and more accessible

Positional Skip-wise Training for Efficient Context Window Extension of LLMs to Extremely Length (ICLR 2024)

Make Praat Picture style plots of acoustic data

This repository contains the official implementation of the research paper, "FastViT: A Fast Hybrid Vision Transformer using Structural Reparameterization" ICCV 2023

🕸️ Web apps in pure Python 🐍

JAX implementation of the Llama 2 model

Transformers with Arbitrarily Large Context

Obtain Word Alignments using Pretrained Language Models (e.g., mBERT)

LoRA for arbitrary JAX models and functions

Conformal classifiers, regressors and predictive systems

Annotation Tool and Image Search

JAX Synergistic Memory Inspector

Automated dense category annotation engine that serves as the initial semantic labeling for the Segment Anything dataset (SA-1B).