graphcore-research / out-of-the-box-fp8-trainingLinks
Demo of the unit_scaling library, showing how a model can be easily adapted to train in FP8.
☆46Updated last year
Alternatives and similar repositories for out-of-the-box-fp8-training
Users that are interested in out-of-the-box-fp8-training are comparing it to the libraries listed below
Sorting:
- Experiment of using Tangent to autodiff triton☆80Updated last year
- ☆87Updated last year
- A place to store reusable transformer components of my own creation or found on the interwebs☆59Updated last week
- A library for unit scaling in PyTorch☆129Updated last month
- PyTorch centric eager mode debugger☆47Updated 8 months ago
- CUDA implementation of autoregressive linear attention, with all the latest research findings☆44Updated 2 years ago
- ☆118Updated last year
- train with kittens!☆62Updated 9 months ago
- Minimal (400 LOC) implementation Maximum (multi-node, FSDP) GPT training☆130Updated last year
- Make triton easier☆47Updated last year
- Triton Implementation of HyperAttention Algorithm☆48Updated last year
- CUDA and Triton implementations of Flash Attention with SoftmaxN.☆73Updated last year
- Utilities for Training Very Large Models☆58Updated 10 months ago
- Automatically take good care of your preemptible TPUs☆36Updated 2 years ago
- Hacks for PyTorch☆19Updated 2 years ago
- Personal solutions to the Triton Puzzles☆19Updated last year
- The evaluation framework for training-free sparse attention in LLMs☆90Updated 2 months ago
- Faster Pytorch bitsandbytes 4bit fp4 nn.Linear ops☆30Updated last year
- This repository contains the experimental PyTorch native float8 training UX☆224Updated last year
- ☆53Updated last year
- An implementation of the Llama architecture, to instruct and delight☆21Updated 2 months ago
- FlexAttention w/ FlashAttention3 Support☆27Updated 10 months ago
- Unofficial but Efficient Implementation of "Mamba: Linear-Time Sequence Modeling with Selective State Spaces" in JAX☆86Updated last year
- ☆21Updated 5 months ago
- Using FlexAttention to compute attention with different masking patterns☆44Updated 11 months ago
- ☆56Updated 10 months ago
- Official repository for the paper "Approximating Two-Layer Feedforward Networks for Efficient Transformers"☆38Updated 2 months ago
- ☆29Updated 2 years ago
- A bunch of kernels that might make stuff slower 😉☆58Updated this week
- JAX bindings for Flash Attention v2☆90Updated 3 weeks ago