foundation-model-stack / fms-fsdpLinks
๐ Efficiently (pre)training foundation models with native PyTorch features, including FSDP for training and SDPA implementation of Flash attention v2.
โ258Updated last week
Alternatives and similar repositories for fms-fsdp
Users that are interested in fms-fsdp are comparing it to the libraries listed below
Sorting:
- ๐ Collection of components for development, training, tuning, and inference of foundation models leveraging PyTorch native components.โ206Updated this week
- This repository contains the experimental PyTorch native float8 training UXโ224Updated last year
- Triton-based implementation of Sparse Mixture of Experts.โ230Updated 8 months ago
- Applied AI experiments and examples for PyTorchโ289Updated 2 months ago
- Load compute kernels from the Hubโ220Updated this week
- Large Context Attentionโ719Updated 6 months ago
- Implementation of ๐ Ring Attention, from Liu et al. at Berkeley AI, in Pytorchโ532Updated 2 months ago
- Scalable toolkit for efficient model reinforcementโ558Updated this week
- ring-attention experimentsโ146Updated 9 months ago
- โ206Updated 5 months ago
- Fast low-bit matmul kernels in Tritonโ338Updated last week
- A Quirky Assortment of CuTe Kernelsโ374Updated this week
- Explorations into some recent techniques surrounding speculative decodingโ275Updated 7 months ago
- Official repository for DistFlashAttn: Distributed Memory-efficient Attention for Long-context LLMs Trainingโ212Updated 11 months ago
- โ227Updated this week
- Cataloging released Triton kernels.โ247Updated 6 months ago
- Fault tolerance for PyTorch (HSDP, LocalSGD, DiLoCo, Streaming DiLoCo)โ372Updated this week
- LLM KV cache compression made easyโ560Updated last week
- Megatron's multi-modal data loaderโ230Updated last week
- ๐ฅ A minimal training framework for scaling FLA modelsโ209Updated last month
- A subset of PyTorch's neural network modules, written in Python using OpenAI's Triton.โ565Updated this week
- โ123Updated 2 months ago
- โ114Updated last year
- Cold Compress is a hackable, lightweight, and open-source toolkit for creating and benchmarking cache compression methods built on top ofโฆโ138Updated 11 months ago
- Collection of kernels written in Triton languageโ139Updated 3 months ago
- Efficient LLM Inference over Long Sequencesโ385Updated last month
- Ring attention implementation with flash attentionโ828Updated last week
- A safetensors extension to efficiently store sparse quantized tensors on diskโ141Updated this week
- [NeurIPS 2024] KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantizationโ365Updated 11 months ago
- โ147Updated 2 years ago