foundation-model-stack / fms-extrasLinks
β24Updated 11 months ago
Alternatives and similar repositories for fms-extras
Users that are interested in fms-extras are comparing it to the libraries listed below
Sorting:
- Code for the paper "QMoE: Practical Sub-1-Bit Compression of Trillion-Parameter Models".β277Updated last year
- π Efficiently (pre)training foundation models with native PyTorch features, including FSDP for training and SDPA implementation of Flashβ¦β262Updated last month
- β217Updated 7 months ago
- This repository contains the experimental PyTorch native float8 training UXβ224Updated last year
- Tree Attention: Topology-aware Decoding for Long-Context Attention on GPU clustersβ129Updated 8 months ago
- A safetensors extension to efficiently store sparse quantized tensors on diskβ153Updated this week
- Load compute kernels from the Hubβ258Updated this week
- β202Updated 8 months ago
- Simple implementation of Speculative Sampling in NumPy for GPT-2.β95Updated 2 years ago
- β74Updated 5 months ago
- The source code of our work "Prepacking: A Simple Method for Fast Prefilling and Increased Throughput in Large Language Models" [AISTATS β¦β61Updated 10 months ago
- Triton-based implementation of Sparse Mixture of Experts.β233Updated last week
- π Collection of components for development, training, tuning, and inference of foundation models leveraging PyTorch native components.β208Updated last week
- CUDA and Triton implementations of Flash Attention with SoftmaxN.β73Updated last year
- Demo of the unit_scaling library, showing how a model can be easily adapted to train in FP8.β46Updated last year
- ring-attention experimentsβ149Updated 10 months ago
- Code for paper: "QuIP: 2-Bit Quantization of Large Language Models With Guarantees"β379Updated last year
- β53Updated 10 months ago
- PB-LLM: Partially Binarized Large Language Modelsβ153Updated last year
- Make triton easierβ47Updated last year
- Experiment of using Tangent to autodiff tritonβ80Updated last year
- Layer-Condensed KV cache w/ 10 times larger batch size, fewer params and less computation. Dramatic speed up with better task performanceβ¦β153Updated 4 months ago
- Official implementation for Training LLMs with MXFP4β77Updated 4 months ago
- β121Updated last year
- β127Updated last year
- β110Updated last year
- Cold Compress is a hackable, lightweight, and open-source toolkit for creating and benchmarking cache compression methods built on top ofβ¦β142Updated last year
- π· Build compute kernelsβ119Updated this week
- A high-throughput and memory-efficient inference and serving engine for LLMsβ267Updated 10 months ago
- β196Updated 3 months ago