Why Low-Precision Transformer Training Fails: An Analysis on Flash Attention
☆84Apr 7, 2026Updated 7 months ago
Alternatives and similar repositories for why-low-precision-training-fails
Users that are interested in why-low-precision-training-fails are comparing it to the libraries listed below. We may earn a commission when you buy through links labeled 'Ad' on this page.
Sorting:
- CroissAttn: FP4 attention forward and backward, with analysis of stochastic rounding, zero-sum softmax gradients, and sampled centering.☆16Oct 25, 2026Updated 2 weeks ago
- Spectral Sphere Optimizer☆135Mar 23, 2026Updated 7 months ago
- [ICLR 2025] Systematic Outliers in Large Language Models.☆11Feb 11, 2025Updated last year
- ☆53May 20, 2025Updated last year
- [NeurIPS 2026] POME: Post Optimization Model Edit via Muon-style Projection☆14Oct 9, 2025Updated last year
- Wordpress hosting with auto-scaling - Free Trial Offer • AdFully Managed hosting for WordPress and WooCommerce businesses that need reliable, auto-scalable performance. Cloudways SafeUpdates now available.
- A lightweight CuTe-based CUDA kernel for DuoAttention, optimized for large language model inference.☆12Dec 25, 2025Updated 10 months ago
- RobuQ: Pushing DiTS to W1.58A2 via Robust Activation Quantization☆17Oct 14, 2026Updated 3 weeks ago
- FA4-based Relative Attention Kernel developed by TML and Colfax☆18Sep 11, 2026Updated last month
- ☆18Sep 27, 2026Updated last month
- ☆18May 27, 2026Updated 5 months ago
- Flash-Muon: An Efficient Implementation of Muon Optimizer☆265Jun 15, 2025Updated last year
- ☆12May 8, 2026Updated 5 months ago
- FLA but cuTile☆27Apr 17, 2026Updated 6 months ago
- Fast Hadamard transform in CUDA, with a PyTorch interface☆357Mar 10, 2026Updated 7 months ago
- Managed Database hosting by DigitalOcean • AdPostgreSQL, MySQL, MongoDB, Kafka, Valkey, and OpenSearch available. Automatically scale up storage and focus on building your apps.
- A framework for training-free asynchronous agents☆22Updated this week
- [EMNLP2024] Rethinking Token Reduction for State Space Models☆11Dec 7, 2024Updated last year
- QuTLASS: CUTLASS-Powered Quantized BLAS for Deep Learning☆204Sep 2, 2026Updated 2 months ago
- ☆12Apr 3, 2023Updated 3 years ago
- ☆37Aug 7, 2025Updated last year
- DeeperGEMM: crazy optimized version☆85May 5, 2025Updated last year
- ☆139May 29, 2025Updated last year
- Code for the papers: “Four Over Six: More Accurate NVFP4 Quantization with Adaptive Block Scaling” and “Adaptive Block-Scaled Data Types”☆211Apr 21, 2026Updated 6 months ago
- Muon fsdp 2☆65Aug 8, 2025Updated last year
- Deploy on Railway without the complexity - Free Credits Offer • AdConnect your repo and Railway handles the rest with instant previews. Quickly provision container image services, databases, and storage volumes.
- ☆129Mar 18, 2026Updated 7 months ago
- Quartet II Official Code☆85May 1, 2026Updated 6 months ago
- ☆40Dec 14, 2025Updated 10 months ago
- [ICML 2026 Spotlight] Official implementation of TetraJet-v2: Accurate NVFP4 Training for LLMs, with fully-NVFP4 linear layer with unbias…☆19Jul 3, 2026Updated 4 months ago
- Official implementation for the paper "Controlled Sparsity via Constrained Optimization"☆12Aug 10, 2022Updated 4 years ago
- Discover, compare, and contribute to cutting-edge residual connections designed for large-scale deep learning and depth scaling.☆24Updated this week
- Repo for ECHO: Efficient KV Cache Offloading with Lossless Prefetching for Serving Native Sparse Attention LLMs (OSDI'26)☆42Jul 12, 2026Updated 3 months ago
- ☆47Aug 3, 2026Updated 3 months ago
- SpInfer: Leveraging Low-Level Sparsity for Efficient Large Language Model Inference on GPUs☆70Mar 25, 2025Updated last year
- Managed Database hosting by DigitalOcean • AdPostgreSQL, MySQL, MongoDB, Kafka, Valkey, and OpenSearch available. Automatically scale up storage and focus on building your apps.
- ☆18Sep 22, 2024Updated 2 years ago
- Triton kernels and PyTorch ops for Block Attention Residuals (AttnRes)☆91May 29, 2026Updated 5 months ago
- ☆32Jul 2, 2025Updated last year
- [ICLR 2024] This is the official PyTorch implementation of "QLLM: Accurate and Efficient Low-Bitwidth Quantization for Large Language Mod…☆31Mar 12, 2024Updated 2 years ago
- TiledLower is a Dataflow Analysis and Codegen Framework written in Rust.☆13Nov 23, 2024Updated last year
- Compact and Agent-Native MoE Training System☆360Updated this week
- Comparative study and experimentation on standard vs mHC vs attention residual (full and block)☆13Mar 17, 2026Updated 7 months ago