Popular GitHub Repositories for Flash Mla

中文LLaMA-2 & Alpaca-2大模型二期项目 + 64K超长上下文模型 (Chinese LLaMA-2 & Alpaca-2 LLMs with 64K long context models)

Official release of InternLM series (InternLM, InternLM2, InternLM2.5, InternLM3).

?A curated list of Awesome LLM/VLM Inference Papers with codes: WINT8/4, Flash-Attention, Paged-Attention, Parallelism, etc. ??

FlashInfer: Kernel Library for LLM Serving

MiniMax-01 is a simple implementation of the MiniMax algorithm, a widely used strategy for decision-making in two-player turn-based games like Tic-Tac-Toe. The algorithm aims to minimize the maximum possible loss for the player, making it a popular choice for developing AI opponents in various game scenarios.

MoBA: Mixture of Block Attention for Long-Context LLMs

Your PyTorch AI Factory - Flash enables you to easily configure and run complex AI recipes for over 15 tasks across 7 data domains

Deploy your private Gemini application for free with one click, supporting Gemini 1.5, Gemini 2.0 models.

An agentic company research tool powered by LangGraph and Tavily that conducts deep diligence on companies using a multi-agent framework. It leverages Google Gemini 2.0 Flash and Chat GPT-4o-mini on the backend for inference.

InternEvo is an open-sourced lightweight training framework aims to support model pre-training without the need for extensive dependencies.

Vanilla JS web interface for Gemini 2.0 flash-exp Multimodal API with text, audio, camera, screen inputs and audio responses and function calling

TransMLA: Multi-Head Latent Attention Is All You Need

Co-create a PowerPoint presentation with Generative AI

Implementation of Flash Attention in Jax

Implementation of fused cosine similarity attention in the same style as Flash Attention

Shush is an app that deploys a WhisperV3 model with Flash Attention v2 on Modal and makes requests to it via a NextJS app

?FFPA: Extend FA-2 with Split-D for large headdim, 2x↑ vs SDPA.

?FFPA(Split-D): Yet another Faster Flash Prefill Attention with O(1) GPU SRAM complexity for headdim > 256, ~2x↑?vs SDPA EA.

[ICLR 2025] Palu: Compressing KV-Cache with Low-Rank Projection

Train llm (bloom, llama, baichuan2-7b, chatglm3-6b) with deepspeed pipeline mode. Faster than zero/zero++/fsdp.