JT-Ushio / MHA2MLALinks
Towards Economical Inference: Enabling DeepSeek's Multi-Head Latent Attention in Any Transformer-based LLMs
☆192Updated 3 months ago
Alternatives and similar repositories for MHA2MLA
Users that are interested in MHA2MLA are comparing it to the libraries listed below
Sorting:
- Parallel Scaling Law for Language Model — Beyond Parameter and Inference Time Scaling☆443Updated 4 months ago
- TransMLA: Multi-Head Latent Attention Is All You Need (NeurIPS 2025 Spotlight)☆372Updated last week
- Skywork-MoE: A Deep Dive into Training Techniques for Mixture-of-Experts Language Models☆137Updated last year
- [ICLR 2025] DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads☆491Updated 7 months ago
- ☆202Updated 5 months ago
- Implementation for FP8/INT8 Rollout for RL training without performence drop.☆242Updated this week
- Chain of Experts (CoE) enables communication between experts within Mixture-of-Experts (MoE) models☆220Updated 2 weeks ago
- Efficient triton implementation of Native Sparse Attention.☆226Updated 4 months ago
- Unofficial implementation for the paper "Mixture-of-Depths: Dynamically allocating compute in transformer-based language models"☆171Updated last year
- Official implementation of "Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding"☆482Updated 3 weeks ago
- [ICML 2025] |TokenSwift: Lossless Acceleration of Ultra Long Sequence Generation☆113Updated 4 months ago
- [ICLR 2025] COAT: Compressing Optimizer States and Activation for Memory-Efficient FP8 Training☆241Updated last month
- Speed Always Wins: A Survey on Efficient Architectures for Large Language Models☆337Updated last month
- [NeurIPS 2025] Simple extension on vLLM to help you speed up reasoning model without training.☆197Updated 4 months ago
- ☆119Updated 3 months ago
- [ICLR2025] Codebase for "ReMoE: Fully Differentiable Mixture-of-Experts with ReLU Routing", built on Megatron-LM.☆92Updated 9 months ago
- Ling-V2 is a MoE LLM provided and open-sourced by InclusionAI.☆75Updated last week
- A highly capable 2.4B lightweight LLM using only 1T pre-training data with all details.☆217Updated 2 months ago
- Efficient LLM Inference over Long Sequences☆391Updated 3 months ago
- ☆427Updated last month
- MiroMind-M1 is a fully open-source series of reasoning language models built on Qwen-2.5, focused on advancing mathematical reasoning.☆236Updated last month
- SeerAttention: Learning Intrinsic Sparse Attention in Your LLMs☆155Updated last week
- ☆84Updated 6 months ago
- siiRL: Shanghai Innovation Institute RL Framework for Advanced LLMs and Multi-Agent Systems☆217Updated this week
- Lightning Attention-2: A Free Lunch for Handling Unlimited Sequence Lengths in Large Language Models☆328Updated 7 months ago
- Fira: Can We Achieve Full-rank Training of LLMs Under Low-rank Constraint?☆116Updated 11 months ago
- A lightweight reinforcement learning framework that integrates seamlessly into your codebase, empowering developers to focus on algorithm…☆68Updated last month
- 🔥 A minimal training framework for scaling FLA models☆255Updated 3 weeks ago
- Official codebase for "Can 1B LLM Surpass 405B LLM? Rethinking Compute-Optimal Test-Time Scaling".☆271Updated 7 months ago
- qwen-nsa☆75Updated 5 months ago