vLLM (@vllm_project) 's Twitter Profile
vLLM

@vllm_project

A high-throughput and memory-efficient inference and serving engine for LLMs. Join slack.vllm.ai to discuss together with the community!

ID: 1774187564276289536

linkhttps://github.com/vllm-project/vllm calendar_today30-03-2024 21:31:01

327 Tweet

12,12K Takipçi

15 Takip Edilen

vLLM (@vllm_project) 's Twitter Profile Photo
vLLM
@vllm_project

17 days ago

How does DeepSeek Sparse Attention (DSA) work? It has 2 components: the Lightning Indexer and Sparse Multi-Latent Attention (MLA). The indexer keeps a small key cache of 128 per token (vs. 512 for MLA). It scores incoming queries. The top-2048 tokens to pass to Sparse MLA.

How does <a href="/deepseek_ai/">DeepSeek</a> Sparse Attention (DSA) work? It has 2 components: the Lightning Indexer and Sparse Multi-Latent Attention (MLA). The indexer keeps a small key cache of 128 per token (vs. 512 for MLA). It scores incoming queries. The top-2048 tokens to pass to Sparse MLA.