DeepInfer

DeepSeek-V2 is a state-of-the-art Mixture-of-Experts (MoE) language model introduced by DeepSeek-AI, characterized by its economical training and efficient inference capabilities. With a total of 236 billion parameters, of which only 21 billion are active per token, it supports a context length of up to 128K tokens. DeepSeek-V2 employs innovative architectures like Multi-head Latent Attention (MLA) for efficient inference by compressing the Key-Value (KV) cache and DeepSeekMoE for cost-effective training through sparse computation. This model significantly outperforms its predecessor, DeepSeek 67B, by saving 42.5% in training costs, reducing the KV cache by 93.3%, and enhancing generation throughput by 5.76 times. Pretrained on an 8.1 trillion token corpus, DeepSeek-V2 excels in language understanding, coding, and reasoning tasks, making it a top-tier performer among open-source models.

vLLM is a high-performance library designed to facilitate efficient inference and serving of Large Language Models (LLMs). Originally developed in the Sky Computing Lab at UC Berkeley, vLLM has evolved into a community-driven project with contributions from both academia and industry. It offers state-of-the-art serving throughput by efficiently managing attention key and value memory through its PagedAttention mechanism. It supports continuous batching of incoming requests and utilizes optimized CUDA kernels, including integration with FlashAttention and FlashInfer, to enhance model execution speed. Additionally, vLLM provides quantization support for GPTQ, AWQ, INT4, INT8, and FP8, as well as speculative decoding capabilities. Users benefit from seamless integration with popular Hugging Face models, support for various decoding algorithms such as parallel sampling and beam search, and compatibility with NVIDIA GPUs, AMD CPUs and GPUs, Intel CPUs, and more.