vLLM roundup December 2024

Welcome to Neural Magic's monthly vLLM roundup! We are excited to announce the agreement to be acquired by Red Hat. Joining forces with the industry's open source leader will enable us to bring our cutting-edge AI model optimization and accelerated inference technology to a worldwide audience of enterprises adopting open LLM capabilities.

Keep scrolling for exciting vLLM updates and opportunities to engage with the community!

Bi-Weekly vLLM Office Hours

Recent Recordings

vLLM Project Update: 2024 Retrospective and 2025 Roadmap | Watch Now

Exploring Machete, a Mixed-Input GEMM Kernel for Hopper GPUs | Watch Now

Disaggregated Prefill and KV Cache Storage in vLLM | Watch Now

SOTA Tool-Calling Implementation in vLLM | Watch Now

Take Your AI Performance to the Next Level

2:4 Sparse Llama: Smaller Models for Efficient GPU Inference
Large language models (LLMs) are approaching their limits in terms of traditional scaling, with billions of parameters added for relatively small accuracy gains and advanced quantization techniques squeezing out the last possible bits before accuracy plummets.

Keep Reading

We Ran Over Half a Million Evaluations on Quantized LLMs: Here's What We Found
Quantizing models to lower precision formats, such as 8-bit or 4-bit, significantly reduces computational costs and accelerates inference.

Keep Reading

Introducing Machete, a Mixed-Input GEMM Kernel Optimized for NVIDIA Hopper GPUs
Mixed-input quantization is a technique that processes weights and activations at different precisions in neural networks.

Keep Reading

Research From Our Labs 🧪

1️⃣ "Give Me BF16 or Give Me Death"? Accuracy-Performance Trade-Offs in LLM Quantization | Read Here

2️⃣ PV-Tuning: Beyond Straight-Through Estimation for Extreme
LLM Compression | Read Here

3️⃣ QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs | Read Here

4️⃣ The Iterative Optimal Brain Surgeon: Faster Sparse Recovery by Leveraging Second-Order Information | Read Here

5️⃣ MicroAdam: Accurate Adaptive Optimization with Low Space Overhead and Provable Convergence | Read Here

vLLM has surpassed 32,000 stars! 🌟 Be sure to add your star and join the community. Thank you for your support.