Inference Infrastructure & ServingMIT Han Lab (Ji Lin et al.)MIT

AWQ

AWQ

Why it matters

Quantization that protects only the top weights as judged by activations; TinyChat runs more than 3× faster than HF FP16

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2306.00978

Added in Week 35, 2026 · See that week

Same week

RefinedWebPaperDPOPaper

Nearby on the same axis

UC Berkeley releases vLLM + PagedAttentionOfficialllama.cpp community introduces the GGUF formatPressGeorgi Gerganov creates the llama.cpp repositoryOfficialAccelerating LLM Decoding with Speculative SamplingPaper