Home
Chat
Deep Research
News
Pricing
New Chat
Sign in
Sign up
Home
Chat
Deep Research
News
Pricing
New Chat
提
问News
Tech, finance and policy — daily must-reads
Home
Tech
Finance
Policy & Regulation
简体
繁體
EN
Tiwen News
Tags: 254k context benchmark
254k context benchmark
Tag "254k context benchmark" — all articles
22
2026-11
Inkling-2 Mini hits local hardware: single RTX 3090 and dual 16GB 50-series cards run 254k context
The roughly 25B-parameter Inkling-2 Mini has been put through multiple local tests in the AI community: the Unsloth UD-Q4_K_XL quantized build runs on a single RTX 3090 and is claimed to beat larger MoE models, while capy-flash v0.16 adds support and runs 254k context on 2×RTX 5070 Ti 16GB at 149 tok/s chat decode and 194 tok/s code decode, though long-context degradation is pronounced.
X · LiTianleli、Reddit · FeydRowan