Get the latest tech news

Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

I test Unsloth GGUFs of Qwen3.8 27B (Q4_K_M, UD-Q2_K_XL, UD-IQ1_S) with llama.cpp on GPQA Diamond, IFBench, Terminal-Bench 2.1. Q4_K_M matches BF16 abd fits an RTX 4090.

None

Get the Android app

Or read this on Hacker News

Related news:

Grab Hoto's 25-bit electric screwdriver set for under $30 right now, perfect for PC builds and projects — limited-time deal knocks 40% off the usual price for a multi-torque driver with a 1,500 mAh battery and 25 heavy-duty bits for hobbyists

The balkanization of virtualization will de-throne VMware, which doesn't mind a bit

The Pixel 11 Pro Fold is a solid upgrade, but after testing it against other foldables, I wish Google had gone a bit further

« Apple Watch GPS vs GPS + Cellular: What's the difference?

Google’s revived nuclear power plant gets $1.9B loan from US government »