Skip to content
RCreddit.com·

Unsloth UD-quants - Qwen 3.8 27b for example - worth using 8-bit or stick with faster 6 bit for coding?

AI summary

A developer is asking the community about the optimal quantization for coding projects using Unsloth UD-quants, specifically Qwen 3.8 27b. They are debating whether to use 8-bit quants, if memory allows, or to stick with the faster UD-Q6_K_XL (6-bit). The developer notes that 6-bit is noticeably faster on their setup, despite others claiming the difference is imperceptible, and seeks community input on their experiences and test results.

Time & source

Times shown in UTC

Display time zone: UTC

Local time zone unavailable; showing UTC.

PublishedOffset at this time: UTC+0Sep 12, 2026, 04:18 UTC

IngestedOffset at this time: UTC+0Sep 12, 2026, 22:01 UTC

Published
Sep 12, 2026, 04:18
Ingested
Sep 12, 2026, 22:01
Source type
Dev community
Tier
Community
Source status
Healthy

Tier is a per-source editorial setting, not a per-item score.

For those using these models for coding in larger projects where things can get complex, do you find yourself using the 8-bit quants if you have enough memory? Or do you stick with UD-Q6_K_XL?

The 6-bit is faster, noticeably so on my setup. And I keep seeing people say it's imperceptible. I've been doing tests myself, and well, I can't tell, but maybe that's just because I'm an idiot.

That said, can you tell? Have you ever done some tests to see?

Source·reddit.com