Is it worth running Qwen 3.8 Flash Next on 4x3090 vs 27B?
Heat trend
The percentage is based on available heat signal, not comment count or independent people.
A developer is seeking advice on whether to use Qwen 3.8 Flash Next on a 4x3090 setup instead of 27B, citing frustration with 27B's indecisiveness despite its proximity to solving problems. They note that a 4-bit quantization of Flash Next might fit and be faster, but express concern that the architecture may not be fully ready yet, and are looking for guidance from more experienced users.
Can someone please tell me if it's worth running Qwen 3.8 Flash Next on 4x3090 yet over 27B?
27B is good but damn it is indecisive. I am getting frustrated watching it get "so close" to solving a problem, only to do another 2 hours of "let me just check/prove/etc"
It looks like a 4 bit quant of Flash Next should fit with the ngrams in SSD and be a lot faster but it also sounds like the architecture isn't quite there yet
Can someone smarter and more patient than me tell me what to do pls? thanks