radiance-exl3 test models: 8-layer Qwen3.8-Flash-Next containers

Three 8-layer slices of Qwen3.8-Flash-Next as radiance containers, used by the correctness and speed gates of radiance-exl3. They are for testing kernels and engine behaviour, not for chat: with 8 of 48 layers the output is not useful text.

file	what it is
`qwen3.8-next-flash-8L-base.rad`	radiance's own recipe (fp8 trunk, iq4 routed experts), first 8 layers
`qwen3.8-flash-next-8L-exl3.rad`	the same with turboderp's 4.05 EXL3 routed experts
`qwen3.8-flash-next-8L-exl3-shared-l0td.rad`	the same with shared-input-vector EXL3 experts (layer 0 turboderp's)

SHA256SUMS covers the first two; config-48L.orig.json, plans.out and converts.out are the build records. Licence: Qwen Community License 1.0 (LICENSE).

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Dyluhn/radiance-exl3-test-models

Base model

Qwen/Qwen3.8-Flash-Next

Finetuned

(100)

this model