Qwen3.8-Flash-Next MXFP4 for Whallm
This repository contains the text-only installed model used by Whallm, formerly named DeepSeekV4SSD.
The routed experts were converted from the pinned
Qwen/Qwen3.8-Flash-Next-FP8 revision
bcd9f01ddc9cff2316eb84281bebcd5b058bddce to MXFP4. The conversion uses
4-bit values, group size 32, and Whallm conversion version 2.
This is not a Transformers checkpoint. The files use the Whallm installed model format. Install the model from the Whallm app.
The repository contains the complete manifest.json, common tensors,
N-gram store, tokenizer files, and 48 routed expert layer files. The manifest
records the size and SHA-256 of every installed model file.
Qwen vision, video, MTP, and DSpark weights are not included.
The model remains subject to the Qwen Community License 1.0 in LICENSE.
- Downloads last month
- -