Modelpublic

Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Quality

sourceHugging Faceotherupdated 12h agoView on Hugging Face

0likes311downloads

Model Card

Qwen 3.8 Flash-Next Optimized Quality

The 8-bit build of Qwen 3.8 Flash-Next, for Macs with 256 GB or 512 GB. Requires MTPLX 2.12.0 or later.

Qwen's 125B-A6B Flash-Next, the Qwen4-generation hybrid mixture of experts with Qwen Sparse Attention and a 51B-parameter n-gram table, packed for MTPLX with its multi-token prediction head. The main model and the draft head are 8-bit with group size 64, the structural weights stay in BF16, and the n-gram table is 4-bit with group size 32. On a Mac with 128 GB or more, Optimized Speed is the recommended build.

Memory

The model weights, the draft head and the vision tower need about 128.5 GiB, and the 32 GB n-gram table streams from SSD.

—128 GB: Cannot load. The weights, the draft head and the vision tower alone need about 128.5 GiB.
—256 GB and 512 GB: the Macs this pack is for, with about 59.5 GiB left for context and the session cache. The MTPLX app and CLI list it second there, after Optimized Speed.

Speed

Speed on 256 GB and 512 GB Macs is not measured yet. The 8-bit weights move twice the bytes per token of Optimized Speed, so expect slower decoding.

What is in the pack

Tensor class	Stored precision	Size (GB)
attention	Q8/g64 affine; BF16 scales and biases	0.635044
embeddings	Q8/g64 affine; BF16 scales and biases	0.675430
gdn A log	BF16	0.000003
gdn conv	BF16	0.002949
gdn dt bias	BF16	0.000003
gdn projections	Q8/g64 affine; BF16 scales and biases	2.215342
hyper connections	BF16	1.279263
lm head	Q8/g64 affine; BF16 scales and biases	0.675430
mtp attention	Q8/g64 affine; BF16 scales and biases	0.052920
mtp fc	BF16	0.026214
mtp hyper connections	BF16	0.039485
mtp norms	BF16	0.000089
mtp qsa indexer	Q8/g64 affine; BF16 scales and biases	0.001741
mtp routed experts	Q8/g64 affine; BF16 scales and biases	2.673869
mtp router	Q8/g64 affine; BF16 scales and biases	0.001395
mtp shared expert	Q8/g64 affine; BF16 scales and biases	0.005222
ngram table	Q4/g32 affine; BF16 scales and biases	32.000154
norms	BF16	0.002076
ple integer buffers	I64	0.000000
ple projections and conv	BF16	0.065618
qsa indexer	Q8/g64 affine; BF16 scales and biases	0.020890
routed experts	Q8/g64 affine; BF16 scales and biases	128.345702
router	Q8/g64 affine; BF16 scales and biases	0.066977
shared expert	Q8/g64 affine; BF16 scales and biases	0.250675
vision tower	BF16	0.897862

The download is 169.96 GB. size-checksums.json lists the size and SHA-256 of every other file.

Use it

In the Mac app, pick Qwen 3.8 Flash-Next Optimized Quality. From the command line:

bash

pip install mtplx
mtplx serve --model Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Quality --model-id mtplx-flash-next-optimized-quality

MTPLX samples at the official Qwen 3.8 settings (temperature 1.0, top-p 0.95, top-k 20), and drafts are accepted with exact speculative sampling, so the output follows the model's own distribution.

Built with the flash-next-optimized-quality recipe from Qwen/Qwen3.8-Flash-Next at revision de4b8e4d43b917e7706784d8bb445c9af86a3540. Qwen Community License, preserved in LICENSE. The upstream model card is preserved as README-upstream-qwen.md. License and credits are carried from Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed. Conversion and serving: MTPLX.