Qwen 3.8/3.6 Blend with DFlash2 v0.3 for Splash

Installer archive for the Junie Local Splash preview. The target and drafter use Splash packed Q4 weights. Tested with Splash 1.1.0 on Apple silicon.

The ZIP contains one Qwen3.8-3.6-27B-blend-Splash-v0.3/ directory with target, draft, tokenizer, vision weights and a per-file checksum manifest. It is not a Transformers or MLX safetensors export. The runtime selects INT8 or BF16 KV at launch; KV format is not baked into these weights.

Source revisions:

Target: vvsotnikov/Qwen3.8-27B-test-MLX-4bit at 067628273633243c2f39bd6f1958088594118874.
Drafter: vvsotnikov/Qwen3.8-27B-test-DFlash2-v0.3 at 00308d84d850c56ade9158072c5b2d9a321949f0 (step 500).

The package uses seven draft proposals. Validation so far covers text and native tool calls; the presence of vision weights is not image-quality validation. See manifest.json for packed formats and every weight checksum, and SHA256SUMS for the installer ZIP checksum.

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 vvsotnikov/Qwen3.8-3.6-27B-blend-Splash-v0.3

Base model

vvsotnikov/Qwen3.8-27B-test-DFlash2-v0.3

Finetuned

(2)

this model