Launch80/Qwopus3.8-27B-Coder-Flash-Int4-baked

Image-Text-to-Text • 28B • Updated 2 days ago • 112

Note The INT4 checkpoint (baked lm_head + MTP head). Pin a version with --revision v1.0 / v1.1. Study: https://launch80.com/blog/qwopus-coder-int4-bake-study

Note BF16 top-64 reference logits, exact eval token ids, per-item task results, calibration set and scripts: score your own quant on our harness.

Launch80/Qwen3.8-27B-DFlash2-Int4-baked-drafter

2B • Updated 3 days ago • 118

Note INT4 DFlash2 drafter used by the v1.1 fast-decode serving profile (k=7, 128k context).

Jackrong/Qwopus3.8-27B-Coder-Flash

Image-Text-to-Text • 28B • Updated 6 days ago • 1.08k • 31

Note The BF16 source model by Jackrong.