ANEMLL quantized exports of Qwen3.8-27B and its paired DFlash2 drafter for
Core AI conversion

Original model: Qwen/Qwen3.8-27B
Developed by the Qwen Team.
Upstream weights: Copyright 2026 Alibaba Cloud.
Reviewed revision: 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0
https://huggingface.co/Qwen/Qwen3.8-27B
The upstream license is preserved unchanged in LICENSE.

Original drafter: ProCreations/Ternary-Bonsai-2-27B-DFlash2
Reviewed drafter revision: 4cfb6ad03268fed0f60ca96c1a659c0b1c77e50b
https://huggingface.co/ProCreations/Ternary-Bonsai-2-27B-DFlash2
Its original Apache-2.0 LICENSE and NOTICE are preserved unchanged in
drafter/LICENSE and drafter/NOTICE. The ProCreations adaptation credits the
z-lab Qwen3.8-27B-DFlash2 donor and the Prism Ternary-Bonsai-2-27B training
target; drafter/DFLASH2_SOURCE.json records their pinned identities.

ANEMLL independently quantized the weights (mixed-precision GPTQ lookup tables,
per-channel scales, online rotations, rank-64 residual corrections) and copied
the small tensors, configuration and tokenizer files the converter needs without
changes. See MODIFICATIONS.md and inventory.json.

This NOTICE was added by ANEMLL; the reviewed upstream revision had no NOTICE
file. It does not replace, amend, or add restrictions to the Apache License 2.0.

Qwen, Alibaba Cloud, ProCreations, z-lab, Prism ML and Apple are identified for
credit and context. No official
affiliation or endorsement is claimed.

The model-weight license and the ANEMLL Forge source-code license are separate:
ANEMLL Forge source code and documentation are MIT-licensed; these model-derived
assets remain under the upstream Apache-2.0 license.