NOTICE
1.8 KB · 33 lines · plaintext Raw
1 Qwen3.8-27B-pi — attribution and modification notice
2
3 Base model
4 Qwen3.8-27B is provided by the Qwen team / Alibaba Cloud.
5 Copyright 2026 Alibaba Cloud.
6 Source: https://huggingface.co/Qwen/Qwen3.8-27B
7 The root LICENSE is copied unchanged from upstream revision 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0.
8
9 Pi modifications and packaging
10 Copyright 2026 Thomas Kim (bytkim).
11 Original contributions licensed under Apache License 2.0.
12 Upstream copyrights and license notices remain applicable.
13
14 “Pi” refers to the Pi agent harness; this notice covers our model adaptation and packaging, not the harness itself.
15
16 Publisher: bytkim (https://huggingface.co/bytkim).
17 The main Pi model weights differ from the original Qwen checkpoint through
18 Pi-oriented supervised fine-tuning and reinforcement learning. Release work
19 includes evaluation, serving documentation, and format-specific packaging.
20 The main Pi checkpoint has additionally been converted to GGUF and quantized into the supplied target-model variants. The Q8_0 and Q4_0 MTP companions are quantized versions of the upstream BF16 MTP head; that head is not a Pi-fine-tuned draft.
21 This attribution does not claim authorship of upstream weights, architectures,
22 training/inference tools, or third-party companion models.
23
24 DFlash2 companions
25 The dflash2/ checkpoint files are mirrored unchanged from Inco AI / Z Lab:
26 https://huggingface.co/incoai/Qwen3.8-27B-DFlash2-GGUF/tree/51962825493a48b846b40126d35c799ac4093ad0
27 These companions are not Pi-fine-tuned weights. Their license, original model
28 card, attribution, source revision, and checksums are retained in dflash2/LICENSE,
29 dflash2/UPSTREAM_README.md, dflash2/README.md, and dflash2/source.json.
30
31 This NOTICE records attribution and modifications; it does not change the
32 Apache License 2.0 terms or imply endorsement by upstream contributors.
33