Downloads · 30 days
0
openevo-recovery/openevo-ceiling1-qwen25-7b-historical-final-adapters
openevo-ceiling1-qwen25-7b-historical-final-adapters is a machine learning model from openevo-recovery. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for peft. The card lists the license as apache-2.0.
这是 OpenEVO Ceiling-1.0 早期 Qwen2.5-7B historical long-run 的终局参数包。它不是当前 Q17 / Bounded 1.7B 主线,也不是把完整 69.9GB 历史 checkpoint 仓库公开。
Downloads · 30 days
0
Access
Public
Updated Sep 28, 2026
Repo size
2.9 GB
Likes
0
Public
Click a slice to open those files.
.safetensors2.9 GB · 100%
From the Hugging Face model README
这是 OpenEVO Ceiling-1.0 早期 Qwen2.5-7B historical long-run 的终局参数包。它不是当前 Q17 / Bounded 1.7B 主线,也不是把完整 69.9GB 历史 checkpoint 仓库公开。
只包含两段 adapter:
opsd/:7B OPSD final adapter;stage2-final/:Stage2 terminal adapter(round 148 / generation 142 / 143 components)。基础模型直接引用公开的 Qwen/Qwen2.5-7B-Instruct@a09a35458c702b33eeacc393d103063234e8bc28。历史本地 content pin 与该 revision 的 9 个 required files 已逐 SHA256 对齐。
Stage2 当时不是直接从原始 Qwen2.5-7B 开始,而是先把 OPSD adapter 合入 base,再在这个 merged base 上加载累计 Stage2 adapter。因此复现顺序必须是:
Qwen2.5-7B-Instruct
-> load + merge OPSD adapter
-> load Stage2 final adapter
完整 merged-base 权重约 14GB,不需要为了这个 release 再公开一份。
0f1a3f3ad5dc8a50c2156f6416944b951ccb17c87439b9505b9233847f5a7b63SEED 的 89.7 / 78.1% 只作为 paper-reported reference;SEED checkpoint/denominator 没有在本项目里本地重现,因此不能把数值差写成 paired causal comparison。
| File | Bytes | SHA256 |
|---|---|---|
opsd/adapter_model.safetensors | 20,214,760 | f226725cad89f3f138423398168ce9e07d59e86a740a96e946ba1286f8d02190 |
stage2-final/adapter_model.safetensors | 2,886,498,904 | b0368f8c0dba79afa94e9e513c7b4f0023d1ebb95ab0d64adeca70734886199a |
两份 config 和 safetensors structured header 已做 bounded secret/internal-path scan;Hugging Face provider security 字段当时仍为 unscanned,所以这里不声称平台 malware scan PASS。
PUBLIC 包中的 OPSD adapter_config.json 会把 base_model_name_or_path 规范化为公开的 Qwen/Qwen2.5-7B-Instruct;历史原始 config 的精确 SHA 仍保留在内部 provenance,不随 PUBLIC 包发布内部运行时路径。
本 adapter package 以 Apache-2.0 发布。基础模型不随本仓库重新分发;复现时请使用 Qwen/Qwen2.5-7B-Instruct@a09a35458c702b33eeacc393d103063234e8bc28。来源与再分发说明见 NOTICE,完整许可文本见 LICENSE。
This package contains the terminal adapters from an early Qwen2.5-7B historical long-run in OpenEVO Ceiling-1.0. It is not the current Q17 / Bounded 1.7B line, and it does not publish the full 69.9GB historical checkpoint warehouse.
Only two adapters are included:
opsd/ — the final 7B OPSD adapter;stage2-final/ — the terminal Stage2 adapter (round 148 / generation 142 / 143 components).The base model is referenced directly as Qwen/Qwen2.5-7B-Instruct@a09a35458c702b33eeacc393d103063234e8bc28. The historical local content pin was matched file-by-file by SHA256 against the nine required files of that revision.
Stage2 did not start directly from the original Qwen2.5-7B base. The OPSD adapter was first loaded and merged into the base, and the accumulated Stage2 adapter was then loaded on top of that merged base. The reconstruction order is therefore:
Qwen2.5-7B-Instruct
-> load + merge OPSD adapter
-> load Stage2 final adapter
The complete merged-base weights are about 14GB and are not duplicated in this release.
0f1a3f3ad5dc8a50c2156f6416944b951ccb17c87439b9505b9233847f5a7b63SEED's 89.7 / 78.1% values are included only as a paper-reported reference. The SEED checkpoint/denominator was not reproduced locally in this project, so the numerical difference must not be presented as a paired causal comparison.
| File | Bytes | SHA256 |
|---|---|---|
opsd/adapter_model.safetensors | 20,214,760 | f226725cad89f3f138423398168ce9e07d59e86a740a96e946ba1286f8d02190 |
stage2-final/adapter_model.safetensors | 2,886,498,904 | b0368f8c0dba79afa94e9e513c7b4f0023d1ebb95ab0d64adeca70734886199a |
The two configs and the structured safetensors headers were checked with a bounded secret/internal-path scan. At that time the Hugging Face provider security field was still unscanned, so this card does not claim a platform malware-scan PASS.
The PUBLIC OPSD adapter_config.json normalizes base_model_name_or_path to the public Qwen/Qwen2.5-7B-Instruct. The exact SHA of the historical raw config remains in internal provenance and internal runtime paths are not released.
This adapter package is released under Apache-2.0. The base model is not redistributed here; use Qwen/Qwen2.5-7B-Instruct@a09a35458c702b33eeacc393d103063234e8bc28 for reconstruction. See NOTICE for source and redistribution notes and LICENSE for the full license text.