CoolFace
Modelpublic

openevo-recovery/openevo-h140-h141-qwen25-7b-webshop-artifacts

sourceHugging Faceotherupdated 24d agoView on Hugging Face
0likes
Model Card

OpenEVO Asset — openevo-h140-h141-qwen25-7b-webshop-artifacts

<!-- openevo-cross-platform-identity:begin -->

中文说明（默认）

这是 OpenEVO 跨平台实验资产的一部分。本段只补充统一身份与导航信息，不修改仓库已有模型、数据、checkpoint、trajectory 或历史科学语义。

  • —Provider ID: miyuki17/openevo-h140-h141-qwen25-7b-webshop-artifacts
  • —Role: LEGACY_UNRESOLVED
  • —Classification: legacy-unresolved
  • —Canonical: unresolved
  • —Visibility: public
  • —Cross-platform binding: legacy-unresolved
  • —Retention: PRESERVE_PENDING_RESOLUTION
  • —Scientific authority: pinned Git design/preregistration → run manifest/receipt/reconciliation；Hugging Face 是 durable artifact/provider layer。
  • —Governance: mykcs/openevo-experiment@bb57c88eb5de3036dee3ea9b095bc550dd6882c8
  • —Language standard: mykcs/openevo-experiment@a1994980f16a922374f25442a599031825b2cf03

命名和语言约定：repo、title、canonical ID 使用英文；README 默认中文。本“中文说明（默认）”是默认阅读入口，下方已有英文说明作为 English companion / historical detail 保留。若本段与 immutable publication receipt 或精确 remote revision/hash 冲突，以后者为准。

English: This block adds cross-platform identity metadata only. Existing artifact bytes and scientific claims are unchanged. <!-- openevo-cross-platform-identity:end -->

OpenEvo H1.40 / H1.41 WebShop adapter artifact archive / 适配器工件归档

导航 / Navigation：OpenEvo WebShop 公开科研产物

中文

这是 H1.40 / H1.41 的历史实验适配器归档，不是完整模型，也不代表实验已经证明 OpenEvo 有效。 仓库仅保存 24 组历史 SD-LoRA/PEFT adapter 及紧凑 provenance；Qwen2.5-7B-Instruct 基础权重、原始轨迹、replay JSONL、W&B、优化器/RNG 状态和数据集 payload 均不上传。 H1.41 的正式 reconciliation 标记为 MEASUREMENT_INVALID，因此这些权重只能作为可审计的历史实验资产使用。

English

This repository is a historical scientific artifact archive, not a standalone base model and not a claim of final OpenEvo efficacy. It publishes the 24 SD-LoRA/PEFT adapter states retained from H1.40 and H1.41, together with compact provenance and SHA256 manifests.

Identity and scope

  • —Base model: Qwen/Qwen2.5-7B-Instruct (obtain separately; base weights are not redistributed here).
  • —Base content identity recorded by the experiment: 0c32faf81d8f2f83cf79860d2a48111f166348d6dd38cd348576d8b3a0141c1e.
  • —H1.40 source run: phase-h140-g2-r2-20260820; 8 sequential G2 adapter states from formal-g2-training-retry12.
  • —H1.41 source run: phase-h141-magnitude-screen-20260821; 8 C0_exact_h140 states and 8 C1_reset_magnitude states.
  • —Total: 24 adapter directories, 48 safetensors binary files.
  • —Raw rollouts/episodes, replay JSONL, optimizer/RNG state, W&B state, datasets, OpenEvo source code, and unchanged Qwen base weights are excluded.

Scientific claim boundary

H1.40 recorded that G2 adapter training passed structurally, while acquisition, retention, preservation, T2, and multi-generation claims remained unevaluated.

H1.41 was an updater-mechanism magnitude screen, not a fresh-task transfer test. Its retained reconciliation marks the selector MEASUREMENT_INVALID; therefore these artifacts must not be interpreted as a successful H1.41 efficacy result.

The purpose of this repository is byte-preserving reproducibility and later audit.

Layout

text
adapters/
  h1.40/formal-g2-training-retry12/<sd-lora-worker>/
  h1.41/C0_exact_h140/<sd-lora-worker>/
  h1.41/C1_reset_magnitude/<sd-lora-worker>/
provenance/
  publication-manifest.json
  h1.40/...
  h1.41/...

Each adapter directory contains:

  • —adapter_config.json
  • —adapter_model.safetensors
  • —openevo_sd_lora_state.json
  • —openevo_sd_lora_state.safetensors

The top-level publication manifest records the original server-relative path, size, SHA256, and safetensors structural header check for every published adapter file.

Loading

A PEFT adapter can be loaded against the separately obtained pinned base model. The exact historical state is determined by the chosen adapter path and the hashes in provenance/publication-manifest.json.

Licensing and redistribution

The official Qwen2.5-7B-Instruct model page identifies the base model as Apache-2.0. Base weights are intentionally not duplicated here. This archive uses license: other because it is a mixed scientific artifact record and does not infer a license for OpenEvo source code; no OpenEvo source code is redistributed.

Verification policy

Publication is accepted only after every uploaded binary is read back from Hugging Face and its full-file SHA256 is recomputed and matched against the frozen local manifest. A verification receipt is added in a later additive commit; published binary bytes are never replaced under the same path.