CoolFace
Modelpublic

openevo-recovery/openevo-adapter-archive

sourceHugging Faceotherupdated 23d agoView on Hugging Face
0likes
Model Card

OpenEVO Asset — openevo-adapter-archive

<!-- 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-adapter-archive
  • —Role: PUBLIC_CANONICAL_ADAPTER_ARCHIVE
  • —Classification: historical-and-frozen-adapters
  • —Canonical: true
  • —Visibility: public
  • —Cross-platform binding: registry-resolved
  • —Retention: KEEP_STABLE_ID
  • —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 adapter archive / OpenEvo 适配器存档

导航 / Navigation：OpenEvo WebShop 公开科研产物

中文

这是 OpenEvo/WebShop 研究产生的公开、去重 LoRA/SD-LoRA 适配器存档，目的是在服务器清理后仍能定位、校验和恢复有价值的实验资产。它不是基础模型镜像，也不表示仓库里的每个适配器都改善了实验结果。

仓库包含：

  • —各实验的 PEFT adapter、训练状态和来源说明；
  • —`release_manifest.json` 中较早批次的文件清单与 SHA-256；
  • —`h138a/SHA256SUMS` 中 H1.38A 三组适配器、状态权重和 replay 的校验值；
  • —H1.38A 的冻结面板 manifest、训练证据和 combined reconciliation；
  • —h138b/：H1.38B method-control 的冻结 ordinary-LoRA 最终适配器、H1.39 独立加载/复现凭证，以及指向已单独发布 frozen SD-LoRA 的去重指针；不重复上传 SD 权重。
  • —每个 H1.38A 适配器目录里的中英双语实验卡。

仓库不包含 Qwen 基础权重、完整 Docker 镜像、W&B 指标库、原始 WebShop 数据、完整运行目录或任何凭据。H1.36 的精简发布另见 `openevo-h136-qwen25-7b-webshop-lora`。

恢复时先固定本仓库 revision 并校验 SHA-256，再按相邻状态文件定位基础模型内容身份、代码来源和实验 manifest。状态里的 model_revision 是当时本地验证的内容身份；本次归档没有把它提升为 Hugging Face 上游 commit。许可边界见 `LICENSES.md`。

English

This is a public, deduplicated archive of LoRA/SD-LoRA adapters produced by OpenEvo/WebShop research. Its purpose is to keep useful experiment artifacts locatable, verifiable, and restorable after shared-server cleanup. It is not a base-model mirror, and inclusion does not imply that an adapter improved the scientific result.

The repository contains:

  • —PEFT adapters, training state, and provenance for individual experiments;
  • —older-batch file identities and SHA-256 values in `release_manifest.json`;
  • —H1.38A adapter, state-weight, and replay checksums in `h138a/SHA256SUMS`;
  • —frozen-panel manifests, training evidence, and combined reconciliation for H1.38A;
  • —h138b/: frozen H1.38B method-control ordinary-LoRA final adapter, H1.39 independent load/replication evidence, and a deduplicating pointer to the separately published frozen SD-LoRA; the SD weights are not duplicated here.
  • —bilingual experiment cards in each H1.38A adapter directory.

It does not contain Qwen base weights, complete Docker images, the W&B metrics store, raw WebShop data, complete run directories, or credentials. The compact H1.36 release is available separately at `openevo-h136-qwen25-7b-webshop-lora`.

For restoration, pin this repository revision and verify SHA-256 values before following the adjacent state file to the base-model content identity, source code, and experiment manifest. The recorded model_revision is the locally verified content identity used by the run; this archive does not represent it as an upstream Hugging Face commit. See `LICENSES.md` for licensing boundaries.