Go to file

zhousha efee86e49c submit 24 models on zhoushasha: ppu_zw_810e(18)/hygon_k100-ai(4)/MetaX_c-500(1)/Biren_166m(1)

Candidate pool is the 24711 already-downloaded models, which satisfies
mechanism A's precondition that the model be present in platform storage.

Strict and relaxed condition-2 give identical results this round: of the 24007
pooled models with verify records, 24006 already have 已验证 on some card, so
relaxing to include 验证中 adds just one. The binding constraint is now how few
models lack a record on each card - Kunlunxin_p-800, Cambricon_mlu-370-x8 and
Iluvatar_bi-150 are all at zero and are left out of GPU_JOBS.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

2026-09-20 18:27:33 +08:00

.gitignore

update model list

2026-07-16 16:58:36 +08:00

Dockerfile

uodate

2026-06-22 19:00:46 +08:00

main.py

submit 24 models on zhoushasha: ppu_zw_810e(18)/hygon_k100-ai(4)/MetaX_c-500(1)/Biren_166m(1)

2026-09-20 18:27:33 +08:00

README.md

fix: add env vars to Dockerfile

2026-06-12 21:02:56 +08:00

requirements.txt

add platform strategy

2026-06-12 14:24:27 +08:00

README.md

xc_validation_strategy

批量向 ModelHub XC 平台提交模型验证任务的策略服务，之后保持 HTTP 服务存活供平台探活。

功能

自动登录 ModelHub 获取 Token
批量提交模型验证任务（vLLM 框架，Cambricon MLU-370-x8）
提交结果写入 submitted_validation_tasks.txt
暴露 /health 和 /status 接口满足平台运行时契约

项目结构

.
├── main.py            # 主入口：HTTP 服务 + 提交逻辑
├── Dockerfile         # 平台镜像构建配置
├── requirements.txt   # Python 依赖
└── submitted_validation_tasks.txt   # 运行后自动生成，记录提交结果

平台契约说明

本项目满足平台对策略镜像的全部必要约束：

Dockerfile 位于仓库根目录，基于官方轻量基础镜像
暴露 8080 端口并实现 GET /health
通过环境变量 STRATEGY_ID 获取策略 ID
正确处理 SIGTERM 信号，支持优雅停机

README.md Unescape Escape

xc_validation_strategy

功能

项目结构

平台契约说明

README.md