Repository navigation
Gallery 1.1.2
v1.1.2(2026-08-21)
1.1.2 正式版新增 GPU 推理运行时自动准备:CCIP 角色模型与 NVIDIA CUDA ONNX Runtime 会在启动后后台自动下载,并按 CUDA → OpenVINO → CPU 顺序自动选择推理后端。
主要变化
- 字符识别默认
CHARACTER_RECOGNITION_PROVIDER=auto:启动后后台准备 CCIP 模型与 CUDA 运行时(下载固定哈希,可从官方源或国内镜像获取),按可用硬件自动选择后端。 - 默认允许 CPU 回退(
CHARACTER_ALLOW_CPU_FALLBACK=1);旧的CHARACTER_OPENVINO_ALLOW_CPU_FALLBACK变量仍兼容读取。 - 维护页新增“模型与推理”区域:查看 GPU 检测、Provider 选择、模型/CUDA 准备进度、切换下载源并重试失败任务;下载进行时页面保持短轮询。
- 支持离线安装:可手动放置模型与 CUDA 库,
ONNXRUNTIME_AUTO_DOWNLOAD=0关闭自动下载。 gpuProvider 别名仍指 OpenVINO(历史兼容),绝不映射 CUDA。
安装包
- 文件:
gallery_1.1.2_x86_64.fpk - SHA-256:
03389114d5c2ebcb680b6f513a612da0eb4b7011777a5884c6ce4f4cf48c3d85 - 格式:官方 fnpack 生成的 gzip tar FPK,目标架构
x86_64,服务端口8899
English
v1.1.2 (2026-08-21)
The 1.1.2 formal release adds automatic GPU runtime preparation: the CCIP character model and the NVIDIA CUDA ONNX Runtime are downloaded in the background after startup, and the inference backend is chosen automatically as CUDA → OpenVINO → CPU.
Changes
- Character recognition defaults to
CHARACTER_RECOGNITION_PROVIDER=auto: after startup the CCIP model and CUDA runtime are prepared in the background (pinned hashes; official or China mirror), then the best backend is selected from available hardware. - CPU fallback is allowed by default (
CHARACTER_ALLOW_CPU_FALLBACK=1); the legacyCHARACTER_OPENVINO_ALLOW_CPU_FALLBACKvariable is still honored. - The maintenance page adds a "模型与推理" (Models & Inference) section: GPU detection, provider selection, model/CUDA preparation progress, download-source switching, and retry for failed preparation; the page keeps short polling while a download is in progress.
- Offline installs are supported by pre-placing the model and CUDA libs;
ONNXRUNTIME_AUTO_DOWNLOAD=0disables automatic downloads. - The
gpuprovider alias still means OpenVINO (historical compatibility) and never CUDA.
Package
- File:
gallery_1.1.2_x86_64.fpk - SHA-256:
03389114d5c2ebcb680b6f513a612da0eb4b7011777a5884c6ce4f4cf48c3d85 - Format: official fnpack gzip-tar FPK,
x86_64, service port8899