Skip to content

Router reconstruct - #237

Open
yu-zhy wants to merge 2 commits into
itmisx:mainfrom
yu-zhy:RouterReconstruct
Open

Router reconstruct#237
yu-zhy wants to merge 2 commits into
itmisx:mainfrom
yu-zhy:RouterReconstruct

Conversation

@yu-zhy

@yu-zhy yu-zhy commented Aug 7, 2026

Copy link
Copy Markdown
   1、建议重构模型路由策略实现,支持用户自定义模型路由语义元句
          ~/.deepx/router/patterns/   追加4个级别的语义元句定义 deep.toml   complex.toml  medium.toml  simple.toml
   2、支持模型路由回降机制,进一步降低成本
   3、进一步细化模型路由策略

yu-zhy added 2 commits August 6, 2026 08:42
- segmenter.yaml 独立配置文件(不污染 model.yaml)
- embedder: onnx 启动语义模型  默认使用bge-small-zh-v1.5  支持自动下载
- topic_tracking: true 启动主题追踪
- onnx_model: xxx  支持手动配置下载模型

add TopicGraph feature with ollama

fix: PR itmisx#219 issue 全部修复

1. modelPin 判断: m.modelPin == "" || m.modelPin == "auto" (原恒为 false)
2. activeModelID 兜底: else 分支加回退逻辑
3. CI: go vet self-assignment 删除 + TF-IDF 首篇文档向量全零修复
4. 话题切换冲突: 移除 TopicSwitch 提示(保留 TopicGraph 用于路由)
5. ONNX 路径: ortLibName build tag 分平台 + ORT 引擎路径改为 ~/.deepx/ocr/
6. ONNX mean pooling: 按 attention_mask 加权平均 + 归一化
7. 阈值分档: adjustedNewTopicThreshold() 按 embedder 类型返回不同值
8. token_type_ids: 补充缺失的 ONNX 模型输入(全零张量)
   1、建议重构模型路由策略实现,支持用户自定义模型路由语义元句
   2、支持模型路由回降机制,进一步降低成本

	Fixs itmisx#236
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant