Move DeepSeek R1 to LiteRT-LM and remove MediaPipe
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
调研方向
Start with the model URL and file type in lib/data/repositories/ai_model_repository_impl.dart, then check the DeepSeek token limit in lib/data/datasources/llm/local_model_params.dart. Update the listed FlutterGemma.initialize(...) call sites and integration tests, and remove the dependency from pubspec.yaml. Test download and responses on a real phone, including <think> rendering and memory use on a low-RAM device; confirm the old .task file is no longer needed.
由索引模型根据 Issue 内容生成。
描述
What
Move DeepSeek R1 to the LiteRT-LM engine and remove the MediaPipe engine from the app.
Why
- Right now MediaPipe is only in the app for one model: DeepSeek R1. Our catalog downloads it as a
.taskfile (deepseek_q8_ekv1280.task), which only MediaPipe can run. Qwen3 and Gemma 4 already use LiteRT-LM. - Google's own model card says the MediaPipe route is "in maintenance mode", so it won't get new fixes or speedups.
- A LiteRT-LM build of DeepSeek R1 now exists in the same Hugging Face repo:
DeepSeek-R1-Distill-Qwen-1.5B_multi-prefill-seq_q8_ekv4096.litertlm - One engine instead of two means less to maintain, a smaller app, and DeepSeek gets the same GPU handling and fixes as the other models.
Not a drop-in swap
- The new file has a bigger context (4096 vs our 1280), so it uses more memory. Needs checking on a low-RAM phone.
- People who already downloaded the old
.taskfile would need to download the new one. The old file becomes unused (the storage cleanup can remove it).
What to change
lib/data/repositories/ai_model_repository_impl.dart— point DeepSeek R1 at the.litertlmURL, file typelitertlmlib/data/datasources/llm/local_model_params.dart— DeepSeekmaxTokens1280 → 4096- Remove
MediaPipeEngine()fromFlutterGemma.initialize(...)inlib/main.dart,lib/features/shiv/gana/engine/gana_workmanager.dart, and the integration tests - Remove
flutter_gemma_mediapipefrompubspec.yaml
How to test (on a real phone)
- DeepSeek R1 downloads, loads, and answers
<think>reasoning blocks still show correctly- Memory use and speed compared with the old file
- Works on a low-RAM phone without being killed
Depends on #216 (flutter_gemma 1.8.3).
- 主要语言
- Dart
- 星标
- 2
- 派生
- 2
- 平均合并
- 1 天 7 小时
- 30 天内合并 PR
- 35
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
basictech01/uniun 的其他 Issue
-
bug good first issue
难度 2/5 1-3 小时 新手友好度 82/100
basictech01/uniun#300 ·
维护者通常 1 天内回复
-
enhancement good first issue P2-Medium
难度 5/5 一周以上 新手友好度 25/100
basictech01/uniun#296 ·
维护者通常 1 天内回复
-
bug good first issue P2-Medium
难度 4/5 1-2 天 新手友好度 22/100
basictech01/uniun#293 ·
维护者通常 1 天内回复
-
bug P1-High
难度 4/5 3-5 天 新手友好度 18/100
basictech01/uniun#286 · 2 条评论 ·
维护者通常 1 天内回复
-
P3-Low
难度 4/5 1-2 天 新手友好度 30/100
basictech01/uniun#282 · 1 条评论 ·
维护者通常 1 天内回复
查看 basictech01/uniun 的全部 Issue
相似的 Issue
-
bug
难度 1/5 1 小时以内 新手友好度 65/100
chriskelly/wish-together#36 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
AppFlowy-IO/AppFlowy#9075 ·
-
Paste asks for iOS paste permission because the default context menu ignores SystemContextMenu可能已有人在做 @dorklein 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 72/100
maheshj01/searchfield#267 ·
-
难度 2/5 1-3 小时 新手友好度 67/100
-
难度 2/5 1-3 小时 新手友好度 72/100
avdept/JellyBoxPlayer#375 ·
维护者通常 1 天内回复