Test Gemma 4 -gpu model files before switching the catalog
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 3/5
- 预计耗时
- 1-2 天
- 新手友好度
- 55/100
- Issue 类型
- 功能
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- dart
调研方向
Start with #216 and the catalog entry in lib/data/repositories/ai_model_repository_impl.dart. On a real phone, compare the generic and -gpu Gemma 4 E2B files on GPU, then force CPU with the -gpu file; repeat for E4B if E2B looks promising. Done means the load, prefill, decode, and memory comparisons support a clear decision about whether to switch, while confirming whether CPU fallback still works.
由索引模型根据 Issue 内容生成。
描述
What
Test whether the special GPU model files for Gemma 4 are better for us, before changing the catalog.
Why
Both Gemma 4 repos on Hugging Face ship two versions of the model:
| Model | What we download today | Also available |
|---|---|---|
| Gemma 4 E2B | gemma-4-E2B-it.litertlm |
gemma-4-E2B-it-gpu.litertlm |
| Gemma 4 E4B | gemma-4-E4B-it.litertlm |
gemma-4-E4B-it-gpu.litertlm |
Now that the app prefers the GPU on Android (#216), the -gpu file might load faster or run faster.
But the model card doesn't explain what's different, so we should not switch blindly.
Risk to check
The -gpu file may be GPU-only. If it is, our CPU fallback would break — phones where the GPU fails would not be able to run Gemma 4 at all. This has to be tested before switching.
How to test (on a real phone)
- Download both files for Gemma 4 E2B.
- Run the same prompt with each, on GPU, and compare from the log:
- load time
- prefill (reading the prompt) time
- decode (writing the answer) speed
- memory use
- Force CPU with the
-gpufile and check it still loads. If it doesn't, we can't use it as the only download. - Repeat for E4B if E2B looks good.
Decide
- Faster and still works on CPU → switch the catalog URL in
lib/data/repositories/ai_model_repository_impl.dart - Faster but GPU-only → keep the generic file (or download per device, which is a bigger change)
- No real difference → keep the generic file and close this
Depends on #216.
- 主要语言
- Dart
- 星标
- 2
- 派生
- 2
- 平均合并
- 1 天 35 分钟
- 30 天内合并 PR
- 36
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
-
bug good first issue
难度 3/5 1-2 天 新手友好度 52/100
basictech01/uniun#301 ·
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
查看 basictech01/uniun 的全部 Issue
相似的 Issue
-
good-first-issue P3
难度 2/5 1-3 小时 新手友好度 88/100
flutter/dart-intellij-third-party#758 ·
维护者通常 1 天内回复
-
Bug: Incorrect location coordinates on logged Sound Meter data可能已有人在做 @Jashan32 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 72/100
维护者通常 1 天内回复
-
area/archive bug
难度 2/5 1-3 小时 新手友好度 76/100
DarkFlippers/qUnleashed#374 ·
维护者通常 1 天内回复
-
agent:in-progress agent:ready feature P2 · 中 区域: 悬浮窗
难度 2/5 1-3 小时 新手友好度 85/100
ASDW51/DanmuFloat#178 ·
维护者通常 1 天内回复
-
Missing workouts未关闭enhancement
难度 2/5 1-3 小时 新手友好度 67/100
维护者通常 5 天内回复