This is OneKey Neo. Security, made personal.
Your Crypto. Your Style.
Take a closer look 👇🧵
onekey.so/products/onekey-ne…
Frad retweeted
苦 Calibre 久矣,开源一个它的替代品 Dalibre dalibre.randynamic.org/
Frad retweeted
Replying to @FradSer
折腾豆包模型的同好握手。Mac 端想要开箱即用的,自荐我做的 Setem:菜单栏语音输入,云端接了豆包/微信/千问三家,另带本机离线模型,断网可跑。按住快捷键说话、浮窗预览后上屏,Swift 原生,6.1MB,MIT 开源:github.com/zcyoung2000/Setem
Frad retweeted
Apple's Core AI models repo is now accepting third-party contributions 🥳
If you have developed some cool ports, this is the best place to share your recipes with the community. Looking at you @JackdeS11 @mattt @kevinqz 😂
github.com/apple/coreai-mode…
Frad retweeted
Raven 的这个可观测性的效果,太棒了!
Raven’s 🐦⬛ observability system just got a full rebuild.
When an agent team runs a task, the final answer only tells part of the story.
Now you can follow the execution timeline, drilling down from Agent → Turn → Step to see:
• Which agent is working
• What happened at each step
• Which tools were called
• Where errors occurred
A clearer view of how your agents work together, and where to start debugging when they don’t.
github.com/EverMind-AI/Raven
Frad retweeted
火炬之光2 Testflight 的邀请链接来了,1000 个名额。
这个链接同时支持 iPhone iPad Apple TV Vision Pro 和 Mac,不过 Mac 的审核卡了几天没反应了,所以 Mac 暂时还是很老的版本,不支持云存档和联网。
testflight.apple.com/join/fH…
需要优化的小问题应该还不少,欢迎大家反馈,我会尽快更新。
Frad retweeted
antimattr (YC & WTFund backed) is building earbuds + a smart ring so you can control AI agents, dictate, and capture thoughts by voice.
pre book now!
谷歌EmbeddingGemma 2的第三个“歪用”用例,只花了78美元训练费用,就做出了越南语ASR的SOTA级模型
3000小时越南语音频训练,8张H100 约3小时
CPU推理速度达实时语音的200倍,支持越英混合口语
大量被通用模型照顾不到的地方语言/口音,可能会迎来一波低成本专用模型
8个越南语基准平均WER为6.87%居首,略超过Gemini-Flash-3.8的6.92%、Cohere Transcribe的7.01%、ElevenLabs Scribe v2的7.08%
网友复用EmbeddingGemma 2的预训练音频编码能力,仅微调12层音频编码器的顶部2层,再训练一个6层 Transformer读取器,通过CTC解码将音频特征映射为文字
这种设计无需自回归逐Token生成,有利于提高转写吞吐量;Reader采用动态音频分块训练,还支持流式识别
以前要为一个小语种或一个地方方言训练高性能STT,数据和算力都是门槛,现在只微调部分编码层,再接上Reader+CTC,就做到了与Gemini同级的识别精度
这条路线如果能跨语言复现,可能会影响大量小语种、方言等细分语音识别市场的竞争格局
来自 @stablequan
#STT #ASR #EmbeddingGemma2
谷歌EmbeddingGemma 2的第二个“歪用”案例,装进macOS 菜单栏,做成本地语义搜索工具
统一嵌入空间会不断长出你没预见过的新奇用法,上次是STT,这次是多模态语义搜索
相当于语义版Spotlight,用同一个EmbeddingGemma 2 统一空间,把你Mac里所有模态的文件变成一个可以用人话搜索的本地数据库
过去,你需要记住文件名、目录或者关键词;现在,只需要描述你想找什么,就能从照片、视频、音频、文档或代码中找到对应内容
并且不依赖云端大模型,文件索引和搜索在本地完成
EmbeddingGemma 2本身支持多语言,这意味着你不光能用中文描述搜中文文件,还可以跨语言搜
搜索颗粒度能定位到视频里的某个时刻、播客的某一分钟、PDF的某一页
技术上,作者通过llama.cpp+Metal在Mac GPU上本地运行EmbeddingGemma 2,将文本、图像、音频、视频映射到统一的768维向量空间
索引时按内容类型拆分,PDF按页、音频按重叠的30秒窗口、视频按画面和音轨、代码按语义片段建立索引,存储在本地SQLite中
搜索时采用向量语义检索+关键词匹配的混合方案,既能理解“一个人在海边遛狗”,也能精确匹配发票编号、错误代码等字符串
项目来自 @_ARahim_
#EmbeddingGemma2 #MAC本地语义搜索 #DigUp
Frad retweeted
Excited to launch Transitions Builder and Community.
All these components were generated with a single prompt with the Transition UI agent, default setup, no designs provided.
The agent creates on the library’s knowledge, UI craft principles, and libraries-dev libraries, with a focus on UI motion and the details that make an interface feel polished.





























