Usama Ahmed

👨‍💻

[Usama Ahmed @0ssamaak0]

@oyacaro @Hesamation @huggingface @danielhanchen you can go for the 9B version, from the same creator. Downloading it rn

[madroid @Madroidmaq]

🎉 mlx-omni-server v0.4.0 is out!

✨ New /v1/embeddings service (via mlx-embeddings, thx @0ssamaak0) - generate embeddings easily!
🔊 More TTS models (kokoro, bark & more!) with mlx-audio (thx @ zboyles) !
🧠 mlx-lm upgraded for models like qwen3!

https://t.co/iGK9vfbRuK

[Usama Ahmed @0ssamaak0]

I created an open source mac app that mocks the usage of OpenAI API by routing the messages to chatgpt desktop app so it can be used without API key

You can simply change the api base (like if u are using ollama) and select any of the models that u can access from chatgpt app

[Usama Ahmed @0ssamaak0]

I will merge this into main very soon after I fix some issues!

[Usama Ahmed @0ssamaak0]

I have been exploring @cohere Multimodal Embed 3 last week. It's like magic! 🪄

I was able to replace the whole pipeline of CLIPPyxX (CLIP + OCR + Text embedding model) with just Embed 3.
fp

[Usama Ahmed @0ssamaak0]

I created the index once and now I'm can do image search very easily. Before I had to create 2 indices (1 for image embeddings and 1 for text embeddings) and search using 2 routes.

The results are amazing You can try it now using Cohere-embed branch https://t.co/hJPgxb6e5H

[Usama Ahmed @0ssamaak0]

@cedriceberhardt @ollama It’s connecting to your computer that is running the LLMs. No battery drain on the phone itself

[Usama Ahmed @0ssamaak0]

I integrated locally running mistral with Siri, far better than standard Siri. no more "Here's what I found"

Thanks to @ollama for making this possible!
GitHub: https://t.co/ILydRonUGT

[Usama Ahmed @0ssamaak0]

Siri LLama now can answer your questions from a webpage!
Running locally with open source models from @ollama

[Usama Ahmed @0ssamaak0]

I updated Siri LLama so it can run LLaVA on any image + support any model on both @ollama and @FireworksAI_HQ AI. who wants an ai pin now?

[Usama Ahmed @0ssamaak0]

I integrated locally running mistral with Siri, far better than standard Siri. no more "Here's what I found"

Thanks to @ollama for making this possible!
GitHub: https://t.co/ILydRonUGT

[Usama Ahmed @0ssamaak0]

Once CLIPPyX server is running, you can access it from any UI. I made a simple html page and Flow launcher plugin. and @Microsoft Powertoys run plugin will be ready soon!

[Usama Ahmed @0ssamaak0]

CLIPPyX: The new AI powered image search tool

🔵Search by Image Caption
🟡Search by Textual Content in Images
🔵Search by Image Similarity

Video at 1x speed on my laptop (1660 Ti / 16 GB RAM)
GitHub: https://t.co/hJPgxb6e5H

[Usama Ahmed @0ssamaak0]

Tool Overview
Models can be used directly from🤗Transformers, or more efficient options like @Apple Mobile CLIP and GGUF text embedding models through llama.cpp

[Usama Ahmed @0ssamaak0]

@Aminmoe3 فيكوا شبه من بعض

[Usama Ahmed @0ssamaak0]

@Mo7amed_Ma3rouf @MatthewBerman @GroqInc @pinecone @mckaywrigley It's exactly the same I think, he just named the shortcut "Let's Chat" while I called it Siri LLama

[Usama Ahmed @0ssamaak0]

@Mo7amed_Ma3rouf @MatthewBerman @GroqInc @pinecone @mckaywrigley you can get any text input from Siri directly using iPhone dictation.
That's what I've done here: https://t.co/QZXzwuKedC

[Usama Ahmed @0ssamaak0]

I integrated locally running mistral with Siri, far better than standard Siri. no more "Here's what I found"

Thanks to @ollama for making this possible!
GitHub: https://t.co/ILydRonUGT

[Usama Ahmed @0ssamaak0]

@Mo7amed_Ma3rouf @MatthewBerman @GroqInc @pinecone ... and there's no straight forward way to interface with them (as far as I know)

[Usama Ahmed @0ssamaak0]

@Mo7amed_Ma3rouf @MatthewBerman @GroqInc @pinecone I think it's not just about calling the LLM itself. there're other things like memory, input image, and adding features like RAG. requires a python app.

On-device model is a very good idea (I think apple is going to do it next June) however now, model options are limited.....

[Usama Ahmed @0ssamaak0]

@MatthewBerman @GroqInc @pinecone There’s no need for voice model imo since you can use Siri itself

[Usama Ahmed @0ssamaak0]

@MatthewBerman @GroqInc @pinecone I have already tried similar approach (no RAG yet) but it supports multimodality too

Check it out: https://t.co/ILydRonUGT

[Usama Ahmed @0ssamaak0]

@rezk1222 في حاجات في أساسيات الـCS معمولة بشكل كويس جدا، المشكلة إنه بدأ يحاول يعمل دور documentation بديل لـlibraries أو frameworks، ودا شيء شديد الغباء بصراحة