Ivan Fioravanti ᯅ Profile picture
Technology Advisor, Founder & Board Member of @CoreViewHQ GenAI/LLM addicted, Apple MLX, Microsoft 365, Azure, Kubernetes, builder of my personal dreams.
Mar 15, 2025 7 tweets 2 min read
1/🧵 This benchmark M2 Ultra vs M3 Ultra is wrong!
This screenshots is taken from the Matt Tech Talks YouTube Channel. In this thread I'll repeat same steps on my machines to show that M2 Ultra results are just too low in the video. (why????) Image 2/🧵 I have used MLX and GGUF version of DeepSeek R1 Distill Qwen 7B in 4bit, using LM Studio.
Mar 9, 2025 4 tweets 2 min read
1/🧵 The Power of Manus AI? Here it is!
I asked "a thorough analysis of Microsoft stock".
In the first turn it created many detailed files I then asked for a website publicly accessible to share and analyze results and here it is! Replay and link below. Image 2/🧵 Link with final result. …9qn0l4a7av0vp-226a4af3.manus.computer
Jan 22, 2025 6 tweets 2 min read
👀 DeepSeek R1 (right) crushed o1-pro (left) 👀

Prompt: "write a python script for a bouncing yellow ball within a square, make sure to handle collision detection properly. make the square slowly rotate. implement it in python. make sure ball stays within the square" Yes, collision detection can be improved, but it's night and day the difference and moreover, one is 200$ a month and the other is nearly free.

I will try with local version 💪
Jan 5, 2025 6 tweets 2 min read
Guess the wine! 🍷
A simple classification test on multiple LLMs (open, closed, local, online) inspired by an OpenAI cookbook .

Let me share code now and I will write an article later to give more details and results. Here a preview.

1/🧵 Image Repo on GitHub with MIT license, use it as you want.
Goal was just trying the various LLMs API with structured output with a classification problem that from User Reviews must discover variety of the wine.


2/🧵github.com/ivanfioravanti…
Dec 28, 2024 7 tweets 2 min read
Phidata Multimodal AI Agent on Hugging Face Spaces!
It's my first attempt to create an app on HF and it was easy and smooth!

Gemini 2.0 Flash analysis of the video at the end is just INCREDIBLE 🤩

Some parts are speed up up to 40x

1/🧵 2/🧵
You can find it here you can just duplicate the space in your private one and add a gemini api key to secretshuggingface.co/spaces/ivanfio…
Nov 30, 2024 8 tweets 1 min read
Ollama prompt engineering interactive tutorial using Qwen 2.5 14B in progress! 3 chapters ready and more to come 🔥

This was inspired by Haiku tutorial by Anthropic, but Qwen model is simply to powerful (in math and logic) and I had to change few examples here and there. Image Repo here: github.com/ivanfioravanti…