Sunday, October 5, 2025

Large Language Models explained briefly

 Great video which explains LLM simply, by 3Blue1Brown youtube channel which explains maths using animation.atin math



Wednesday, September 24, 2025

Nvidia Jetson ORIN NANO AI/ML Developer Kit

 I finally got my order of the long backordered - Nvidia Jetson ORIN NANO AI/ML Developer Kit. I will be testing and reporting on AI/ML/LLM benchmarks and projects made using this.

I still own the Jetson Nano kit as well.

The NVIDIA Jetson Orin Nano is an edge AI computer available in an 8GB and a 4GB module, featuring an NVIDIA Ampere architecture GPU with 1024 CUDA cores (8GB version) and a 6-core Arm® Cortex®-A78AE CPU. It offers up to 67 INT8 TOPS of AI performance (8GB module), is equipped with 8GB (or 4GB) of LPDDR5 memory, and supports external storage via microSD and NVMe. The device is designed for edge AI applications, offering a balance of performance, power efficiency (7W-15W for the 8GB module), and a compact form factor. 

Here's a breakdown of the key specs for the 8GB Jetson Orin Nano module:

GPU 

Architecture: NVIDIA Ampere

CUDA Cores: 1024

Tensor Cores: 32

AI Performance: Up to 40 INT8 TOPS (dense) / 67 INT8 TOPS (sparse)

CPU

Cores: 6-core Arm® Cortex®-A78AE v8.2 64-bit 

Cache: 1.5MB L2 + 4MB L3 

Clock Speed: Up to 1.5 GHz (original) or 1.7 GHz (Super version) 

Memory 

Capacity: 8GB

Type: 128-bit LPDDR5

Bandwidth: 68 GB/s (original) or 102 GB/s (Super version)

Storage 

Supports microSD card for the base OS

Supports external NVMe SSDs via M.2 Key M slot

Power 

Power Modes: 7W, 15W (and a new 25W "Super" mode)

Key Features

Size: Compact, ~69.6mm x 45mm 

Connectivity: USB ports, GbE, HDMI/DP output, and multiple I/O options 

Applications: Ideal for edge AI applications, robotics, smart vision systems, and more 

Picture:







Picture - Bootup, and firmware updating. I used a 128Gb MicroSD card and downloaded the latest firmware from this link - click here.







Unboxing video:




Saturday, August 30, 2025

Exploring Googles Nano Banana AI - Gemini 2.5 flash preview

 Below are images from my prompts, while exploring Googles Nano Banana AI - gemini 2.5 flash image preview. The prompt ideas were from an Instagram post l saw, l will post the link below.

I took a picture of myself, and entered the first prompt below.









Result









Not bad. Then:









Below took 161.1 seconds and the system crashed.







I reentered the words, this time in quotes. Enugu is the name of the city where l was born.

Result in 227.1 seconds









Then l tried a merge: The image generated wasn't impressive as per the positioning of the can. 









I gave my feedback to Gemini/google system on this and decided to rerun the prompt.

In 98.6 seconds l got:









Which looks much more better! 


Friday, August 1, 2025

Hugging Face - Upgrade to Xet

I received an informational email on - Hugging face upgrading the Hub's storage backend from Git LFS to Xet. It stated that "Xet is our chunk-based backend that already powers 50% of all Hub downloads and serves the Llama, Qwen, Gemma, and Phi model families."

Xet Documentation link:  https://huggingface.co/docs/hub/en/storage-backends#using-xet-storage

Command given:

pip install -U huggingface_hub

Output:



Meet Sohu, the fastest AI chip of all time

 It will be interesting to see how far progress is made on this SOHU chip.




Friday, July 25, 2025

Release Notes: Gemini's multimodality

 


"Ani Baddepudi, Gemini Model Behavior Product Lead, joins host Logan Kilpatrick for a deep dive into Gemini's multimodal capabilities. Their conversation explores why Gemini was built as a natively multimodal model from day one, the future of proactive AI assistants, and how we are moving towards a world where "everything is vision." Learn about the differences between video and image understanding and token representations, higher FPS video sampling, and more."

Large Language Models explained briefly

 Great video which explains LLM simply, by 3Blue1Brown youtube channel which explains maths using animation. atin math