Self proclaimed software engineer...
Rivers State, Nigeria
Joined November 2019
- Tweets485
- Following323
- Followers75
- Likes933
Excited to be one of the winners of the Gemma 4 Challenge on @ThePracticalDev 🎉
I wrote about how @googlegemma 4 uses soft tokens to efficiently process images, reducing thousands of image patches into a much smaller set of tokens for the LLM.
dev.to/eteimz/gemma-4-soft-t…
Gemma 4 by @googlegemma introduced effective models: E2B/E4B.
These models take advantage of Per-Layer Embeddings. Instead of having embeddings only in the input layer, they are shared across layers of the network.
Watch my video explaining it here: youtu.be/tOwaPCDKkbc?si=zu-T…
This entire claude code clean code saga is reminding of the scene from Halt and Catch Fire where they were reverse engineering IBM BIOS.
Eteims retweeted
LLM = CPU (data: tokens not bytes, dynamics: statistical and vague not deterministic and precise)
Agent = operating system kernel
I’ve been playing around with the idea of combining RAG with image generation. Thanks to the in-context learning capabilities of modern image generation models, this is possible.
I call it RAGE!!!
Read my article to learn more:
neurlcreators.substack.com/p…
Eteims retweeted
BERT is just a Single Text Diffusion Step! (1/n)
When I first read about language diffusion models, I was surprised to find that their training objective was just a generalization of masked language modeling (MLM), something we’ve been doing since BERT from 2018.
The first thought I had was, “can we finetune a BERT-like model to do text generation?”
I ran the same tests from the GPT-4o Image announcement blog on Nano Banana. Overall, the results were impressive. Its biggest strength is image editing, while its weakest point is text rendering.
Read my full breakdown 👉🏿 neurlcreators.substack.com/p…
My graphics skills are limited, yet I managed to outperform state-of-the-art models. Benchmarks aside, there’s still a long road ahead.
Had an amazing time at #OSCAFest25 by @osca_hq
Day 1 highlight → the mentorship session 🙌🏿
Day 2 highlight → giving my talk and sharing Whot Happened 🎤
Big congrats to the organizers, volunteers, and speakers. Already looking forward to the next #OSCAFest
A Story of Imperfection: Understanding Imperfect Information Games
Or
A Story of Imperfection: Decision-Making When the Pieces are Meeting
Day 1 of #OSCAFest25 by @osca_hq was 🔥🔥🔥
Day 2 is definitely going to be a bang 💥
I’ll be speaking today in the Orbital Room at 4:35 PM. The title of my talk is “Here’s Whot Happened”.
Whot isn’t just a card game. It’s a game of interaction. Most software implementations miss that, so I built an open-source version you can play over a phone call.
I’ll be speaking about it at @osca_hq 25 in my talk “Here’s Whot Happened.” See you there
#oscafest25
Here’s an idea: take Nigeria’s most popular card game, turn it into a web app, and let people play just by sharing a link.
And you know Whot? Make it open source.
Can you guess whot I’ll be speaking about at #oscafest25 by @oscahq.
Source code: github.com/EteimZ/whot/tree/…
