d.amuna

and Blog:

[BBC Sport @BBCSport]

Wayne Rooney - A man of his word!

[d.amuna @dna4ty]

KenLM cut WER by 16.76% overall in our evaluation. Dagbani reached 14.56%, Akan 36.43%, and Ga 13.12%.
There’s still plenty to improve, especially for Akan, but the model is open, local, and ready for people to test and build on.

Full technical breakdown: https://t.co/FFhZt0L9Z0

[d.amuna @dna4ty]

Been working on this for a while, and it’s finally out: Griot Nano 1 🎙️

It’s a 153M open-source speech recognition model for Akan, Dagbani, Ewe, Ghanaian English, and Ga.

It runs locally too, so your audio doesn’t have to leave your machine.

🧵 1/3

[d.amuna @dna4ty]

Main model: https://t.co/Y6RXHDkDPP

CPU/ONNX: https://t.co/cLW9kVKi7T

KenLM: https://t.co/DH4OHYeygr

Main model: audio files + live mic. ONNX: local CPU, no PyTorch. KenLM: more context and better word accuracy.

[d.amuna @dna4ty]

@JFPuget Even in French? Is there not enough speech data to have almost perfect ASR models for this?

[PyTorch @PyTorch]

The PyTorch-Triton 3.7 release introduces the Triton Plugin Extensions system, a framework for dynamically loading custom compiler passes, dialect ops, and DSL extensions into upstream Triton at runtime, without forking or recompiling.

Read the technical deep dive and how to get started here 👉 https://t.co/9BXHtatNyf

[d.amuna @dna4ty]

@SpaceXAI already a Gork build out. https://t.co/XF5j0uq7yA

[d.amuna @dna4ty]

@valangde Edon cast

[d.amuna @dna4ty]

@thdxr https://t.co/Y6RXHDkDPP An ASR model for some Ghanaian languages

[d.amuna @dna4ty]

@JFPuget you work in France with Nvidia?

[Corri @FPLCorrigan]

Sepp Blatter:

[Ed @bakaryshaedy]

@Varticoo This is the end🤣🤣🤣🤣

[Hater Report @HaterReport]

Referees after seeing Cape Verde had a chance to beat Argentina

[OpenAI @OpenAI]

Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.

https://t.co/OoM83SyISN

[Yun-Ta Tsai @yunta_tsai]

Many people think any given ML project is 99% training.

In reality, it’s 50% evaluation, 40% data cleaning, 8% integration, and 2% training.

The first two set the noise floor for learning. No ML magic matters; the model cannot lower the noise floor, as that’s the optimal bound of Shannon encoding of your data.

Thus, not a single day goes by without me thinking about ontology. Even the old labels have to be constantly reviewed.

[driss guessous @drisspg]

Holy chart crime

[💻🎒 @CodingNoobie]

@amritwt bro said

[d.amuna @dna4ty]

@PiotrZelasko Well i know what I'm finetuning today

[d.amuna @dna4ty]

I guess this uses the latest multimodal gemini embedding model

[Google Gemma @googlegemma]

Meet Gemma 4 12B!

A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license.

Bridging the gap between edge efficiency and advanced reasoning. Here is what’s new with Gemma 4 12B: 👇

[Google Gemma @googlegemma]

Meet Gemma 4 12B!

A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license.

Bridging the gap between edge efficiency and advanced reasoning. Here is what’s new with Gemma 4 12B: 👇