omneity-labs/spellbench
Viewer β’ Updated β’ 29.7k β’ 16
Omneity Labs
community
AI & ML interests
None defined yet.
Recent Activity
omarkamaliΒ
updated a
dataset 2 days ago
omarkamaliΒ
published a
dataset 2 days ago
omarkamaliΒ
posted an update 6 days ago
Post
4496
We got Qwen 3.5 to count Rs in Strawberry correctly! π¨
Building on Sawtone, weβve been testing a different way to feed language into an LLM to build the next generation of multilingual AI.
The usual setup gives the model tokenized text and asks it to perform various linguistic tasks. That works surprisingly well, until it doesnβt. Accents disappear. Words get mangled. Internal structure gets blurred away. And the cost of that gets higher once you move into multilingual and lower-resource settings.
So we tried adding a second path.
In addition to the normal text input, the model also receives Sawtone: a byte-level word representation that preserves how a word is written, how it sounds, and how it is structured.
Same LLM. Better interface.
In this proof of concept with Qwen 3.5 0.8B, that pushed our eval from 64% to 88%. The gains showed up exactly where tokenized models usually get shaky: diacritics, character order, exact spelling, and other form-sensitive behavior.
Sawtone itself is tokenizer-free, byte-level, and pre-trained across 507 languages.
Still early, but promising!
Building on Sawtone, weβve been testing a different way to feed language into an LLM to build the next generation of multilingual AI.
The usual setup gives the model tokenized text and asks it to perform various linguistic tasks. That works surprisingly well, until it doesnβt. Accents disappear. Words get mangled. Internal structure gets blurred away. And the cost of that gets higher once you move into multilingual and lower-resource settings.
So we tried adding a second path.
In addition to the normal text input, the model also receives Sawtone: a byte-level word representation that preserves how a word is written, how it sounds, and how it is structured.
Same LLM. Better interface.
In this proof of concept with Qwen 3.5 0.8B, that pushed our eval from 64% to 88%. The gains showed up exactly where tokenized models usually get shaky: diacritics, character order, exact spelling, and other form-sensitive behavior.
Sawtone itself is tokenizer-free, byte-level, and pre-trained across 507 languages.
Still early, but promising!
- 5 replies
omarkamaliΒ
updated a
Space 12 days ago
omarkamaliΒ
posted an update 13 days ago
Post
217
π LID Benchmark update:
β’ 10 Regional Leaderboards
β’ 17 LID models (+7 new, incl. non-fastText based)
β’ 449 languages in total (200+ additional)
β’ Fixed: F1 macro reporting error
β’ Normalized language codes for more accurate results
The dataset is also updated, now with individual model predictions to reproduce and validate our findings.
omneity-labs/lid-benchmark
β’ 10 Regional Leaderboards
β’ 17 LID models (+7 new, incl. non-fastText based)
β’ 449 languages in total (200+ additional)
β’ Fixed: F1 macro reporting error
β’ Normalized language codes for more accurate results
The dataset is also updated, now with individual model predictions to reproduce and validate our findings.
omneity-labs/lid-benchmark
omarkamaliΒ
updated a
dataset 13 days ago
omarkamaliΒ
posted an update 24 days ago
Post
218
Omneity Labs LID Benchmark is live π₯
- 8 Evals
- 10 Models (GlotLID, OpenLID, our own Gherbal and others)
- 200+ Languages
- One Leaderboard To Rule Them All!
Come find your language and which LID model supports it best in this space π
omneity-labs/lid-benchmark
- 8 Evals
- 10 Models (GlotLID, OpenLID, our own Gherbal and others)
- 200+ Languages
- One Leaderboard To Rule Them All!
Come find your language and which LID model supports it best in this space π
omneity-labs/lid-benchmark
omarkamaliΒ
updated a
dataset 24 days ago
omarkamaliΒ
published 2
datasets 24 days ago
omarkamaliΒ
published a
Space 24 days ago
omarkamaliΒ
posted an update 25 days ago
Post
1872
I just might have cracked tokenizer-free LLMs. No vocab, no softmax.
I'm training a 22M params LLM rn to test this "thing" and it's able to formulate coherent sentences π€―
Bear in mind, this is a completely new, tokenizer-free LLM architecture with built-in language universality.
Check the explainer video to understand what's happening. Feedback welcome on this approach!
I'm training a 22M params LLM rn to test this "thing" and it's able to formulate coherent sentences π€―
Bear in mind, this is a completely new, tokenizer-free LLM architecture with built-in language universality.
Check the explainer video to understand what's happening. Feedback welcome on this approach!
- 14 replies
omarkamaliΒ
posted an update about 1 month ago
Post
337
You're probably training on outdated Wikipedia data right now and don't know it. π‘
In June last year, a friend from the Moroccan Wikipedia community slid into my DMs: "Are you using the current version? The official dataset is severely outdated. We added so many articles nowhere to be found on HuggingFace."
He was right. I was running a 2023 snapshot. In 2025. The official Wikipedia dataset, the one hundreds of labs and researchers grab by default without a second thought, was frozen in time.
β’ For English, that's 700,000 missing articles.
β’ For Moroccan Arabic, 30% of the language's entire Wikipedia.
β’ For 31 other languages, there was literally no text corpus at all until recently.
I could've shrugged and moved on. Instead I spent the next months building a monthly automated pipeline for 340+ languages, on my personal laptop, nearly killing it several times in the process (100% disk, frozen screen, the works).
Nous Research trained Hermes 4 on it. INRIA cited it. It's now three years ahead of what most people are training on.
Here's the full story of how I built Wikipedia Monthly π
https://omarkamali.com/blog/wikipedia-monthly-pipeline
In June last year, a friend from the Moroccan Wikipedia community slid into my DMs: "Are you using the current version? The official dataset is severely outdated. We added so many articles nowhere to be found on HuggingFace."
He was right. I was running a 2023 snapshot. In 2025. The official Wikipedia dataset, the one hundreds of labs and researchers grab by default without a second thought, was frozen in time.
β’ For English, that's 700,000 missing articles.
β’ For Moroccan Arabic, 30% of the language's entire Wikipedia.
β’ For 31 other languages, there was literally no text corpus at all until recently.
I could've shrugged and moved on. Instead I spent the next months building a monthly automated pipeline for 340+ languages, on my personal laptop, nearly killing it several times in the process (100% disk, frozen screen, the works).
Nous Research trained Hermes 4 on it. INRIA cited it. It's now three years ahead of what most people are training on.
Here's the full story of how I built Wikipedia Monthly π
https://omarkamali.com/blog/wikipedia-monthly-pipeline
omarkamaliΒ
posted an update 4 months ago
Post
1700
New year, new dataset π
I just released omarkamali/wikipedia-labels, with all the structural labels and namespace from wikipedia in 300+ languages. A gift for the data preprocessors and cleaners among us.
Happy new year 2026 everyone! π
I just released omarkamali/wikipedia-labels, with all the structural labels and namespace from wikipedia in 300+ languages. A gift for the data preprocessors and cleaners among us.
Happy new year 2026 everyone! π
omarkamaliΒ
posted an update 4 months ago
Post
302
Picomon v0.2.0 released! π«
- Supports all of AMD, Nvidia and Apple Silicon π§βπ§βπ§βπ§
- Beautiful TUI with themes (who said monitoring should be boring?) π
- Shareable Rig Cards! Boast to friends, family and foes alike π«¨
Get it now!
- Supports all of AMD, Nvidia and Apple Silicon π§βπ§βπ§βπ§
- Beautiful TUI with themes (who said monitoring should be boring?) π
- Shareable Rig Cards! Boast to friends, family and foes alike π«¨
Get it now!
uvx picomon or pip install picomon then picomon 
- 3 replies
omarkamaliΒ
posted an update 4 months ago
Post
3498
Hello
Just run
Repo at https://github.com/omarkamali/picomon
Or pypi at https://pypi.org/project/picomon
picomon! AMD GPU Monitoring made easyJust run
uvx picomon and behold:ββββββββββββββββββββββββββββββββββββββββββββ ββββββββββββββββββββββββββββββββββββββββββββ
β GPU 0 GFX 42% UMC 21% β β GPU 1 GFX 78% UMC 66% β
β PWR 135/250W (54%) VRAM 10.0/16.0GB 62% β β PWR 210/250W (84%) VRAM 14.5/16.0GB 90% β
β β β β
β GFX βββββββ
βββββββ
ββββ β β GFX ββββ
βββββββ
βββββ
β β
β PWR ββββββββ
βββββββ
βββ β β PWR βββββ
βββββββ
βββββ β
β VRM ββββββββ
ββββββββ
ββ β β VRM ββββ
βββββββββ
ββββ β
ββββββββββββββββββββββββββββββββββββββββββββ ββββββββββββββββββββββββββββββββββββββββββββRepo at https://github.com/omarkamali/picomon
Or pypi at https://pypi.org/project/picomon
omarkamaliΒ
posted an update 5 months ago
Post
5233
Exciting updates to the Wikipedia Monthly dataset for November! π
γ» Fixed a bug to remove infobox leftovers and other wiki markers such as
γ» New python package https://pypi.org/project/wikisets: a dataset builder with efficient sampling so you can combine the languages you want seamlessly for any date (ideal for pretraining data but works for any purpose)
γ» Moved the pipeline to a large server. Much higher costs but with better reliability and predictability (let me know if you'd like to sponsor this!).
γ» Dataset sizes are unfortunately missing for this month due to shenanigans with the migration, but should be back in December's update.
Check out the dataset:
omarkamali/wikipedia-monthly
γ» Fixed a bug to remove infobox leftovers and other wiki markers such as
__TOC__γ» New python package https://pypi.org/project/wikisets: a dataset builder with efficient sampling so you can combine the languages you want seamlessly for any date (ideal for pretraining data but works for any purpose)
γ» Moved the pipeline to a large server. Much higher costs but with better reliability and predictability (let me know if you'd like to sponsor this!).
γ» Dataset sizes are unfortunately missing for this month due to shenanigans with the migration, but should be back in December's update.
Check out the dataset:
omarkamali/wikipedia-monthly
omarkamaliΒ
posted an update 6 months ago
Post
306
Another month, another Wikipedia Monthly release! π
Highlights of October's edition:
Β· π£οΈ 341 languages
Β· π 64.7M articles (+2.5%)
Β· π¦ 89.4GB of data (+3.3%)
We are now sampling a random subset of each language with a reservoir sampling method to produce splits
Now you can load the english (or your favorite language) subset in seconds:
Happy data engineering! π§°
omarkamali/wikipedia-monthly
Highlights of October's edition:
Β· π£οΈ 341 languages
Β· π 64.7M articles (+2.5%)
Β· π¦ 89.4GB of data (+3.3%)
We are now sampling a random subset of each language with a reservoir sampling method to produce splits
1000, 5000, and 10000 in addition to the existing train split that contains all the data.Now you can load the english (or your favorite language) subset in seconds:
dataset = load_dataset("omarkamali/wikipedia-monthly", "latest.en", split="10000")Happy data engineering! π§°
omarkamali/wikipedia-monthly
- 2 replies
omarkamaliΒ
posted an update 7 months ago
Post
1609
**Wikipedia Monthly's September edition is now live π**
Highlights of this edition:
Β· π£οΈ 341 languages
Β· π 63.1M articles
Β· π¦ 86.5GB of data
This update also solves upload issues in the August edition where some languages had missing parts. Happy data engineering!
omarkamali/wikipedia-monthly
Highlights of this edition:
Β· π£οΈ 341 languages
Β· π 63.1M articles
Β· π¦ 86.5GB of data
This update also solves upload issues in the August edition where some languages had missing parts. Happy data engineering!
omarkamali/wikipedia-monthly
- 2 replies