Hugging Face's logo

ml-fw-prerelease
community
Activity Feed

AI & ML interests

None defined yet.
omarkamali 
posted an update 1 day ago
view post
Post
94
Just sharing a little breakthrough with Gherbal LID where we managed to distinguish the 15 variants of Arabic with 6 variants above 90%, 10 variants above 85% accuracy, practically distinguishing Moroccan and Algerian (which overlap massively).

It also embraces the duality of MSA and arabic variants pioneered in ALDi by @AMR-KELEG et al.

Now we're only bottlenecked by the availability of high quality data for the low scoring variants such as Iraqi, Libyan, Sudanese, Adeni ...

More on Gherbal at:
https://omneitylabs.com/models/gherbal

  • 1 reply
·
omarkamali 
posted an update 8 days ago
view post
Post
4514
We got Qwen 3.5 to count Rs in Strawberry correctly! 🚨

Building on Sawtone, we’ve 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 doesn’t. 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
·
omarkamali 
posted an update 15 days ago
view post
Post
220
🌐 LID Benchmark update:

• 10 Regional Leaderboards
• 17 LID models (+7 new, incl. non-fastText based)
• 449 languages in total (200+ additional)
• Fixed: F1 macro reporting error
• 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
omarkamali 
posted an update 26 days ago
view post
Post
220
Omneity Labs LID Benchmark is live 🔥

- 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 👇

omneity-labs/lid-benchmark
omarkamali 
posted an update 27 days ago
view post
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 🤯

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
·
omarkamali 
posted an update about 1 month ago
view post
Post
337
You're probably training on outdated Wikipedia data right now and don't know it. 💡

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.
• For English, that's 700,000 missing articles.
• For Moroccan Arabic, 30% of the language's entire Wikipedia.
• 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 👇

https://omarkamali.com/blog/wikipedia-monthly-pipeline
omarkamali 
posted an update 4 months ago
view post
Post
1700
New year, new dataset 🚀

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! 🎆
omarkamali 
posted an update 4 months ago
view post
Post
302
Picomon v0.2.0 released! 💫

- Supports all of AMD, Nvidia and Apple Silicon 🧑‍🧑‍🧒‍🧒
- Beautiful TUI with themes (who said monitoring should be boring?) 💅
- Shareable Rig Cards! Boast to friends, family and foes alike 🫨

Get it now! uvx picomon or pip install picomon then picomon
  • 3 replies
·
omarkamali 
posted an update 5 months ago
view post
Post
3498
Hello picomon! AMD GPU Monitoring made easy

Just run uvx picomon and behold:
┌──────────────────────────────────────────┐  ┌──────────────────────────────────────────┐
│ GPU 0  GFX  42%  UMC  21%                │  │ GPU 1  GFX  78%  UMC  66%                │
│ PWR 135/250W (54%)  VRAM 10.0/16.0GB 62% │  │ PWR 210/250W (84%)  VRAM 14.5/16.0GB 90% │
│                                          │  │                                          │
│ GFX ▁▂▂▃▄▄▅▆▆▇█▇▆▅▄▃▂▁                   │  │ GFX ▂▃▄▅▆▇██▇▆▅▄▂▂▃▅▆                    │
│ PWR ▁▁▂▂▃▄▄▅▆▇██▇▆▅▄▂▁                   │  │ PWR ▂▂▃▄▅▆▇██▇▆▅▄▃▂▂▃                    │
│ VRM ▁▁▂▂▃▄▄▅▆▇███▇▆▅▄▂                   │  │ VRM ▂▃▄▅▆▆▇███▇▆▅▄▃▂▂                    │
└──────────────────────────────────────────┘  └──────────────────────────────────────────┘


Repo at https://github.com/omarkamali/picomon
Or pypi at https://pypi.org/project/picomon
omarkamali 
posted an update 5 months ago
view post
Post
5233
Exciting updates to the Wikipedia Monthly dataset for November! 🚀

・ Fixed a bug to remove infobox leftovers and other wiki markers such as __TOC__
・ 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)
・ 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!).
・ 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
SivilTaram 
authored a paper 5 months ago
nouamanetazi 
posted an update 6 months ago
view post
Post
4726
After training 𝐒𝐦𝐨𝐥𝐋𝐌𝟑 on 𝟑𝟖𝟒 𝐇𝟏𝟎𝟎𝐬 for nearly a month, I've come to realize something most people overlook: 𝐢𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐢𝐬 𝐭𝐡𝐞 𝐦𝐚𝐤𝐞-𝐨𝐫-𝐛𝐫𝐞𝐚𝐤 𝐟𝐚𝐜𝐭𝐨𝐫 𝐢𝐧 𝐋𝐋𝐌 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠. 🔥

Everyone talks about model architecture and data quality. And yes, those matter immensely. But here's what nobody tells you: when your training run fails at 2 AM because of mysterious 𝐍𝐂𝐂𝐋 𝐞𝐫𝐫𝐨𝐫𝐬, or when your expensive GPU cluster is running at 𝟔𝟎% 𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐜𝐲, the problem isn't your model. It's most probably a 𝐦𝐢𝐬𝐮𝐬𝐞 𝐨𝐟 𝐭𝐡𝐞 𝐡𝐚𝐫𝐝𝐰𝐚𝐫𝐞. 🛠️

Questions that seemed simple but had no clear answers: Why is 𝐌𝐨𝐄 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐬𝐥𝐨𝐰𝐞𝐫 𝐭𝐡𝐚𝐧 𝐝𝐞𝐧𝐬𝐞 𝐦𝐨𝐝𝐞𝐥𝐬? Which 𝐍𝐂𝐂𝐋 𝐟𝐥𝐚𝐠𝐬 should we actually set? How often should we checkpoint without killing throughput?

That's why we built 𝐓𝐡𝐞 𝐒𝐦𝐨𝐥 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐏𝐥𝐚𝐲𝐛𝐨𝐨𝐤 📖: a complete guide covering everything from model architecture and data curation to the SmolLM3 training marathon, post-training techniques, and crucially, the 𝐢𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐥𝐚𝐲𝐞𝐫 that most teams get wrong.

We validated real vs theoretical bandwidth across the entire stack: 𝐇𝐁𝐌𝟑 𝐡𝐢𝐭𝐭𝐢𝐧𝐠 𝟑 𝐓𝐁/𝐬, 𝐍𝐕𝐋𝐢𝐧𝐤 𝟒.𝟎 𝐫𝐞𝐚𝐜𝐡𝐢𝐧𝐠 𝟕𝟖𝟔 𝐆𝐁/𝐬, 𝐏𝐂𝐈𝐞 𝐆𝐞𝐧𝟒 𝐚𝐭 𝟏𝟒.𝟐 𝐆𝐁/𝐬. Then we ran collective operations across 𝟏𝟐𝟖 𝐆𝐏𝐔𝐬 (16 nodes, 8xH100s each) and measured how performance degrades at scale: all-reduce drops from 𝟒𝟖𝟎 𝐆𝐁/𝐬 on a single node to 𝟑𝟐𝟎-𝟑𝟓𝟎 𝐆𝐁/𝐬 across 16 nodes.

If you've ever wondered why your training runs are slower than they should be, or you're planning to scale up and want to avoid expensive mistakes, this guide might save you weeks of debugging.

𝐓𝐡𝐞 𝐒𝐦𝐨𝐥 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐏𝐥𝐚𝐲𝐛𝐨𝐨𝐤: https://lnkd.in/e5MKXUHS

Shared with ❤️ by the HuggingFace team
Zaid 
authored a paper 6 months ago
lbourdois 
posted an update 6 months ago
omarkamali 
posted an update 6 months ago
view post
Post
306
Another month, another Wikipedia Monthly release! 🎃

Highlights of October's edition:
· 🗣️ 341 languages
· 📚 64.7M articles (+2.5%)
· 📦 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! 🧰

omarkamali/wikipedia-monthly
  • 2 replies
·
BramVanroy 
posted an update 6 months ago
view post
Post
632
What are currently the best multilingual models with at most 72B parameters? Are Llama 3.3 70B and Qwen 2.5 72B still king?
  • 1 reply
·
Zaid 
authored a paper 6 months ago
omarkamali 
posted an update 7 months ago
view post
Post
1609
**Wikipedia Monthly's September edition is now live 🎉**

Highlights of this edition:
· 🗣️ 341 languages
· 📚 63.1M articles
· 📦 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
·
SivilTaram 
authored a paper 8 months ago
alibayram 
authored a paper 8 months ago