Argilla Warehouse

Fine-tune a SmolLM on domain-specific synthetic data from a LLM

Blog: https://huggingface.co/blog/davidberenstein1957/fine-tune-a-smollm-on-synthetic-data-of-llm

Fine-tuning ModernBERT for text classification using synthetic data generation

From prompt to model in 3 steps.
1 dataset description
20 minutes of generating
60 minutes of fine-tuning on my Macbook Pro

Tutorial: https://nbsanity.com/static/552eb50cbd91bedb4e5b73fddca2664a/fine-tune-modernbert-classifier.html

Introducing 📐𝐅𝐢𝐧𝐞𝐌𝐚𝐭𝐡: the best public math pre-training dataset with 50B+ tokens!
HuggingFaceTB/finemath

Math remains challenging for LLMs and by training on FineMath we see considerable gains over other math datasets, especially on GSM8K and MATH.

We build the dataset by:
🛠️ carefully extracting math data from Common Crawl;
🔎 iteratively filtering and recalling high quality math pages using a classifier trained on synthetic annotations to identify math reasoning and deduction.

We conducted a series of ablations comparing the performance of Llama-3.2-3B-Base after continued pre-training on FineMath and observe notable gains compared to the baseline model and other public math datasets.

We hope this helps advance the performance of LLMs on math and reasoning! 🚀
We’re also releasing all the ablation models as well as the evaluation code.

HuggingFaceTB/finemath-6763fb8f71b6439b653482c2

People are flexing their end of year stats, so I made this app to show hub stats in a tidy design!

Thanks @Ameeeee and @jfcalvo for the feature from Argilla!
burtenshaw/recap

🐇 Tumble down the AI rabbit hole without any technical knowledge!

Explore AI models on the Hub by a simple and quick search

Demo: davidberenstein1957/transformers-pipeline-playground

If you are still wondering how the FineWeb2 annotations are done, how to follow the guidelines or how Argilla works, this is your video!

I go through a few samples of the FineWeb2 dataset and classify them based on their educational content. Check it out!

https://www.youtube.com/watch?v=_-ORB4WAVGU

Introducing the Synthetic Data Generator, a user-friendly application that takes a no-code approach to creating custom datasets with Large Language Models (LLMs). The best part: A simple step-by-step process, making dataset creation a non-technical breeze, allowing anyone to create datasets and models in minutes and without any code.

Blog: https://huggingface.co/blog/synthetic-data-generator
Space: argilla/synthetic-data-generator

How do your annotations for FineWeb2 compare to your teammates'?

I started contributing some annotations to the FineWeb2 collaborative annotation sprint and I wanted to know if my labelling trends were similar to those of my teammates.

I did some analysis and I wasn't surprised to see that I'm being a bit harsher on my evaluations than my mates 😂


Do you want to see how your annotations compare to others?
👉 Go to this Gradio space: nataliaElv/fineweb2_compare_my_annotations
✍️ Enter the dataset that you've contributed to and your Hugging Face username.

How were your results?
- Contribute some annotations: data-is-better-together/fineweb-c
- Join your language channel in Rocket chat: HuggingFaceFW/discussion

Quick update from week 1 of smol course. The community is taking the driving seat and using the material for their own projects. If you want to do the same, join in!

- we have ongoing translation projects in Korean, Vietnamese, Portuguese, and Spanish
- 3 chapters are ready for students. On topics like, instruction tuning, preference alignment, and parameter efficient fine tuning
- 3 chapters are in progress on evaluation, vision language models, and synthetic data.
- around 780 people have forked the repo to use it for learning, teaching, sharing.

⏭️ Next step is to support people that want to use the course for teaching, content creation, internal knowledge sharing, or anything. If you're into this. Drop an issue or PR

REPO: https://buff.ly/3ZCMKX2
discord channel: https://buff.ly/4f9F8jA

Open Preference Dataset for Text-to-Image Generation by the 🤗 Community

Open Image Preferences is an Apache 2.0 licensed dataset for text-to-image generation. This dataset contains 10K text-to-image preference pairs across common image generation categories, while using different model families and varying prompt complexities.

https://huggingface.co/blog/image-preferences

🌐 Announcing Global-MMLU: an improved MMLU Open dataset with evaluation coverage across 42 languages, built with Argilla and the Hugging Face community.

Global-MMLU is the result of months of work with the goal of advancing Multilingual LLM evaluation. It's been an amazing open science effort with collaborators from Cohere For AI, Mila - Quebec Artificial Intelligence Institute, EPFL, Massachusetts Institute of Technology, AI Singapore, National University of Singapore, KAIST, Instituto Superior Técnico, Carnegie Mellon University, CONICET, and University of Buenos Aires.

🏷️ +200 contributors used Argilla MMLU questions where regional, dialect, or cultural knowledge was required to answer correctly. 85% of the questions required Western-centric knowledge!

Thanks to this annotation process, the open dataset contains two subsets:

1. 🗽 Culturally Agnostic: no specific regional, cultural knowledge is required.
2. ⚖️ Culturally Sensitive: requires dialect, cultural knowledge or geographic knowledge to answer correctly.

Moreover, we provide high quality translations of 25 out of 42 languages, thanks again to the community and professional annotators leveraging Argilla on the Hub.

I hope this will ensure a better understanding of the limitations and challenges for making open AI useful for many languages.

Dataset: CohereForAI/Global-MMLU

This is amazing for cheap models fine-tunes without the hassle of actual deployment! TIL: LoRA fine-tunes for models on the Hub can directly be used for inference!

The Data Is Better Together community is set to release the first Apache 2 licensed image preference dataset!

Great work and let's give this a final push :)

@aashish1904 congrats on your month of HF pro. There is more to win during this sprint!

@aashish1904 @AnyaDesdein @davidberenstein1957 @Malalatiana @beta3 @fffiloni @munish0838 @Reza2kn @bbunzeck @Creazycreator @andrei-saceleanu @jafhaponiuk @rca-etl @kf120 @burtenshaw @mmhamdy @grib0ed0v @Doopus @AnyaDes @ttkap @Xceron @Lewox @davanstrien @Azazelle @adirik @Ashish08 @AntonVic @kenantang @sdiazlor @g-ronimo @dennis-rall @prithivMLmods @girtss3 @flozi00 @WaveCut @Taylor658 @Wildminder @Sara9999 @phaelishall @sararob @dvilasuero @pgabrys @plaguss @CDS899 @timajwilliams @rudzinskimaciej @pavel-ai @aggr8 @ignacioct @MouseAI @Leeps @MaksKul @NicolasDmln @Muinez @kusht55 @caiolang @Jakub-Brand24 @loamy @Demijan @eliab96 @Viewegger @JosephCatrambone @p1atdev @mrshu @o639 @Targezed @Aviv-anthonnyolime @thliang01 @Ahmed-Amine @glards @pranaykoppula @nataliaElv @MaPirlet @alvarobartt @gabrielmbmb @zlicastro @Jaydip @Chouettecheveche @lilcheaty @ruyrdiaz @robintema @fdaudens @ggcristian @a-r-r-o-w @pates @joheras @stopsatgreen @bezo97 @chachi902 @iamyann @liamcripwell @dmb23 @korbih @anonymous7743 @akbdx18 @OVAWARE @severo @akontra @lichorosario @lhoestq @SebastianBodza @Vishnou @ameerazam08 @appoose @Mukei @mearco @joaquincabezas @Fizzarolli @thomastraum @igortopolski @OxxoCodes @patrickfleith @asoria @bn22 @sitammeur @Krodolf @bergr7f @Sbxxn @wietsevenema @sugatoray @Iamladi @MikeTrizna @feveromo @mokady @Bolero @prath @Dowwie @kfahn @decodingchris @alili2050 @RahulRaman @yzimmermann @Ameeeee @ecyht2 @MattMC001 @hemanthkumarak @Thegorgibus @akos2 @LawRun @ramithuh @SuperMuel @sjans @peterizsak @mosama @Eyel @mtr3 @cfahlgren1 @legentil @clem @Citaman @Aurelien-Morgan @AntoineBourgois @TotoB12 @Stanmey @osanseviero @multimodalart @maxiw @ariG23498 @ngk89 @femboysLover @dvs @tacohiddink @blanchon @DavidJimenez

We're so close to reaching 100 languages! Can you help us cover the remaining 200? Check if we're still looking for language leads for your language: nataliaElv/language-leads-dashboard

For anyone looking to boost their LLM fine-tuning and alignment skills this decemeber. We're running this free and open course called smol course. It’s not big like Li Yin and @mlabonne , it’s just smol.

👷 It focuses on practical use cases, so if you’re working on something, bring it along.

👯‍♀️ It’s peer reviewed and open so you can discuss and get feedback.

🤘 If you’re already a smol pro, feel free to drop a star or issue.

> > Part 1 starts now, and it’s on instruction tuning!

https://github.com/huggingface/smol-course

[SATURDAY ROUNDUP] ☕️🧑‍🎓

In case you missed everything this week. It’s all about vision language models and image preference datasets. Here are the models and datasets you can use in your projects.

QWQ-32B-Preview is the first open weights model to reason like o1 with comparable performance. It’s large but is acing some of the hardest tasks.

https://bsky.app/profile/philschmid.bsky.social/post/3lbylz6nzqk25

SmolVLM is a vision implementation of the recently released SmolLM2. It uses the Idefics3 approach to add a vision encoder. The main difference being the smaller language model (8b > 1.7b) and more compression of images. This results in a model that is very accurate for its memory footprint.

https://huggingface.co/blog/smolvlm

ColSmolVLM is a vision embedding model based on SmolVLM using the Colbert approach from ColPali. This is shown to be great at document retrieval and everyone should test it out in their RAG setups.

https://huggingface.co/posts/merve/663466156074132

In an effort to build a FLUX level open source image generation model, the community is building a dataset of image preferences. The dataset is already open and the project is still running. Join in!

https://huggingface.co/posts/davidberenstein1957/405018978675827

TRL tutorial Drop - This week I dropped a load of tutorials on finetuning and aligning models with TRL. If you’re upskilling in this space, you should check these out.

https://bsky.app/profile/benburtenshaw.bsky.social/post/3lbrc56ap3222

SmolVLM speeding locally on a laptop thanks to mlx-vlm and
@Gradio ! Try it with two lines:
pip install git+https://github.com/andimarafioti/mlx-vlm.git@stream-generate-fix
python -m mlx_vlm.chat_ui --model mlx-community/SmolVLM-Instruct-8bit

Gotta love the MLX community! Big thanks to @pcuenq and @prince_canuma !

🔥 Dataset Drop - Open Image Preferences

BlackForest Labs Flux Dev VS. Stability AI Stable Diffusion Large 3.5

Together with the ⁠data-is-better-together community, we've worked on an Apache 2.0 licensed open image preference dataset based on the fal ai imgsys prompts dataset. Thanks to the awesome community, we have managed to get 5K preference pairs in less than 2 days. The annotation alignment among annotators is great too.

Aashish Kumar won a month of Hugging Face Pro by making the most contributions! Congrats from the entire team 🥇

The best thing?! We are not done yet! Let's keep the annotations coming for 5K more in the second part of the sprint! (with more prices to go around).

Dataset: https://huggingface.co/datasets/data-is-better-together/image-preferences-results

Let's go! We are releasing SmolVLM, a smol 2B VLM built for on-device inference that outperforms all models at similar GPU RAM usage and tokens throughputs.

- SmolVLM generates tokens 7.5 to 16 times faster than Qwen2-VL! 🤯
- Other models at this size crash a laptop, but SmolVLM comfortably generates 17 tokens/sec on a macbook! 🚀
- SmolVLM can be fine-tuned on a Google collab! Or process millions of documents with a consumer GPU!
- SmolVLM even outperforms larger models in video benchmarks, despite not even being trained on videos!

Check out more!
Demo: HuggingFaceTB/SmolVLM
Blog: https://huggingface.co/blog/smolvlm
Model: HuggingFaceTB/SmolVLM-Instruct
Fine-tuning script: https://github.com/huggingface/smollm/blob/main/finetuning/Smol_VLM_FT.ipynb