Can't wait to see multimodal LLama 3!

We released a resource that might come in handy: The Cauldron🍯

The Cauldron is a massive manually-curated collection of 50 vision-language sets for instruction fine-tuning. 3.6M images, 30.3M query/answer pairs.

It covers a large

Hugging Face PRO or Enterprise user?

Then you can combine distilabel 1.0.0 and Llama 3 recent releases to generate a synthetic text dataset using serverless endpoints 🤗

🦫 We have just released `Capybara-Preferences` in collaboration with KAIST AI and Hugging Face

A new synthetic preference dataset built using `distilabel` on top of LDJ Capybara dataset

More details 🧵

huggingface.co/datasets/argil…

Big announcement: pleias releases a massive open corpus of 2 million Youtube videos in Creative Commons (CC-By) on Hugging Face. Youtube-Commons features 30 billion words of audio transcriptions in multiple languages, and soon other modalities huggingface.co/datasets/PleIA…

Never doubt the impact of an emoji!

Did you know that Argilla and distilabel datasets have over 6 million hub downloads on the Hub? 🤯

Now, distilabel datasets will be even easier to identify thanks to the new icon added to the Hugging Face Hub—a nice addition to yesterday's release!

github.com/argilla-io/dis…

We've just added a new icon to indicate datasets created using Argilla's Distilabel on the Hugging Face Hub!

Good data is vital for AI so I'm very excited to see the growing number of data tools integrating with the Hub 🚀

💥After months of work, we're thrilled to introduce ⚗️distilabel 1.0.0!

🚀More flexible, robust, and powerful.

🙌 Let's empower the community to build the most impactful datasets for Open Source AI!

Blogpost: argilla.io/blog/introduci…
Github: github.com/argilla-io/dis…

It's not about being GPU poor or rich; it's about being data poor or rich.

Could DPO-style preference data be crucial for enhancing open LLMs across different languages?

Leveraging a 7k preference dataset, Argilla, Hugging Face, and KAIST AI utilized KAIST AI's ORPO technique with the latest Mistral AI MOE model to create a very high-performing

New multimodal model in town: Idefics2!

💪 Strong 8B-parameters model: often on par with open 30B counterparts.
🔓Open license: Apache 2.0.
🚀 Strong improvement over Idefics1: +12 points on VQAv2, +30 points on TextVQA while having 10x fewer parameters.
📚 Better data:

Could DPO-style preference data be crucial for enhancing open LLMs across different languages?

Leveraging a 7k preference dataset, Argilla, Hugging Face, and KAIST AI utilized KAIST AI's ORPO technique with the latest Mistral AI MOE model to create a very high-performing

Tgi 2.0 is out!

-back to fully open source for good (apache 2.0)
- Fastest inference server in existence (110 tok/s for cohere R+, with medusa speculation)
- fp8 support
- mixtral 8x22b support ! (also the fastest medusa on the way)

And much more to come
github.com/huggingface/te…

Welcome Zephyr 141B to Hugging Chat🔥

🎉A Mixtral-8x22B fine-tune
⚡️Super fast generation with TGI
🤗Fully open source (from the data to the UI)

huggingface.co/chat/models/Hu…

DS-1000 (ds1000-code-gen.github.io) code generation data format has now been simplified and hosted on Hugging Face datasets.

1⃣Simplified format: github.com/xlang-ai/DS-10…
2⃣DS-1000 Hugging Face: huggingface.co/datasets/xlang…

Credits: Yuhang Lai and Sida Wang

🔥Mixtral-8x22B-base + ORPO🔥

7k data & 1.3 hours to build a strong human-aligned 140B chat model🦾

👉IFEval: 65%
👉BBH: 59%
👉MT-Bench: 8.17

More models will be added to the Zephyr-ORPO collection with Argilla and Hugging Face , stay-tuned😃