Hugging Face, Inc. is a French-American company based in New York City that develops computer tools for building applications using machine learning. It is most notable for its transformers library built for natural language processing applications and its platform that allows users to share machine learning models and datasets and showcase their work.

Hugging Face, Inc.

Lux Capital

Betaworks Ventures

A.Capital Ventures

Alex Wang

Florian Douetteau

Kevin Durant

Olivier Pomel

Paul St. John

Rich Kleiman

Richard Socher

Betaworks

Coatue

Sequoia Capital

Addition

AIX Ventures

Thirty Five Ventures

Cygni Capital

Salesforce Ventures

Amazon

Sound Ventures

Google

Intel

NVIDIA

Qualcomm

Hugging Face

We're excited to announce a practical, powerful integration between #FiftyOne and Hugging Face's transformers library. - https://lnkd.in/gMsqVt9H

This collaboration makes it easier than ever to apply state-of-the-art transformer models directly to computer vision datasets — and it's all OPEN SOURCE! 

* Direct compatibility: Apply Hugging Face transformers models DIRECTLY to FiftyOne datasets

* Comprehensive task support: Image Classification, Object Detection, Depth Estimation, and Segmentation

* Multimodal: Zero-shot/open vocabulary Image Classification and Object Detection

* Video Inference: Run inference directly on the frames of videos

* Embeddings & Visualization: Visualize transformer model embeddings with dimensionality reduction techniques for insightful data exploration.

* Similarity Searches: Use transformer models to index your data, turning unstructured into searchable (including semantic search)

Shoutout to Daniel van Strien, Julien SIMON, Omar Sanseviero, Quentin Lhoest & Sylvain Lesage for helping to make this happen!

#AI #ML #transformers #ComputerVision #OpenSource #MachineLearning #DataScience #HuggingFace

What kind of datasets did the community build in <2 weeks? 

Announcing `10k_prompts_ranked`, the first dataset release from Data Is Better Together. Created in <2 weeks by the community. Includes:

- 10,000+ prompt quality rankings
- Human and synthetic data rankings
- Generated by 300+ contributors

How + why collaborative datasets?

It's no secret that high-quality open data is essential for creating better open models. The open source community shares 100s of models, datasets and demos openly weekly, but collectively building open datasets has been less explored.

Datasets have a massive role in shaping what models can be created. If we want more high-quality models for all languages, domains and tasks, we need more and better open datasets for all languages, domains and tasks!

To explore how the community could build impactful datasets collectively, Argilla added support for HF authentication for Argilla instances hosted on a @huggingface Space. Anyone with an HF login could begin contributing to a dataset in <1 minute.

To test this new workflow, we launched a task to rank the quality of prompts (human and synthetically generated).

In less than two weeks, we built a community of over 300 contributors for this dataset. 

This dataset became a reality thanks to the dedication of all the individuals who lent their support ❤️ To see the amazing people behind this dataset, visit https://lnkd.in/euKWdkmQ

This is just the beginning! We aim to empower the community to build new datasets collaboratively. This could include:

📊 Preference datasets for a low-resource language
🔍 Evaluations for a specific domain
🆕 Datasets for novel tasks

If this sounds interesting, keep your eyes peeled for further announcements later this week 🤗

📢 AI Game-Changer Alert: FiftyOne & Hugging Face Integration 📢

Excited to announce a practical, powerful integration between Voxel51's FiftyOne and Hugging Face's transformers library. This collaboration brings unparalleled ease to applying state-of-the-art transformer models directly to computer vision datasets — and it's all OPEN SOURCE! (Data 🤝 Models)

Key Highlights:
• Direct compatibility: Apply Hugging Face transformers models DIRECTLY to FiftyOne datasets
• Comprehensive task support: Image Classification, Object Detection, Depth Estimation, and Segmentation
• Multimodal: Zero-shot/open vocabulary Image Classification and Object Detection
• Video Inference: Run inference directly on the frames of videos
• Embeddings & Visualization: Visualize transformer model embeddings with dimensionality reduction techniques for insightful data exploration.
• Similarity Searches: Use transformer models to index your data, turning unstructured into searchable (including semantic search)

👇 Dive into the details and see the integration in action in our latest blog post (link in comments). Discover how this can transform your computer vision and AI projects.

#AI #ML #transformers #ComputerVision #OpenSource #MachineLearning #DataScience #HuggingFace

🚨 NEW COMPETITION ALERT 🚨
https://lnkd.in/dZvSB3-w
Building3D Challenge -- 1st Workshop on Urban Scene Modeling: Where Vision Meets Photogrammetry and Graphics

Introducing OpenHermesPreferences 1M! 🦋 We just released the largest open AI preference dataset on Hugging Face! 🤯 Together with Argilla, we extended the OpenHermes (Teknium) dataset into a pair-wise comparison dataset for RLHF and DPO. 🧠

Dataset:
📦 Size: ~1 million AI preferences
🤖 AI Feedback: Mixtral-8x7B-Instruct-v0.1 & Nous-Hermes-2-Yi-34B to generate AI feedback
👨⚖️ Judge: PairRM to score chosen & rejected pairs from the original data + AI feedback
🧠 Use Cases: RLHF, can be used with Direct Preference Optimization
🔨 Tools: Created with distilabel (Argilla) and llm-swarm (Hugging Face)
📄 License: same as OpenHermes dataset
🤗 Available on Hugging Face

Limitations: 
3️⃣ Sampling from only three models may limit diversity.
🥇 Rankings might be biased due to using one model trained on the source dataset (Yi). 
🚧 PairRM ranking model may truncate long inputs/outputs due to its 512 token context length limit.
🙅♂️ PairRM bias towards longer response, e.g. Mixtral-8x7B-Instruct-v0.1

Dataset: https://lnkd.in/eRDUqkww

If you want to learn how to train models with DPO check out my latest blog:
Blog: https://lnkd.in/eWfTtiVP

Much Kudos to Lewis Tunstall, Costa Huang, Kashif Rasul, Agustín Piqueres Lajarín and Daniel Vila Suero for this great release and collaboration

For the last five days, Gemma 7B has been the top trending text-gen model on Hugging Face. But if the last year in AI has taught us anything, it's only matter of time until the next hot model comes along and captures our collective attention. With some help from the Internet Archive, our team built the LLM Time Capsule (https://lnkd.in/g2XbNy6G) to chronicle the most popular text-gen models on Hugging Face since they starting tracking "trending" models last July. Looking back helps drive home just how rapid the rate of change in AI truly is. The world went from 14,000 OSS text-gen models to 59,000 models in 8 months 🤯  (By the time you read this, it will probably be north of 60,000).

At OctoAI we help our customers build resilient AI programs that turn the head-spinning rate of change into their strategic advantage. Check us out at octoai.cloud and have fun with the time capsule!

At Hugging Face, we provide expert support to enterprises willing to use open-source and open machine-learning models. Discover in ⬇️ case study how we advised Fetch in using Open-Source AI on Amazon Web Services (AWS).

Hugging Face worked together with Fetch to take their document #AI solutions to production on Amazon Web Services (AWS).
And guess what ? 👉 "With Yifeng’s guidance, Fetch was able to cut its development time by 30 percent."
https://lnkd.in/eh-U45HX

Yesterday, Google released Gemma, an open version based on Gemini! 🤗 Did you know you can try and experiment with it for free using the Hugging Face API and the OpenAI SDK? 🤯 Yes! With the new Message API of Text Generation Inference, you can change 2 parameters and switch from closed to open

```python
from openai import OpenAI

# initialize the client but point it to TGI
client = OpenAI(
base_url="https://lnkd.in/eA_eFfNR",
  api_key="hf_xxx") # Replace with your token

chat_completion = client.chat.completions.create(
    model="google/gemma-7b-it",
    messages=[
        {"role": "user", "content": "Why is open-source software important?"},
    ],
    stream=True,
    max_tokens=500
)

# iterate and print stream
for message in chat_completion:
    print(message.choices[0].delta.content, end="")
```


Gemma Blog: https://lnkd.in/e5nBPMnz
Message API Blog: https://lnkd.in/eVsFDkFS

If you hit rate limits, you can create a dedicated endpoint on Inference Endpoint with 1-click of a button or use Google Cloud. 🤗

CO:HUGGINGFACE

Hugging Face, Inc.