Hugging Face

Hugging Face has made it seriously easy to deploy Gradio apps through the use of Spaces, would recommend everyone to give it a shot!

I'm thrilled to share Multi-LoRA Serving on 🤗Text Generation Inference. Are you navigating the complexities of managing multiple AI models? Say goodbye to the hassle and high costs! With Multi-LoRA, you deploy once and dynamically serve multiple specialized models using the same base model. This means significant cost savings and reduced operational complexity. 📉💡 In this blog you will learn: 🚀 How to deploy a single base model and use it to serve up 30+ specialized models 🧠 Build deep intuition on how Multi-LoRA serving works 💡 Gain insights on the cost efficiency and operational benefits of using Multi-LoRA serving https://lnkd.in/eWeDTsaV

You can now embed a Hugging Face datasets viewer preview as an iframe. I used this to create a Space for quickly viewing all the datasets inside a Hugging Face collection. For example, this Space https://lnkd.in/epyq_T6P allows you to explore all the datasets inside the Alpaca Style Datasets collection: https://lnkd.in/edWHzAFT. This can be very helpful for quickly exploring which datasets might be useful for a project. You can duplicate this Space to explore datasets in any HF collection!

HOT & NEW - NVIDIA's BigVGAN! In essence it is an AI-powered sound synthesizer! - It extracts the mel-spectrogram from the input audio and uses it to produce the actual matching sound waves! - Its power lies in its ability to do this for a wide range of sounds with high quality and flexibility. - BigVGAN is a universal neural vocoder model that generates audio waveforms using mel spectrogram as inputs for a wide variety of inputs like music, voice, sounds, and even those which are outside its training data. - BigVGAN Gradio demo is live on Hugging Face Spaces: https://lnkd.in/g4aBv6iP - Model collection Hugging Face Hub: https://lnkd.in/grvmb26z

We just integrated dataset viewer in AutoTrain 💥 So, now you can look into your dataset, identify correct splits and columns before training the model, without leaving the page 🚀

Google presents YouTube-SL-25 A Large-Scale, Open-Domain Multilingual Sign Language Parallel Corpus paper page: https://lnkd.in/eUAHknyb Even for better-studied sign languages like American Sign Language (ASL), data is the bottleneck for machine learning research. The situation is worse yet for the many other sign languages used by Deaf/Hard of Hearing communities around the world. In this paper, we present YouTube-SL-25, a large-scale, open-domain multilingual corpus of sign language videos with seemingly well-aligned captions drawn from YouTube. With >3000 hours of videos across >25 sign languages, YouTube-SL-25 is a) >3x the size of YouTube-ASL, b) the largest parallel sign language dataset to date, and c) the first or largest parallel dataset for many of its component languages. We provide baselines for sign-to-text tasks using a unified multilingual multitask model based on T5 and report scores on benchmarks across 4 sign languages. The results demonstrate that multilingual transfer benefits both higher- and lower-resource sign languages within YouTube-SL-25.

State of Open AI - June Edition! https://lnkd.in/db6HTysc June is a great month for the open ML space, but it's hard to keep up! I curated a list of new model releases and grouped them into categories. Let me know if this is useful :)

Alibaba presents Qwen2 Technical Report paper page: https://lnkd.in/eVHdeiZS This report introduces the Qwen2 series, the latest addition to our large language models and large multimodal models. We release a comprehensive suite of foundational and instruction-tuned language models, encompassing a parameter range from 0.5 to 72 billion, featuring dense models and a Mixture-of-Experts model. Qwen2 surpasses most prior open-weight models, including its predecessor Qwen1.5, and exhibits competitive performance relative to proprietary models across diverse benchmarks on language understanding, generation, multilingual proficiency, coding, mathematics, and reasoning. The flagship model, Qwen2-72B, showcases remarkable performance: 84.2 on MMLU, 37.9 on GPQA, 64.6 on HumanEval, 89.5 on GSM8K, and 82.4 on BBH as a base language model. The instruction-tuned variant, Qwen2-72B-Instruct, attains 9.1 on MT-Bench, 48.1 on Arena-Hard, and 35.7 on LiveCodeBench. Moreover, Qwen2 demonstrates robust multilingual capabilities, proficient in approximately 30 languages, spanning English, Chinese, Spanish, French, German, Arabic, Russian, Korean, Japanese, Thai, Vietnamese, and more, underscoring its versatility and global reach. To foster community innovation and accessibility, we have made the Qwen2 model weights openly available on Hugging Face1 and ModelScope2, and the supplementary materials including example code on GitHub3. These platforms also include resources for quantization, fine-tuning, and deployment, facilitating a wide range of applications and research endeavors.

Competition admins can now access admin control right from the competition space. Want to host a competition? Use the competition creator or contact us 🚀

Yesterday, Julien Chaumond tweeted about how the 🤗 Huggingface datasets viewer is now embeddable (https://lnkd.in/d-sDKrXR). Inspired, I made a quick PR to integrate the view in the ZenML dashboard. Wrote how I did it in a blog today, thought it be useful for the community: https://lnkd.in/dZzGH3Cj tldr; Return a 🤗 Huggingface dataset from a ZenML step, and see an embedded view of it right in the dashboard.