Hugging Face, one of the most popular platforms for AI tools, has hit a new milestone: it now hosts over 1 million models. The platform, which began as a chatbot app in 2016, shifted focus in 2020 to become a key resource for open-source machine-learning models. Its growth has been rapid as demand for artificial intelligence models tailored to specific use cases has surged.
CEO Clément Delangue emphasized on X that the strength of Hugging Face lies in its wide array of specialized models, not a one-size-fits-all approach. Many companies use these models to build custom solutions, fine-tuning them to meet specific needs such as language, domain, and hardware. Hugging Face's repository has quickly become essential for developers looking to create more personalized AI applications.
A Diverse AI Model Ecosystem
Hugging Face doesn't just focus on large language models like ChatGPT. The platform offers tools for a variety of tasks, including image generation, audio processing, and text analysis. It hosts well-known models such as Meta's Llama and OpenAI's CLIP, along with many others that serve more niche purposes.
We just crossed 1,000,000 free public models on Hugging Face!
That's the ones the media covers like Llama, Gemma, Phi, Flux, Mistral, Phi, Starcoder, Qwen, Stable diffusion, Grok, Whisper, Olmo, Command, Zephyr, OpenELM, Jamba, Yi but also 999,984 others. Why?
Because contrary…
— clem 🤗 (@ClementDelangue) September 26, 2024
Delangue highlighted the diversity on Hugging Face's platform by listing models like Whisper and Stable Diffusion, alongside hundreds of thousands more that have been fine-tuned for unique tasks. Fine-tuning involves taking an existing AI model and training it further on specialized data, which allows developers to tweak it for more targeted use cases.
The flexibility has attracted a global audience of researchers and engineers, leading to rapid growth. Product engineer Caleb Fahlgren recently shared that a new repository—whether it's a model, dataset, or application—is added every 10 seconds to Hugging Face's growing collection.
Popular AI Models on the Platform
The platform's most downloaded models offer insight into which technologies developers find most useful. Hugging Face's top spot is held by MIT's Audio Spectrogram Transformer, which has been downloaded 163 million times and helps with audio classification, including recognizing speech and music. Google's BERT, another popular choice, has been downloaded 54.2 million times and is primarily used for language tasks like sentence prediction.
Models designed for visual tasks also perform well on Hugging Face. The Vision Transformer, which processes images by breaking them into smaller patches for classification, ranks highly in terms of downloads. Another notable entry is all-MiniLM-L6-v2, which helps map sentences and paragraphs to dense vectors for semantic search applications.
HuggingChat: A New macOS AI App
As the company continues to grow, Hugging Face has also launched HuggingChat, an open-source macOS app that competes with ChatGPT. Still in its beta phase on GitHub, HuggingChat sports a straightforward interface with a conversation prompt at the top and a simple chat history below. Though the app is basic for now, Hugging Face has plans to add more features, such as image uploads, which are expected to roll out soon.
Users can already experiment with different open-source models like Llama 3.1 and Qwen 2.5, and there are a few customization options for power users, including keyboard shortcuts and appearance tweaks. Although additional themes are not yet available, the core features are functional, providing a solid base for AI-assisted conversations.
The app's appeal to open-source enthusiasts lies in its simplicity and cost—HuggingChat is entirely free to use. Hugging Face's entry into the desktop AI app space is a clear indication that the company aims to diversify its offerings beyond just hosting models and tools for developers.
