Hugging Face

Hugging Face is a collaborative hub for an AI community, focused on democratising AI.. It is a platform for sharing tools, models, datasets, not specifically targeted AI image creation, but generative AI more broadly.

From chatbot to AI hub

Hugging Face initially set out in 2016 by French entrepreneurs Clément Delangue, Julien Chaumond, and Thomas Wolf.. Already in 2017 they received their first round of investment of $1,2 million, as it was stated in the press release, Hugging Face is a "new chatbot app for bored teenagers. The New York-based startup is creating a fun and emotional bot. Hugging Face will generate a digital friend so you can text back and forth and trade selfies." In 2021 they received a $40 million investment to develop its "open source library for natural language processing (NLP) technologies." There were "10,000 forks" (i.e., branches of development projects) and "around 5,000 companies [...] using Hugging Face in one way or another, including Microsoft with its search engine Bing." This shows how the company has gradually moved from providing a service (a chatbot) to becoming a major (if not the) platform for AI development – now, not only in language technologies, but also in speech synthesis, text-to-video. image-to-video, image-to-3D and much more within generative AI.

A community of users

Hugging Face is a platform, but what it offers is more resembling an infrastructure for, in particular, training models. Stability AI, and also others involved in training foundation models would typically have their own infrastructures, but they may make them available on Hugging Face, allowing the platform users to 'post train' and create LoRAs, for instance. It also allows users to experiment in other ways, for instance to create 'pipelines' of models, meaning that the outcome of one model can become the input for another model; or to upload their datasets to experimentation with the many models in Hugging Face. At the time of writing there are nearly 500,000 datasets and 2,000,000 models freely available for experimentation.

Many amateur users are signed up on the platform in order to do specific visual experiments. For instance the user 'mgane', who has shared a dataset of "76 cartoon art-style video game character spritesheets." The images are "open-source 2D video game asset sites from various artists", mgane has used them on Hugging Face to build LoRAs on Stable Diffusion, that is "for some experimental tests on Stable Diffusion XL via LORA and Dreambooth training methods for some solid results post-training."

One of "76 cartoon art-style video game character spritesheets."

A user like mgane is arguably both embedded in a specific 2D gaming culture, and also has the developer skills necessary to access and experiment with models in the command line interface. However, users can also access the many models in Hugging Face through more graphical user interfaces like Draw Things that allows for accessing and combining models and LoRAs to generate images, and also to train one's own LoRAs.

The extraction of value on Hugging Face

There are also professional corporations using Hugging Face to develop their own services. The 'Enterprise Hub' on Hugging Face offers, for instance, advanced computing at higher scale with a more dedicated hardware setup ('ZeroGPU', see also GPU), and also 'Priority Support'. In these cases of commercial purposes, access is typically more restricted.

Notably, Hugging Face also collaborates with a number of key corporations within AI. The company has, for instance, beenteaming up with, for instance, Meta to boost European startups in an "AI Accelerator Program". The company also collaborates with Amazon Web Services to allow users to deploy the trained and publicly available models in Hugging Face through Amazon SageMaker. Nasdaq Private Market lists a whole range of investors in Hugging Face (Amazon, Google, Intel, IBM, NVIDIA, etc.), and its estimated market value was $4.5 billion in 2023.

The excessive (and growing) market value of Hugging Face reflects, on the one hand, how the company capitalises on the various communities of developers in, for instance, visual culture who experiment on the platform and share their datasets and LoRAS. However, it also reflects the hight degree of expertise that has accumulated within a company that consistently has sought to accommodate these communities. Managing an infrastructure of both hardware and software for AI models at this large scale is, if not unprecedented, then very rare in the business of generative AI.

A diagram by the European Business review representing Hugging Face business model https://www.europeanbusinessreview.com/hugging-face-why-do-most-tech-companies-in-ai-collaborate-with-hugging-face/

Autonomy and 'platformisation'

Looking at Hugging Face, delinking from commercial interests in autonomous generative AI does not seem to be an either-or. Rather, dependencies on communities and attempts to decentralise generative AI seem to be in a constant movement, gravitating away from 'pee-to-peer' communities, towards platformisation and 'client-server relations'. This may be due to the high level of expertise and technical requirements involved in generative AI, but is not without consequence.

When, as noted by the European Business Review, most tech-companies in AI want to collaborate with Hugging Face, it is because the company offers an infrastructure for AI. Or, rather, it offers a platform that performs as an infrastructure for AI, or "linchpin" that keeps everything in the production in position. As also noted by Paul Edwards, platform seems to be, in a more general view, the new mode of handling infrastructures in the age of data and computation. Working with AI models is demanding task that requires both expertise, hardware and organisation of labour, and what Hugging Face offers is speed, reliability, and not least agility in a world of AI that is in constant flux, and where new models and techniques are introduced almost at a monthly basis.

With their 'linchpin status' they create social, commercial and other dependencies. That is, Hugging Face builds on existing infrastructures: they depend on already existing infrastructures, such as the flow of energy or water, necessary to make the platform run; they also build on social and organisational infrastructures, such as those of both start-ups and cultural communities. At the same time, however, they also reconfigure these relations – creating social and commercial dependencies on Hugging Face as a new 'platformed' infrastructure for AI.

++++++++++++++++++++++++++

CARD TEXT

Hugging Face

Hugging Face is a platform sharing tools, models, datasets, focused on democratizing AI. It is a sort of collaborative hub for an AI community, but it is also registered as a private company teaming up with, for instance, Meta to boost European startups in an "AI Accelerator Program". The company also collaborates with Amazon Web Services to allow users to deploy the trained and publicly available models in Hugging Face through Amazon SageMaker. Nasdaq Private Market lists a whole range of investors in Hugging Face (Amazon, Google, Intel, IBM, NVIDIA, etc.), and its estimated market value was $4.5 billion in 2023, which of course also reflects the high expertise the company has in managing models, hosting them, and making them available.

IMAGES: Business + interface/draw things + mgane + historical/contemporary interfaces.

++++++++++++++++++++++++

NOTES

A platform > extraction

. In this sense, it can also be compared to platforms such as Github, and what it really offers is an infrastructure, and not least an expertise, in handling all of this at a very large scale, involving nearly 2 million models,

Complex pipelines .. . you have access meaning you can have one model - outpost for another -

Foundation models training - customised set up / aim for smth that cannot be replicated

… value of HF . Pipeline of models / experimentation with

For instance, one might find a dataset for creating 2-D, cartoon-line video game characters.

Access to, and not least integration of models is a complicated issue that demands high technical expertise. More than anything, what Hugging Face offers is therefore an infrastructure.

I had used this same image set for some experimental tests on Stable Diffusion XL via LORA and Dreambooth training methods for some solid results post-training

Upload datasets - train models

.... training models at large scale

+ customisation for clients / manually racks of machines

Upload a dataset / accesss for others to download

But also to access models on HF

https://huggingface.co/datasets/mgane/2D_Video_Game_Cartoon_Character_Sprite-Sheets

"mgane/2D_Video_Game_Cartoon_Character_Sprite-Sheets" - a "76 cartoon art-style video game character spritesheets"

"All images editted using Tiled image editting software as most assets are typically downloaded individually and not in sequence. I compiled each animation sequence into one img to display animations frame-by-frame evenly distributed across some common animations seen in 2D video game art (Idle, Attack, Walk, Running, etc). I had used this same image set for some experimental tests on Stable Diffusion XL via LORA and Dreambooth training methods for some solid results post-training."

https://huggingface.co/datasets/mgane/2D_Video_Game_Cartoon_Character_Sprite-Sheets

but it is also registered as a private company teaming up with, for instance, Meta to boost European startups in an "AI Accelerator Program". The company also collaborates with Amazon Web Services to allow users to deploy the trained and publicly available models in Hugging Face through Amazon SageMaker. Nasdaq Private Market lists a whole range of investors in Hugging Face (Amazon, Google, Intel, IBM, NVIDIA, etc.), and its estimated market value was $4.5 billion in 2023, which of course also reflects the high expertise the company has in managing models, hosting them, and making them available.

What is the network that sustains this object?

• How does it move from person to person, person to software, to platform, what things are attached to it (visual culture)
• Networks of attachments
• How does it relate / sustain a collective? (human + non-human)

How does it evolve through time?

Evolution of the interface for these objects. Early chatgpt offered two parameters through the API: prompt and temperature. Today extremely complex object with all kinds of components and parameters. Visually what is the difference? Richness of the interface in decentralization (the more options, the better...)