India's Rising Influence in AI Development and Open Source Innovation

Hugging face's Jeff Boudier discusses India's ascendance in AI development and open-source innovation - ET Telecom

Jeff Boudier, the head of products at AI startup Hugging Face, on India's rapid rise as the second largest base of AI developers in the world and the state of play in the rapidly evolving sector India is a leading nation for AI builders, second only to the USA in terms of users of the Hugging Face Hub, Jeff Boudier, chief product and growth officer of Hugging Face says. In his first interaction with Indian media, the top executive of the $4.5 billion-valued startup speaks with ET about the benefits of open-source versus closed-source AI, the company's investments in making AI accessible, and the competition ecosystem. The New York-based AI startup has become the world's largest open-source community platform for AI builders, hosting a repository of 773,235 models, 179,623 open datasets, downloaded over a billion times by one million developers across the world as of Friday. Founded by Clément Delangue, Julien Chaumond and Thomas Wolf back in 2016, Hugging Face counts Google, Amazon, Nvidia, Intel, AMD, Qualcomm, IBM and Salesforce among its backers. The platform offers free CPU to GPU accelerators and inference endpoints at $0.032 per hour, which is among the cheapest in the world. On India's significance India is a leading country for AI builders. We can see this first-hand, as India is the number 2 country after the US in terms of users of the Hugging Face hub. On AI commercialisation We just came out of the discovery phase of the new capabilities of AI that was spurred by ChatGPT and LLMs. And now we're getting into the industrialisation phase. We're at a moment where AI is spilling over the initial audience of experts, which started with AI researchers through the transform[1]ers library...and then AI was quickly adopted by data scientists at companies. And today, we're at a phase where software developers are starting to build with AI. On open-source vs closed-source The world needs AI that is not centralised, where companies (and) nations can build their own models and then their own AI systems. We don't want any one company or handful of companies to be the gatekeepers of AI and decide what is appropriate. So, to do that, you need to build the tools that people can use and try to make them as accessible and easy to use as possible. In terms of perfor[1]mance, over the past year, the best open models are catching up to the best closed models. The best closed models are getting better; and the best open models are getting better faster than them. So, I think that the perceived gap is shrinking. Maybe you have, I don't know, a 10% difference. But then, if you adapt the open model, because you can fine-tune it to your use case without a lot of compute, without a lot of data, then you get even better per[1]formance and impact than a closed model. On Hugging Face investments We are investing a lot in technologies and tools that can really help with accessibility, and making sure that models can run as fast as possible whenever possible on the customer's device. So, you are not relying on expensive server costs, on high speed bandwidth, and the benefits of AI can be evenly distributed amongst the population. Hugging Face has also launched the 'Zero GPU' programme to support startups with $10 million of compute capacity. On the cost of running AI The efforts we're putting on local inference of models, I think, are very important. So, the transformers, JS (JavaScript) library, our work in collaboration with our partners...make it possible to run more models locally. And then another piece is also de[1]creasing the cost of running AI on the server. And we are partnering with Amazon, Google Cloud, Azure and Cloudflare to make that a reality. On the hardware side, we are working with Nvidia, AMD, Intel, (and) Dell. One of our main contributions is our text generation inference, open-source library, which can really decrease by orders of magnitude the cost of running large language models. On the possibility of a takeover We have built bridges to all third-party clouds - Amazon SageMaker, Azure AI Studio, Google Cloud Vertex AI etc. But we also offer our own independent cloud, hosting hundreds of thousands of public mod[1]els. But GitHub is owned by Microsoft. So, in that sense, the situation is a bit different. On rivals like Ollama It (Ollama) is really, really cool. And we're contributing a lot to that ecosystem. One thing which we help HF users to set up is, which local hardware they have available to them. Okay, I have this GPU card on my laptop, I have access to this accelerator, and based on what you have, you can, in just two clicks, go from a model to using it in llama. cpp locally, or we have a variety of different apps that have one click into integrations. On growth of the business I'm really focused on enterprise use cases... How companies are using it, giving them more ways to build their own AI with open source, whichever cloud they're using, whichever kind of hardware they have available. That's where I'm personally focused on.

Hugging face's Jeff Boudier discusses India's ascendance in AI development and open-source innovation

India is a leading nation for AI builders, second only to the USA in terms of users of the Hugging Face Hub, Jeff Boudier, chief product and growth officer of Hugging Face says. In his first interaction with Indian media, he talked about the benefits of open-source versus closed-source AI, Jeff Boudier, the head of products at AI startup Hugging Face, on India's rapid rise as the second largest base of AI developers in the world and the state of play in the rapidly evolving sector India is a leading nation for AI builders, second only to the USA in terms of users of the Hugging Face Hub, Jeff Boudier, chief product and growth officer of Hugging Face says. In his first interaction with Indian media, the top executive of the $4.5 billion-valued startup speaks with ET about the benefits of open-source versus closed-source AI, the company's investments in making AI accessible, and the competition ecosystem. The New York-based AI startup has become the world's largest open-source community platform for AI builders, hosting a repository of 773,235 models, 179,623 open datasets, downloaded over a billion times by one million developers across the world as of Friday. Founded by Clément Delangue, Julien Chaumond and Thomas Wolf back in 2016, Hugging Face counts Google, Amazon, Nvidia, Intel, AMD, Qualcomm, IBM and Salesforce among its backers. The platform offers free CPU to GPU accelerators and inference endpoints at $0.032 per hour, which is among the cheapest in the world. On India's significance India is a leading country for AI builders. We can see this first-hand, as India is the number 2 country after the US in terms of users of the Hugging Face hub. On AI commercialisation We just came out of the discovery phase of the new capabilities of AI that was spurred by ChatGPT and LLMs. And now we're getting into the industrialisation phase. We're at a moment where AI is spilling over the initial audience of experts, which started with AI researchers through the transform[1]ers library...and then AI was quickly adopted by data scientists at companies. And today, we're at a phase where software developers are starting to build with AI. On open-source vs closed-source The world needs AI that is not centralised, where companies (and) nations can build their own models and then their own AI systems. We don't want any one company or handful of companies to be the gatekeepers of AI and decide what is appropriate. So, to do that, you need to build the tools that people can use and try to make them as accessible and easy to use as possible. In terms of perfor[1]mance, over the past year, the best open models are catching up to the best closed models. The best closed models are getting better; and the best open models are getting better faster than them. So, I think that the perceived gap is shrinking. Maybe you have, I don't know, a 10% difference. But then, if you adapt the open model, because you can fine-tune it to your use case without a lot of compute, without a lot of data, then you get even better per[1]formance and impact than a closed model. On Hugging Face investments We are investing a lot in technologies and tools that can really help with accessibility, and making sure that models can run as fast as possible whenever possible on the customer's device. So, you are not relying on expensive server costs, on high speed bandwidth, and the benefits of AI can be evenly distributed amongst the population. Hugging Face has also launched the 'Zero GPU' programme to support startups with $10 million of compute capacity. On the cost of running AI The efforts we're putting on local inference of models, I think, are very important. So, the transformers, JS (JavaScript) library, our work in collaboration with our partners...make it possible to run more models locally. And then another piece is also de[1]creasing the cost of running AI on the server. And we are partnering with Amazon, Google Cloud, Azure and Cloudflare to make that a reality. On the hardware side, we are working with Nvidia, AMD, Intel, (and) Dell. One of our main contributions is our text generation inference, open-source library, which can really decrease by orders of magnitude the cost of running large language models. On the possibility of a takeover We have built bridges to all third-party clouds - Amazon SageMaker, Azure AI Studio, Google Cloud Vertex AI etc. But we also offer our own independent cloud, hosting hundreds of thousands of public mod[1]els. But GitHub is owned by Microsoft. So, in that sense, the situation is a bit different. On rivals like Ollama It (Ollama) is really, really cool. And we're contributing a lot to that ecosystem. One thing which we help HF users to set up is, which local hardware they have available to them. Okay, I have this GPU card on my laptop, I have access to this accelerator, and based on what you have, you can, in just two clicks, go from a model to using it in llama. cpp locally, or we have a variety of different apps that have one click into integrations. On growth of the business I'm really focused on enterprise use cases... How companies are using it, giving them more ways to build their own AI with open source, whichever cloud they're using, whichever kind of hardware they have available. That's where I'm personally focused on.