Speaker unverified: belief

Transcript context

…ain its sense of humor. So not very funny, but at least it can run on device. Yeah, so I think now we have good small models, but we also have like good frameworks and tools to use these small models. So I think we're really close to having like really on edge and on device models that are really good. And I think for a while we've had this narrative. But just training larger models is better. Of course, this is supported by science scaling laws. As you can see here, for example, when we scale the model size, the loss is lower and obviously you get a better model. But and we can see this, for example, in the GPT family of models, how we went from just a hundred million parameters to more than a trillion. parameters. And of course, we all observed the performance improvement when using the latest model. But [00:19:00] one thing that we shouldn't forget is that when we scale the model, we also scale the inference costs and time. And so the largest models were are going to cost so much more. So I think now instead of just building larger models, we should be focusing on building more efficient models. It's no longer a race for the largest models since these models are really expensive to run and they require like a really good infrastructure to do that and they cannot run on, for example, consumer hardware. And when you try to build more efficient models that match larger models, that's when you can really unlock some really interesting on device use cases. And I think a trend that we're noticing now is the trend of training smaller models longer. For example, if you compare how much, how long LLAMA was trained compared to LLAMA3, there is a huge increase in the pre training length. LLAMA was trained on 1 trillion tokens, but LLAMA3 8b was trained on 15 trillion tokens. So Meta managed to get a model that's the same size, but But it performs so much [00:20:00] better by choosing to like spend the sacrifice during training, because as we know, training is a one time cost, but inference is something that's ongoing. If we want to see what are like the small models reads in 2024, I think this mobile LLM paper by Meta is interesting. They try to study different models that are like have the less than 1 billion parameters and find which architecture makes most sense for these models. For example, they find that depth is more important than width. So it's more important to have models that have like more layers than just one. making them more wide. They also find that GQA helps, that tying the embedding helps. So I think it's a nice study overall for models that are just a few hundred million parameters. There's also the Apple intelligence tech report, which is interesting. So for Apple intelligence, they had two models, one that was like on server and another model that was on device. It had 3 billion parameters. And I think the interesting part is that they trained this model using [00:21:00] pruning. And then distillation. And for example, they have this table where they show that, like, using pruning and distillation works much better than training from scratch. is model using [00:21:00] pruning. And then distillation. And for example, they have this table where they show that, like, using pruning and distillation works much better than training from scratch. And they also have some interesting insights about, like, how they specialize their models on specific tasks, like, for example, summarization and rewriting. There's also this paper by NVIDIA that was released recently. I think you've already had a talk about, like, hybrid models that was all interesting. And this model, they used, like, a hybrid architecture between state space models and transformers. And they managed to train a 1B model that's really performant without needing to train it on a lot of tokens. And regarding our work, we just recently released SmallM2, so it's a series of three models, which are the best in class in each model size. For example, our 1. 7b model outperforms Lama 1b and also Qt 2. 5. And how we managed to train this model is the following. That's where you spent a lot of time trying to curate the pre training datasets. We did a lot of [00:22:00] ablations, trying to find which datasets are good and also how to mix them. We also created some new math and code datasets that we're releasing soon. But you basically really spent a lot of time trying to find what's the best mixture that you can train these models on. And then we spent some time trying to like we also trained these models for very long. For example, small M1 was trained only on 1 trillion tokens, but this model is trained on 11 trillion tokens. And we saw that the performance kept improving. The models didn't really plateau mid training, which I think is really interesting. It shows that you can train such small models for very long and keep getting performance gains. What's interesting about SmallLM2 is that it's fully open. We also released, like the pre training code base, the fine tuning code, the datasets, and also evaluation in this repository. Also there's, like, really interesting small models for text, but also for vision. For example, here you can see SmallVLM, which is a 2B model that's really efficient. It doesn't consume a lot of RAM, and it also has a good performance. There's also Moondream 0. [00:23:00] 5b, which was released recently. It's like the smallest visual language model. And as you can see, there isn't like a big trade off compared to Moondream 2b. So now I showed you that we have some really good small models. We also have the tools to use them, but why should you consider using small models and when? I think, like, small models are really interesting because of the on device feature. Because these models are small and they can run fast, you can basically run them on your laptop, but also on your mobile phone. And this means that your dataset stays locally. You don't have to send your queries to third parties. And this really enhances privacy. That was, for example, one of the big selling points for Apple Intelligence. ns that your dataset stays locally. You don't have to send your queries to third parties. And this really enhances privacy. That was, for example, one of the big selling points for Apple Intelligence. Also, right now, we really have a lot of work to do. So many frameworks to do on device inference. For example, there's MLX, MLC, Llama, CPP, Transformers, JS. So we have a lot of options and each of them have like great features. So you have so many options for doing that. Small models are also really powerful if you choose to specialize them. For example, here there's a startup called Numind, which took small LM and then they fine tuned it on text extraction datasets. And they managed to get a model that's not very far from models that are much larger. So I think text extraction is like one use case where small models can be really performant and it makes sense to use them instead of just using larger models. You can also chat with these models in browser. For example, here, you can go there, you can load the model, you can even turn off your internet and just start chatting with the model locally. Speaking of text extraction, if you don't want to fine tune the models, there's a really good method of structure generation. We can basically force the models to follow a JSON schema that you defined. For example, here, we try to force the model to follow a schema for extracting key information from GitHub issues. So you can input free text, which is a complaint about a GitHub repository, something not working. And then you can run it there and the model can extract anything that is relevant for your GitHub issue creation. For example, the [00:25:00] priority, for example, here, priority is high, the type of the issue bug, and then a title and the estimation of how long this will take to fix. And you can just like do this in the browser, you can transform your text into a GitHub issue that's properly formatted. So what's next for synthetic data and small models? I think that domain specific synthetic data is going to be, it's already important, it's going to be even more important. For example, generating synthetic data for math. I think this really would help improve the reasoning of a lot of models. And a lot of people are doing it, for example, Quint 2. 12 math, everyone's trying to reproduce a one. And so I think for synthetic data, trying to specialize it on some domains is going to be really important. And then for small models, I think specializing them through fine tuning, it's also going to be really important because I think a lot of companies are just trying to use these large models because they are better. But on some tasks, I think you can already get decent performance with small models. So you don't need to Pay like a [00:26:00] cost that's much larger just to make your model better at your task by a few percent. And this is not just for text. And I think it also applies for other modalities like vision and audio.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.