Speaker unverified: preference

Transcript context

…odel [00:03:00] collapse. Is this the end? If you look at the media and all of like, for example, some papers in nature, it's really scary because there's a lot of synthetic data out there in the web. And naturally we train on the web. So we're going to be training a lot of synthetic data. And if model collapse is going to happen, we should really try to take that seriously. And the other issue is that, as I said, we think, a lot of people think the web is polluted because there's a lot of synthetic data. And for example, when we're building fine web datasets here at Guillerm and Hinek, we're interested in like, how much synthetic data is there in the web? So there isn't really a method to properly measure the amount of synthetic data or to save a webpage synthetic or not. But one thing we can do is to try to look for like proxy words, for example, expressions like as a large language model or words like delve that we know are actually generated by chat GPT. We could try to measure the amount of these words in our data system and compare them to the previous years. For example, here, we measured like a, these words ratio in different dumps of common crawl. [00:04:00] And we can see that like the ratio really increased after chat GPT's release. So if we were to say that synthetic data amount didn't change, you would expect this ratio to stay constant, which is not the case. So there's a lot of synthetic data probably on the web, but does this really make models worse? So what we did is we trained different models on these different dumps. And we then computed their performance on popular, like, NLP benchmarks, and then we computed the aggregated score. And surprisingly, you can see that the latest DOMs are actually even better than the DOMs that are before. So if there's some synthetic data there, at least it did not make the model's worse. Yeah, which is really encouraging. So personally, I wouldn't say the web is positive with Synthetic Data. Maybe it's even making it more rich. And the issue with like model collapse is that, for example, those studies, they were done at like a small scale, and you would ask the model to complete, for example, a Wikipedia paragraph, and then you would train it on these new generations, and you would do that every day. iteratively. I think if you do that approach, it's normal to [00:05:00] observe this kind of behavior because the quality is going to be worse because the model is already small. And then if you train it just on its generations, you shouldn't expect it to become better. But what we're really doing here is that we take a model that is very large and we try to distill its knowledge into a model that is smaller. And in this way, you can expect to get like a better performance for your small model. And using synthetic data for pre-training has become really popular. After the textbooks are all you need papers where Microsoft basically trained a series of small models on textbooks that were using a large LLM. thetic data for pre-training has become really popular. After the textbooks are all you need papers where Microsoft basically trained a series of small models on textbooks that were using a large LLM. And then they found that these models were actually better than models that are much larger. So this was really interesting. It was like first of its time, but it was also met with a lot of skepticism, which is a good thing in research. It pushes you to question things because the dataset that they trained on was not public, so people were not really sure if these models are really good or maybe there's just some data contamination. So it was really hard to check if you just have the weights of the models. [00:06:00] And as Hugging Face, because we like open source, we tried to reproduce what they did. So this is our Cosmopedia dataset. We basically tried to follow a similar approach to what they documented in the paper. And we created a synthetic dataset of textbooks and blog posts and stories that had almost 30 billion tokens. And we tried to train some models on that. And we found that like the key ingredient to getting a good data set that is synthetic is trying as much as possible to keep it diverse. Because if you just throw the same prompts as your model, like generate like a textbook about linear algebra, and even if you change the temperature, the textbooks are going to look alike. So there's no way you could scale to like millions of samples. And the way you do that is by creating prompts that have some seeds that make them diverse. In our case, the prompt, we would ask the model to generate a textbook, but make it related to an extract from a webpage. And also we try to frame it within, to stay within topic. For example, here, we put like an extract about cardiovascular bioimaging, [00:07:00] and then we ask the model to generate a textbook related to medicine that is also related to this webpage. And this is a really nice approach because there's so many webpages out there. So you can. Be sure that your generation is not going to be diverse when you change the seed example. One thing that's challenging with this is that you want the seed samples to be related to your topics. So we use like a search tool to try to go all of fine web datasets. And then we also do a lot of experiments with the type of generations we want the model to generate. For example, we ask it for textbooks for middle school students or textbook for college. And we found that like some generation styles help on some specific benchmarks, while others help on other benchmarks. For example, college textbooks are really good for MMLU, while middle school textbooks are good for benchmarks like OpenBookQA and Pico. This is like a sample from like our search tool. er benchmarks. For example, college textbooks are really good for MMLU, while middle school textbooks are good for benchmarks like OpenBookQA and Pico. This is like a sample from like our search tool. For example, you have a top category, which is a topic, and then you have some [00:08:00] subtopics, and then you have the topic hits, which are basically the web pages in fine web does belong to these topics. And here you can see the comparison between Cosmopedia. We had two versions V1 and V2 in blue and red, and you can see the comparison to fine web, and as you can see throughout the training training on Cosmopedia was consistently better. So we managed to get a data set that was actually good to train these models on. It's of course so much smaller than FineWeb, it's only 30 billion tokens, but that's the scale that Microsoft data sets was, so we kind of managed to reproduce a bit what they did. And the data set is public, so everyone can go there, check if everything is all right. And now this is a recent paper from NVIDIA, Neumatron CC. They took things a bit further, and they generated not a few billion tokens, but 1. 9 trillion tokens, which is huge. And we can see later how they did that. It's more of, like, rephrasing the web. So we can see today that there's, like, some really huge synthetic datasets out there, and they're public, so, [00:09:00] like, you can try to filter them even further if you want to get, like, more high quality corpses. So for this, rephrasing the web this approach was suggested in this paper by Pratyush, where basically in this paper, they take some samples from C4 datasets, and then they use an LLM to rewrite these samples into a better format. For example, they ask an LLM to rewrite the sample into a Wikipedia passage or into a Q& A page. And the interesting thing in this approach is that you can use a model that is Small because it doesn't, rewriting doesn't require knowledge. It's just rewriting a page into a different style. So the model doesn't need to have like knowledge that is like extensive of what is rewriting compared to just asking a model to generate a new textbook and not giving it like ground truth. So here they rewrite some samples from C4 into Q& A, into Wikipedia, and they find that doing this works better than training just on C4. And so what they did in Nemo Trans CC is a similar approach. [00:10:00] They rewrite some pages from Common Crawl for two reasons. One is to, like improve Pages that are low quality, so they rewrite them into, for example, Wikipedia page, so they look better. And another reason is to create more diverse datasets. So they have a dataset that they already heavily filtered, and then they take these pages that are already high quality, and they ask the model to rewrite them in Question and Answer format. into like open ended questions or like multi choice questions.…

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