Comments URL: https://news.ycombinator.com/item?id=49785783

Points: 5

# Comments: 2

Comments URL: https://news.ycombinator.com/item?id=49785429

Points: 217

# Comments: 81

Comments URL: https://news.ycombinator.com/item?id=49785397

Points: 111

# Comments: 4

Comments URL: https://news.ycombinator.com/item?id=49784706

Points: 90

# Comments: 39

Comments URL: https://news.ycombinator.com/item?id=49784336

Points: 75

# Comments: 39

Comments URL: https://news.ycombinator.com/item?id=49783999

Points: 150

# Comments: 66

Comments URL: https://news.ycombinator.com/item?id=49783495

Points: 222

# Comments: 47

So, first of all it does work and you can see the sample from the whole training run here: https://raw.githubusercontent.com/volotat/mini-AGI/refs/head...

Here is the scaling law graph I have so far, and it looks very promising: https://github.com/volotat/mini-AGI/blob/main/assets/scaling...

The model was built under my deep dissatisfaction so we cannot really train even moderately big models (1B+ scale) on the consumer's hardware. We can inference and fine-tune them for sure, but I would like to have full control over what the model sees over the training run, so it is fully aligned with my interests, not some corporations.

I was thinking about for some time and come up with two interesting ideas I thought worth pursuing: MoE with a lot of experts that gets added and pruned from the model while it trains, where only a small subset of of experts are actually in use at any particular moment + batch 1 training on the single continuous stream of data.

First allows us to be bounded only by the disk space in terms of number of parameters and load and unload experts only when they are needed. The second (if figured out and it turns out to be doable) allows us to get aways with small VRAM capacity because we do not need to store big randomized batches and their respective gradients.

I started brainstorming with Claude and after some time we found an approach that seems to be promising, and low and behold, a few weeks pass and you can see the results yourself.

Obviously, I did use AI in the process of making this project and I am pretty sure it would be completely impossible for me to do something like this without it, so I hope it is more than justified.

The model is still running over the first of 7.8B characters corpus I selected for training, so the weights are not out yet, and it's about a couple weeks of waiting until they are cooked at the current reading speed. And yeah, the model just read continuous interleaved passages from the dataset, each by 32K characters long each as a single stream. Just as you or I would do.

The set up seems to be really simple so you can git clone the project, run it and observe everything for yourself.

Thanks for your attention.

Comments URL: https://news.ycombinator.com/item?id=49783133

Points: 130

# Comments: 19

Comments URL: https://news.ycombinator.com/item?id=49783101

Points: 52

# Comments: 15

Comments URL: https://news.ycombinator.com/item?id=49781693

Points: 50

# Comments: 5

Comments URL: https://news.ycombinator.com/item?id=49781436

Points: 112

# Comments: 38

Comments URL: https://news.ycombinator.com/item?id=49780820

Points: 499

# Comments: 341

Comments URL: https://news.ycombinator.com/item?id=49780797

Points: 529

# Comments: 231

Comments URL: https://news.ycombinator.com/item?id=49780022

Points: 90

# Comments: 31

Comments URL: https://news.ycombinator.com/item?id=49779329

Points: 190

# Comments: 148

Comments URL: https://news.ycombinator.com/item?id=49778195

Points: 169

# Comments: 80

Comments URL: https://news.ycombinator.com/item?id=49778162

Points: 148

# Comments: 43

Comments URL: https://news.ycombinator.com/item?id=49778029

Points: 491

# Comments: 342

Comments URL: https://news.ycombinator.com/item?id=49778022

Points: 125

# Comments: 76

Comments URL: https://news.ycombinator.com/item?id=49777467

Points: 256

# Comments: 186