Viewing profile — ramesh1994

Sorry I don't know how hackernews notifications work, didn't see this reply. But if you see this here's another reminder for you to clean this up into a public demo :)

This is really cool! Appreciate sharing this work and the explanation. You mentioned that massaging the data into shape as one of the problems, which I think is possibly one of the…

Hey this sounds pretty cool! If this is public would you mind sharing a link?

I just found about this project from this comment, absolutely excited to try this out. As someone who's never used any of the infrastructure tools, I'm thinking of pyinfra as a way…

I think distillation in the original sense isn't being done anymore but finetuning on outputs from larger models like GPT-4 is a form of distillation (top-1 logit vs all logits and…

Head over to the elueuther.ai discord and discuss with some of the folks there. Tons of small experiments with LLMs can use the $10k in compute

It prohibits anything that competes with OpenAI services i.e as long as you're not literally providing an LLM API commercially you should be fine

> This means it is way cheaper to look something up in a vector store than to ask an LLM to generate it. E.g. “What is the capital of Delaware?” when looked up in an neural informa…

I think parts of the write-up are great. There are some unique assumptions being made in parts of the gist > 10: Cost Ratio of OpenAI embedding to Self-Hosted embedding > 1: Cost R…

I've been looking for a course like this! Especially great given how much of the recent progress in training large models is made possible with the aid of flash attention and fused…

Was it for seeding/hosting torrents or from just downloading them? How long did the whole thing take to play out? I've always assumed that consuming torrents has been low stakes to…

It is a pretty fun game https://wiki-race.com/

The term "chinchilla" predates llama/alpaca. It doesn't directly map to a specific model, rather a family of compute-optimal models.

That fact that OpenAI decided to not even reveal the parameter counts / training data composition in their "technical" report is surely a sign that their moat isn't as big as it wo…

I would also highly recommend the blog post series [1] from Cliqz talking about the tech behind the search. [1] - https://0x65.dev/

I think it is definitely a hard problem to solve on a large scale to address latency, quality and size of the index they plan to address. It definitely isn't as easy as spinning up…

While I do agree with what you've said, I think they should be able to price these into a bill that they can send the app developers monthly. They have shown competency in running …