overview for Gurrako

Is there an interest in resurrecting technical discussions of the latest research? [D] in c/machinelearning@academy.garden

[–] Gurrako@alien.top 1 points 9 months ago

I'm fairly certain it is permanent. Same thing happened with WallStreetBets. Occasionally browsing that subreddit was a lot of fun before the GME madness (and even in the early part of). Now the sub has lost a lot of the character it had before.

I imagine the same thing will happen here. Changes in the atmosphere of the subreddit will slowly push pre-ChatGPT members to look elsewhere for research / project related discussion and even after the hype dies down, they likely won't come back here having founds / made communities elsewhere.

Bill Gates told a German newspaper that GPT5 wouldn't be much better than GPT4: "there are reasons to believe that we have reached a plateau" [N] in c/machinelearning@academy.garden

[–] Gurrako@alien.top 1 points 9 months ago (1 children)

I don’t think so. I doubt GPT-4 will be able to convince someone who is trying to determine whether or not the if the think they are talking to is a human.

[R] Announcing Distil-Whisper - 6x faster than Whisper-large-v2 and performs within 1% WER on out-of-distribution in c/machinelearning@academy.garden

[–] Gurrako@alien.top 1 points 10 months ago

The ideas in the paper seem to be using a very similar concept from a NMT paper called Deep Encoders, Shallow Encoders. Surprised to see no citation in the related work.

[D] Why choose an H100 over an A100 for LLM inference? in c/machinelearning@academy.garden

[–] Gurrako@alien.top 1 points 10 months ago (2 children)

At first I thought that number was almost unbelievably high. It appears it can be 8x faster when using FlashAttention and a multi-GPU setup. Without multi-gpu and flash attention, it is a bit more than 2x faster.

Source: https://lambdalabs.com/blog/flashattention-2-lambda-cloud-h100-vs-a100