@jankais3r You were right: the 65B model failed. The process was killed. Here’s also a screenshot of asitop 5min into the process.

Eventually, the swap file reached the size of 70GB and more.

I’ll try the 30B. https://t.co/dZEgs8Ssrk

@jankais3r My apologies: I didn’t realize that your implementation and @ggerganov would be different, but I saw your last tweet and so I’d be happy to test both.

To use yours, do I have to repeat the quantization process I did for llama.cpp? https://t.co/cIbw4aYhcm

#LLaMA working on a smartphone. Yeah. https://t.co/G8csbCUNg7

@ggerganov Stupid question: do I have to repeat the quantization process after the interactive chat mode update? (or after any update)

The best MidJourney v5 (pre-release) photos I’ve seen so far. Very impressive, especially the last two. https://t.co/p0se0UFPFm

@rauchg @EMostaque @thomasglopes Not just programming code migration, but automation workflow migration (Ansible, Puppet, Chef, Terraform, etc.), cloud provider migration, etc.

@rauchg @EMostaque @thomasglopes Migrating to the next tech platform, whichever it will be, provided decent conditions, will not be impossible anymore because of huge migration costs and complexity lock-in.

@rauchg @EMostaque @thomasglopes The other side of this is that the value proposition of those vendors claiming to protect customers from lock-in (all lies anyway) will become even weaker than now.

Now, this is a game changer. https://t.co/hFWdnN3qvn

The new issue of Synthetic Work is out, people.

This week’s free edition is titled Issue #2 – Law firms’ morale at an all-time high now that they can use AI to generate evil plans to charge customers more money

Read it here: https://t.co/v2Y8o5706M https://t.co/qX6SmyYfMN