HackerTrans
TopNewTrendsCommentsPastAskShowJobs

emikulic

no profile record

comments

emikulic
·3 years ago·discuss
It's over for decels.
emikulic
·3 years ago·discuss
Three epochs means it sees each token three times. The dataset is ~1T like you said.
emikulic
·3 years ago·discuss
I want flying cars and moving sidewalks like I was promised by The Jetsons. :(
emikulic
·3 years ago·discuss
Try https://huggingface.co/ehartford/WizardLM-7B-Uncensored and related models. They're not even trained on smut, just the neutered responses were removed before the RLHF stage (IIUC)
emikulic
·3 years ago·discuss
I used to be bad at math but then I did a 360° on that.
emikulic
·3 years ago·discuss
Emad said SD cost $600,000 to train. I wonder if Midjourney also had to pay that to train from scratch.
emikulic
·3 years ago·discuss
What benefits does the Huggingface diffusers(?) implementation have over A1111?
emikulic
·3 years ago·discuss
I like to run A1111 in --api mode and write my own script to drive it over HTTP.
emikulic
·3 years ago·discuss
Wow, Reddit is going to lose a lot of pagerank.
emikulic
·3 years ago·discuss
You're thinking of archive.is
emikulic
·3 years ago·discuss
You mean the one that defeated two superpowers?
emikulic
·3 years ago·discuss
Last time I looked into this, the answer was "because huggingface transformers and torch.load are written to do it this way"

You could absolutely do something streaming, or mmap the weights instead of loading them into system RAM. Just the default interfaces don't.
emikulic
·3 years ago·discuss
This used to be the case but nowadays you can just pip install e.g. https://libraries.io/pypi/nvidia-cuda-runtime-cu11
emikulic
·4 years ago·discuss
I hope you will run your own locally instead of letting Midjourney block and/or mess with your prompts.
emikulic
·4 years ago·discuss
If edge devices includes my gamer PC then yes. For apps I'd recommend https://github.com/AUTOMATIC1111/stable-diffusion-webui