https://huggingface.co/vicgalle/Configurable-Llama-3-8B-v0.3
#30
by
deleted
- opened
The freeze and koboldcpp issues are bugs in those programs, for koboldcpp there is a fixed version available, but the correct configuration must be used. the "fix" that is usually applied breaks the tokenizer, even though it superficially works.
not that i mind quanting v.03, of course :) it's in the queue and should be available in a few hours. cheers!
mradermacher
changed discussion status to
closed
@mradermacher Thanks for quantizing v0.3, and I respect your decision to stick to the standard. It's been over a month since GPT4All has been updated, which is about a year in LLM time, but a new release appears imminent.