Just chat.deepseek.com with prompts adapted from this gist.
- For the
qX_0
variants, they are actually quite straight-forward so deepseek can come up with a correct result in 1 shot. - For the
qX_K
it's more complicated, I would say most of the time I need to re-prompt it 4 to 8 more times. - The most difficult was
q6_K
, the code never works until I ask it to only optimize one specific part, while leaving the rest intact (so it does not mess up everything)