request for generation steps
#5
by kartikvyas1 - opened
hi, this is too cool. could you please tell how this was generated from bf16 mode. would love to learn. thanks in advance.
like how many sample inputs were taken. did you also use the deepcompressor framework shared by svdquant authors. if so how to tweak it for qwen2511.
quantization is not same as training. You can check the docs on huggingface for quantization and diffusers. There is no other model/lora external layers added , its still the same original Qwen.
if you mean qwen-2512, the nf4 is already provided by me in another repo.
ovedrive changed discussion status to closed