Smaller versions
#2
by snake8888 - opened
Any chance of a version based on minimax_h3_ref2va_pruned_int8_convrot.safetensors?
Or could a lora be pulled out (instead of a checkpoint) of the full bf16 version that can be used when using the smaller models?
Yeah but this is just a first test version, something that would evolve into a beta version thats worth doing that.
Looking forward to the beta version then... I'm curious to see the LTX and Wan2.2 influence on your model, but the 40GB size is too much for me at this point, so, thanks for your hard work and I'll keep an eye on this,
期待一个pruned int8的版本，感谢