mradermacher

/

model_requests

Like 239

patrickbdevaney/GLM-5.3-Flash-REAP50-FP8-v2 brings the size of the powerful GLM-5.3-Flash model down within reach of hobbyist multi-GPU rigs.

I would love to see GGUFs of this REAPed model, so we can use it with llama.cpp. I can barely get the unREAPed model working on my HPC servers, and even then only with llama.cpp options which circumvent filesystem caching and add about twenty-five minutes of load-time overhead before inference even begins.

Using this REAPed version should allow me to keep the model's weights cached.

Thanks for all you do!