https://huggingface.co/nightmedia/Qwen3.8-27B-MindMeld
#2941
by nightmedia - opened
Dear Team Radermacher
Could you please quant this model?
https://huggingface.co/nightmedia/Qwen3.8-27B-MindMeld
arc arc/e boolq hswag obkqa piqa wino
bf16 0.742,0.897,0.913,0.840,0.514,0.834,0.791
mxfp8 0.743,0.899,0.911
qx86-hi 0.743,0.896,0.910
qx64-hi 0.747,0.899,0.907
mxfp4 0.739,0.891,0.916,0.827,0.522,0.826,0.790
Quant Perplexity Peak Memory Tokens/sec
bf16 3.693 ± 0.022 60.75 GB 235
mxfp8 3.733 ± 0.023 34.74 GB 186
qx86-hi 3.697 ± 0.022 33.25 GB 187
qx64-hi 3.713 ± 0.023 27.03 GB 190
mxfp4 3.813 ± 0.023 21.30 GB 191
Thank you, Richard! :)
-G
sorry for delay, it's cooking =)
You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#Qwen3.8-27B-MindMeld-GGUF for quants to appear.
Excellent, thank you Richard!
It just delayed the inevitable: now some people will have more fun :)
It just delayed the inevitable: now some people will have more fun :)
great =) can I have some fun ?
If you use MindMeld, you might learn Kung Fu. Some say :)