https://huggingface.co/nightmedia/Qwen3.8-27B-Metaphor
Dear Team Radermacher,
Could you please quant this model?
https://huggingface.co/nightmedia/Qwen3.8-27B-Metaphor
This model is an experimental merge of:
- nightmedia/Qwen3.8-27B-MindMeld
- schneewolflabs/B0-27B
- OmniJev/OneJev1.6-27B
- shunzou05/TerminalHorizon-27B
- etemiz/Ostrich-27B-261028-Qwen3.8-Abliterated
- inclusionAI/ZwZ2-27B
- gordonhu/Tev-27B
Brainwaves
arc arc/e boolq hswag obkqa piqa wino
mxfp8 0.743,0.902,0.915,0.839,0.540,0.834,0.794
q6-hi 0.743,0.900,0.915
qx64-hi 0.743,0.906,0.910
mxfp4 0.740,0.893,0.917
Quant Perplexity Peak Memory Tokens/sec
mxfp8 3.620 ± 0.022 34.64 GB 201
q6-hi 3.615 ± 0.022 30.44 GB 212
qx64-hi 3.629 ± 0.022 26.93 GB 219
q4-hi 3.652 ± 0.022 23.72 GB 214
mxfp4 3.706 ± 0.022 21.20 GB 221
This is a Halloween model 😃
-G
It's queued! cmon win all benchmarks =)
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-Metaphor-GGUF for quants to appear.
Excellent, now the fun shall begin, metaphorically speaking :)
how much free time do you have for these?
Practical time spent is maybe a few hours a day, the rest is waiting for tests to finish.
When it comes to publishing, I am actually holding back, to avoid flooding HF with clones :)
Every day I pick one out of 3-4 good merges that is distinctive enough to make space for itself, or interesting to me in the long run, publish the best or most representative quant of the emergent behavior I am studying, share a trace for the lab notes which I conveniently keep on my model cards, and flood LinkedIn with the whole trace so people get used to how a model should work, for free.
It's rewarding in its lack of compensation :)
oh nice, mr senior software engineer =)
good luck, waiting for all the benchmarks to be solved by one of your 4b models