winnow
6 TagsUpdated Apache-2.0by EldanRing
Decision models by EldanRing, fine-tuned from Google's Gemma 4 and published as GGUF. Winnow reads the answer labels' logits after its own prompt; Ollaya runs the author's file on llama.cpp, on NVIDIA GPUs, Vulkan GPUs on Windows, Apple silicon or the CPU.
decisionmultilingualfine-tunedgguf4.6b7.5b12b
Tags
| Name |
|---|
| winnowlatestSame as winnow:12b.12.7 GB · 8192 ctx · 100+ languages |
| winnow:12bWinnow-12B, Q8_0 GGUF: 0.702 on typed decisions, 85.7 % on the author's JevBench set; a 16 GB GPU holds it.12.7 GB · 8192 ctx · 100+ languages |
| winnow:12b-visionWinnow 12B Q8_0 with the author's matching vision projector: PNG image decisions through libmtmd.12.8 GB · 8192 ctx · 100+ languages |
| winnow:e2bWinnow-E2B, Q8_0 GGUF (5 GB), uncalibrated: the smallest Winnow, 0.616 on typed decisions in 104 ms on an RTX 5090.5.0 GB · 8192 ctx · 100+ languages |
| winnow:e4bWinnow-E4B, Q8_0 GGUF, with the author's fitted temperature: 0.724 on typed decisions in 113 ms on an RTX 5090. Ollaya's recommended model.8.0 GB · 8192 ctx · 100+ languages |
| winnow:e4b-visionWinnow E4B Q8_0 with the author's matching vision projector: PNG image decisions through libmtmd.9.0 GB · 8192 ctx · 100+ languages |
Tags without a suffix load fp16 on a CUDA GPU and fp32 on CPU; add -fp16 or -fp32 to pin one precision.