pdurlej/ basal:1.5-4.5b-q4_K_M

234 3 weeks ago

Polish/English typed-decision models by Remek Kinas (rkinas/basal). omni-31b (ultra): #2 open-weights on JevBench, 87.9% on Werdykt; omni-12b matches the Jev API, locally and free. The author's GGUFs for basal-serve --mode ollama.

tools thinking
57460bde2bd4 ยท 1.4kB
basal
Copyright 2026 Remigiusz Kinas
This product includes models fine-tuned from the following works, licensed under the Apache License, Version 2.0:
basal-1.5-4.5B <- speakleash/Bielik-4.5B-v3.0-Instruct (https://huggingface.co/speakleash/Bielik-4.5B-v3.0-Instruct)
basal-1.5-max <- speakleash/Bielik-PL-11B-v3.0-Instruct (https://huggingface.co/speakleash/Bielik-PL-11B-v3.0-Instruct)
basal-1.5-mini <- speakleash/Bielik-1.5B-v3.0-Instruct (https://huggingface.co/speakleash/Bielik-1.5B-v3.0-Instruct)
basal-1.0-4.5B <- speakleash/Bielik-4.5B-v3.0-Instruct (https://huggingface.co/speakleash/Bielik-4.5B-v3.0-Instruct)
basal-1.0-1.5B <- speakleash/Bielik-1.5B-v3.0-Instruct (https://huggingface.co/speakleash/Bielik-1.5B-v3.0-Instruct)
The inference engine in this repository is original work licensed under the Apache License, Version 2.0.
Training data attribution (basal-1.0 models):
English decision items in the training data were converted from the public corpus
avbiswas/bev-decision-150K (https://huggingface.co/datasets/avbiswas/bev-decision-150K), which aggregates questions
derived from many upstream sources. Its maintainers ask users to check and attribute those upstream sources; row-level
source identifiers are not available, so the upstream sources of these items are not documented individually.
basal-1.5 models: released under the Apache License, Version 2.0.