| --- |
| license: apache-2.0 |
| base_model: |
| - decisionmodels/reax-1.1 |
| base_model_relation: finetune |
| library_name: pytorch |
| pipeline_tag: text-classification |
| tags: |
| - decision-model |
| - classification |
| - noncausal |
| - joint-questions |
| - sn92 |
| --- |
| |
| # c1 |
|
|
| c1 is a 27B decision model (joint-readout decider format). It is a fine-tune of |
| [decisionmodels/reax-1.1](https://huggingface.co/decisionmodels/reax-1.1) (revision `142b19355af00251f08db3d287307dd8b711c133`), itself derived from |
| pplx-decider-v1.2 (Perplexity AI, Apache-2.0) and Qwen3.8-27B (Apache-2.0). Not affiliated with or endorsed by |
| Perplexity AI, the Qwen team or REAX. |
|
|
| ## Protocol |
| Unchanged from the base: joint protocol v3 (template `joint-question-inline-readout-v1`), noncausal full attention, |
| one readout per question via the separate readout head (`readout.safetensors`, [255, 5120]); probabilities are a |
| softmax over the option codes. Loading uses the bundled, unmodified `pplx_decider` package in `source/src/` |
| (Perplexity AI, Apache-2.0). The fitted logit scale (x2.8) is folded into the backbone's final RMSNorm (equivalent to scaling every readout row), and the saved temperature in `decision_config.json` is scaled by the same factor and then refitted on held-out native-protocol rows (temperature 10.5667), so native probabilities stay calibrated. |
|
|
| ## Training |
| LoRA fine-tuning (all linear layers of the text backbone) plus the readout letter rows, on decisionmodels/sn92-generalist-train-v1 and decisionmodels/sn92-specialists (both CC-BY-4.0) plus public permissively licensed datasets listed in DATA_SOURCES.md, in the SN92 single-question prompt and the native pplx joint format; weighted soup of several runs; the logit scale fitted on held-out items. |
| |
| ## Intended use and limitations |
| Structured judgements where the answer is one of a closed set (routing, triage, moderation, grading, reranking). |
| Not for sole decision-making in high-stakes settings without human review. Same input limits as the base. |
| |
| ## Licence and attribution |
| Apache-2.0. Based on decisionmodels/reax-1.1, pplx-decider-v1.2 (Perplexity AI) and Qwen3.8-27B (Qwen team); see NOTICE and |
| MODIFICATIONS.md. Training data licences: decisionmodels/sn92-generalist-train-v1 and decisionmodels/sn92-specialists (CC-BY-4.0, credit: REAX / decisionmodels); public datasets under MIT, Apache-2.0, CC-BY-3.0/4.0 (DATA_SOURCES.md). |
|
|
