Automated README summary
Automatically summarized from the README on 6 Oct 2026; not a run by jevbooks. Run instructions and author-reported outcomes still need verification.
What it is
Jevstral is a decision model that reads a document and a set of typed questions, returning a calibrated probability for each option in one forward pass. It is for software that needs to act on thresholds, such as routing support tickets.
How it uses Jev
Jevstral uses Jev's decision-model design: each question becomes a token row with reserved Mistral tokens; a frozen Ministral 3 8B decoder with LoRA adapters processes it; a pointer head scores each option and a softmax with temperature T=2.04 gives calibrated probabilities. A state cache reuses the document for multiple questions.
Primitives:choicescorenoul
Technique worth stealing
Use a frozen decoder with LoRA and a pointer head to produce calibrated option probabilities in one forward pass.
Try it
git clone https://github.com/AmirBraham/jevstral && cd jevstral; uv sync --extra inference; uv run hf download AmirBraham/jevstral-8b --include "stage4/final/*" --local-dir weights; then use Predictor.
Model judgments from README excerpts
Project signals use repository metadata, an existing summary and the first 700 README characters, not the full project. “Not found” means not detected in that material; it does not establish absence. These are the original judgments with recorded models and dates, not source audits or run results.
Models & evaluation dates
- Inclusion gate
jev-1.13.0- Project signals
jev-1.13.0- Category
jev-1.13.0- Pattern tags
jev-1.13.0
Evidence
These signals use repository metadata, an existing summary and at most 700 cleaned README characters. ≥ 0.60 means reported in that excerpt; ≤ 0.40 means not found there. The middle band is inconclusive. None is a source-code audit or a test result. The runnable-tool question excludes research experiments; the recommendation question asks for a learning shortlist, not a quality verdict.
- Jev is core to the projectnot found in the excerpt0.27
- Shows a System One patternFound in material0.68
- Describes handling uncertaintynot found in the excerpt0.05
- Reports measured numbersnot found in the excerpt0.06
- Is a runnable toolunclear0.50
- Top-10 learning shortlist signalnot found in the excerpt0.25
- Is a replica of the modelFound in material0.89
- Solves a common problem (0–2)score on a 0–2 scale1.42 / 2
- About JevFound in material0.66
no-judgment band (0.40–0.60)
no-judgment band (0.40–0.60)
The Jev model and date for each judgment stage are listed above. Card written by DeepSeek V4.1 Flash from the README on 6 Oct 2026.