# JevAlt decision model by Mert Kaya (Eschatia Labs)

> JevAlt offers local 4B decision models for English, Turkish and German. Read its unknown-answer option, licence and recorded evaluation limits.

Updated 2026-11-12 (Australia/Adelaide)

## Quick answers

### What is JevAlt by Mert Kaya (Eschatia Labs)?

[JevAlt](<https://decisionmodels.ai/models/jevalt>): Three [Intern-Decision-4B](<https://decisionmodels.ai/models/intern-decision>) fine-tunes: Deem-4B for English, Karar-4B for Turkish and Wähler-4B for German.

[JevAlt facts](<https://decisionmodels.ai/models/jevalt>)

[Source](<https://github.com/mertkayacs/jevalt>)

### Who makes JevAlt?

[JevAlt](<https://decisionmodels.ai/models/jevalt>): Mert Kaya (Eschatia Labs)

[JevAlt facts](<https://decisionmodels.ai/models/jevalt>)

[Source](<https://github.com/mertkayacs/jevalt>)

### Are JevAlt's weights open?

[JevAlt](<https://decisionmodels.ai/models/jevalt>): Full weights, Q4\_K\_M/Q5\_K\_M/Q8\_0 GGUF, Transformers and llama.cpp server Licence: Apache-2.0 code and weights; dataset terms remain separate

[JevAlt facts](<https://decisionmodels.ai/models/jevalt>)

[Source](<https://github.com/mertkayacs/jevalt>)

By Mert Kaya (Eschatia Labs) · October weights; catch-up discovery: 2026-10-01 · OPEN

Three [Intern-Decision-4B](<https://decisionmodels.ai/models/intern-decision>) fine-tunes: Deem-4B for English, Karar-4B for Turkish and Wähler-4B for German. They return per-option probabilities through the System One API. Optional extensions add unknown, conformal answer sets or a short generated reasoning trace; reasoning is off by default. This is a catch-up October release.

## Facts

- Size: 4B per language model
- Inputs: Text; English, Turkish or German by model
- Limits: 8k context; up to 62 choice options; score uses 2 to 10 levels
- Licence: Apache-2.0 code and weights; dataset terms remain separate
- Access: Full weights, Q4\_K\_M/Q5\_K\_M/Q8\_0 GGUF, Transformers and llama.cpp server

## Claims and sources

- **VENDOR-REPORTED**: The maker reports Deem-4B accuracy 94.7% and Brier 0.091 on held-out English decisions from the same pipeline as training. In 11 held-out missing-fact cases, both [JevAlt](<https://decisionmodels.ai/models/jevalt>) and its [Intern-Decision](<https://decisionmodels.ai/models/intern-decision>) base choose unknown in 9; this does not establish a fine-tuning gain. A separate 4 Oct live test records 122/130 correct. [source](<https://huggingface.co/mertkayacs/Deem-4B>)
- **VENDOR-REPORTED**: The Deem Q4\_K\_M file is 2.71 GB; the maker measures peak RAM of 3.04 GB at 4k context. The maker reports Q4\_K\_M selected-option agreement with full precision of 95.2% to 97.5%, on 62 English, 61 Turkish and 40 German validation decisions. These are tested configurations, not universal hardware minima. [source](<https://huggingface.co/mertkayacs/Deem-4B-GGUF>)

## Links

- [Source: Code, licence and serving](<https://github.com/mertkayacs/jevalt>)
- [Source: Deem English card](<https://huggingface.co/mertkayacs/Deem-4B>)
- [Source: Karar Turkish card](<https://huggingface.co/mertkayacs/Karar-4B>)
- [Source: Wähler German card](<https://huggingface.co/mertkayacs/Wahler-4B>)
- [Source: Recorded live requests](<https://huggingface.co/datasets/mertkayacs/jevalt-bench/tree/main/results/tested>)

## Make a first local request

The local guide follows [JevAlt](<https://decisionmodels.ai/models/jevalt>)'s documented Python setup, default Deem GGUF server and typed support-ticket request. Saved author-run examples are linked separately from the setup commands.

[First local JevAlt request](<https://decisionmodels.ai/local-decision-models#first-local-request>)

[Source](<https://github.com/mertkayacs/jevalt>)

## Related projects

### JevAlt local decisions

[JevAlt local decisions](<https://decisionmodels.ai/projects#project-jevalt>)

[Source: mertkayacs/jevalt](<https://github.com/mertkayacs/jevalt>)

Local System One server and CLI for Deem, Karar and Wähler, with optional unknown answers and short reasoning. The maker publishes 390 live requests across the three languages, including misses. Emberwick plays recorded model decisions in a browser village; those replays are not fresh inference.

- Creator: Mert Kaya

- Stars: 7

- Licence: Apache-2.0

- Model: [JevAlt](<https://decisionmodels.ai/models/jevalt>)

- Tags: local, multilingual, server

- [Source: Licence](<https://github.com/mertkayacs/jevalt/blob/main/LICENSE>)
- [Source: Decision engine](<https://github.com/mertkayacs/jevalt/blob/main/src/jevalt/engine.py>)
- [Source: Recorded live requests](<https://huggingface.co/datasets/mertkayacs/jevalt-bench/tree/main/results/tested>)