User repository on GitHub
A Bangla-focused adaptation of Laya for typed decisions, classification, and routing.
- Forks
- 6
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Repos & SDKs
Laya Bangla
A Bangla-focused adaptation of Laya, the non-autoregressive typed-decision model: targeted data preparation, fine-tuning and evaluation for classification, routing and scoring on Bangla, Banglish and code-switched text, with the checkpoint still to come.
- Stars
- 40
- Language
- Python
- License
- APACHE-2.0
- Reading time
- 3 min
- Published by
- Community
- Author
- shamspias
- Last verified
- Oct 2026
Highlights
- Laya is a non-autoregressive decision model answering typed questions: choice for classification, noul for yes/no probability, and score for ordinal scoring.
- Available now: a pinned inference toolchain, locked Python 3.12 environment, a 36-message smoke set, and baseline results measured on Apple M2 Pro / MPS.
- The baseline evaluates Bangla and English question schemas on the same 36 messages, producing 216 typed decisions with a reproducibility manifest.
- Fine-tuning, reviewed datasets, probability calibration and a released laya-bangla checkpoint all remain planned; no Hugging Face checkpoint exists yet.
- Apache-2.0; the pinned upstream baseline is convaiinnovations/laya-multilingual at a recorded revision; the first run downloads about 678 MB.
Quickstart
bash
uv sync --locked
uv run --frozen laya-bangla doctor
uv run --frozen laya-bangla validate
uv run --frozen laya-bangla baseline --output outputs/baseline-001Watch out
README reviewed 2026-10-17; the first local results show strong sensitivity to question wording and poor churn/urgency decisions on several diagnostic cases, and the 36-message synthetic set is for diagnostics, not a representative Bangla benchmark. No fine-tuned model or training dataset has been released.
Questions about this entry
Short answers drawn from the linked project or page.
A Bangla-focused adaptation of Laya from shamspias, aiming to improve the typed-decision model's performance on real-world Bengali text through data preparation, fine-tuning and evaluation, with the resulting checkpoint intended for publication on Hugging Face.
Laya is a typed-decision model in the same class as TypeSafe's Jev, and the repo's topics include jev-alternative and jev-model. The digest shows no Jev API, endpoint or TypeSafe dependency, so the relationship is a shared decision-model shape, not an integration.
Available: documentation, a local baseline CLI with a locked Python environment, a synthetic 36-message smoke set, and first baseline measurements. Planned: reviewed training and held-out datasets, fine-tuning, probability calibration, and the released laya-bangla checkpoint on Hugging Face.
Install uv, then run uv sync --locked, followed by laya-bangla doctor, validate and baseline --output <dir>. The runner picks Apple MPS, CUDA or CPU automatically, or you can pass --device mps or --device cpu. The first run downloads about 678 MB of checkpoint and tokenizer files.
Both Bangla and English question schemas are evaluated on the same 36 messages, producing 216 typed decisions with per-script and per-task metrics, raw answers, timings and a reproducibility manifest. Each run records source, data, dependency-lock and checkpoint file hashes.
The measurements, completed on an Apple M2 Pro with MPS, show strong sensitivity to question wording and poor churn and urgency decisions on several diagnostic cases. The project presents them as a reproducible starting point for data work and fine-tuning, not as a success claim.
It is an independent adaptation by shamspias, not a Convai Innovations release. Original code and docs are Apache-2.0, matching the upstream Laya model cards; datasets keep their own licences. The digest shows 40 stars, 5 forks, created 2023-08-12 and last pushed 2026-09-28.
More Jev resources like this
A System One layout optimizer for decision-model workloads: it reorders rows and fields so a prefix-caching backend reuses input computation, then returns per-record decisions in original order. Its bundled backend serves AutoTrust JEV-9B through vLLM.
Jev Repos & SDKs#community#github#python
Community
0814wdwd/solo_jevPython
0814wdwd/solo_jevPythonA Python library that turns Jev-style decision models into ordinary function calls: check returns a bool, classify picks one option, label picks every option that applies, and score rates on a scale, with probabilities behind every answer.
Jev Repos & SDKs#community#github#python
Community
jlowin/vibecheckPython
jlowin/vibecheckPythonHTTP server for the Mica v0.5 4B Preview decision model: GGUF weights run through llama.cpp and answer noul, choice and score questions with calibrated probabilities over the TypeSafe /v1/systemone format.
Jev Repos & SDKs#community#github#python
Community
akivet/MicaPython
akivet/MicaPythonA Chinese-language decision model that returns a probability for each fixed candidate option in one forward pass, with no generated text or chain of thought. It is a LoRA fine-tune of Qwen3.5-0.8B on 13,048,000 CJ-Dataset rows, benchmarked on CJ-Bench against Chinese-Jev.
Jev Models & Reimplementations#community#github#python
Community
yanqiangmiffy/chinese-decisionPython
yanqiangmiffy/chinese-decisionPythonServing code for the Jeff 1.0 family of Jev-style decision models: a state, a typed question and its options go in, and one forward pass returns a probability for each option. Four open models from 4B to 31B parameters are served the same way.
Jev Repos & SDKs#community#github#python
Community
j-geuter/jeff-servePython
j-geuter/jeff-servePythonAn open-source, locally executed decision engine in the spirit of Jev (TypeSafe AI): state becomes features, and Noul, Choice and Score heads return typed, calibrated probabilities without generating text. It ships a cheap-to-strong cascade and a full train-calibrate-evaluate-deploy cycle.
Jev Repos & SDKs#community#github#python
Community
slavabars/decision-cascadePython
slavabars/decision-cascadePythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.