Jev directory
CommunityJev Models & Reimplementations
Laya Typed-Decisions on ModelScope
The typed-decisions checkpoint of the Laya family on ModelScope: a 421M ModernBERT-large fine-tune reporting 0.766 accuracy over 2,000 decisions, above TypeSafe Jev 1.13.0's published 0.727 and the 0.735 teacher ceiling, Apache-2.0.
- Reading time
- 2 min
- Published by
- Community
- Author
- Convai Innovations
- Last verified
- Oct 2026
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- A 421M ModernBERT-large model fine-tuned on typed-decisions workflows: agent-trace observability, customer service, invoice processing, and security incidents.
- Benchmark over 400 test cases and 2,000 decisions: accuracy 0.766, soft accuracy 0.471, Brier 0.062, ECE 0.213, and score MAE 0.242.
- Reported +3.9 points over TypeSafe Jev 1.13.0's published 0.727 accuracy, above the 0.735 teacher self-agreement ceiling.
- Single 842.6 MB F16 safetensors file; its 1,024-token context is twice the base laya checkpoint's 512.
- The comparison table includes a ModernBERT-base specialist at 0.646 and per-question majority class at 0.461 on the same split.
Watch out
Apache-2.0 per the metadata, under a community-mirror organisation. The card explicitly says the Jev figures are third-party published numbers not measured in this project, with no TypeSafe API access and differing sample sizes and prompts, so the +3.9-point claim is directional. ECE 0.213 is the weakest reported metric, and the laya base checkpoint scores 0.362 untuned on the same split.
Questions about this entry
Short answers drawn from the linked project or page.
A non-autoregressive System 1 decision model in the Laya family, fine-tuned on the typed-decisions workflows from the LocalLLaMA/typed-decisions dataset: agent-trace observability, customer service, invoice processing, and security incidents.
400 test cases and 2,000 decisions on the official test split, where the card reports accuracy 0.766, soft accuracy 0.471, Brier 0.062, ECE 0.213, and score MAE 0.242.
The card reports 0.766 accuracy against TypeSafe Jev 1.13.0's published 0.727, a +3.9-point gap, with 2.4x better Brier and 1.6x better score MAE, but it explicitly says the Jev figures are third-party published rather than measured in this project.
0.735, the point the card says this checkpoint scores above on the test split. For context, the untuned laya checkpoint scores 0.362 accuracy on the same split and laya-multilingual 0.342.
A published ModernBERT-base specialist at 0.646, per-question majority class at 0.461, random guess at 0.318, and the untuned laya and laya-multilingual checkpoints at 0.362 and 0.342. This checkpoint tops the table on accuracy, Brier, and score MAE.
One 842,609,220-byte safetensors file in F16 with a model_size figure of 421,293,830 bytes, sharing the Laya family's head while reading 1,024 tokens against the base laya checkpoint's 512.
The convaiinnovations organisation, whose description says it was created as a community mirror to host open artifacts and invites original owners to contact ModelScope about claiming ownership and management rights.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
Bespoke Labs' 9B decision model on Ollama, fine-tuned from Qwen3.5-9B: Apache-2.0, up to 64 typed questions per call, answers in under 100 ms on a MacBook Pro with an M5 Max, and 75.7% mean accuracy across 13 human-labelled datasets.
Jev Models & Reimplementations#community#open-models#local
Community
A Laya fine-tune from Convai Innovations on the 1,200-case typed-decisions benchmark: it reports 0.777 accuracy on the official 400-case test set, above TypeSafe Jev 1.13.0 at 0.727 and the teacher ceiling at 0.735.
Jev Models & Reimplementations#community#open-models#local
Community
J33.AI's small System One decision model: an EmbeddingGemma 2 encoder plus a 0.59M scorer, 271.6M parameters total, reporting 75.2% accuracy and ECE 0.019 on 2,000 typed-decisions test decisions while deciding alone on 39.9%.
Jev Models & Reimplementations#community#open-models#calibration
Community
Cloudflare's 9B multimodal decision model on Ollama: text, JSON, or images in one non-autoregressive forward pass, Apache-2.0, and the lowest measured latency of any decision model in Cloudflare's Decision Index runs at 38.8 ms median.
Jev Models & Reimplementations#community#open-models#local
Community
ActBro's 2B calibrated System 1 decision model on Ollama, built on OpenBMB's MiniCPM5-2B: 78.79% on JevBench among sub-2B open-weight models, 86.41% on the 10-source decision-v7 set, and one output token per decision via /v1/systemone.
Jev Models & Reimplementations#community#open-models#local
Community
AutoTrust's combined System 1 and System 2 bundle on Gemma-4-26B-A4B-it: Decision Index 0.2.1 of 58.05, a 24-slot head with noul, choice and score, and the base chat model left bit-identical.
Jev Models & Reimplementations#community#open-models#local
Community
From 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.