CommunityJev Models & Reimplementations
JevK5-9B
The 9B JevK5: the same v0.3 data on Qwen3.5-9B, retrained for 1.5 epochs by a pre-registered sweep, reaching index proxy 0.794 and a 0.859 hand-written hard set, Apache-2.0.
- Published by
- Community
- Author
- alibiserikbay
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Same v0.3 data as the 4B on Qwen3.5-9B; v0.3.3 trained 1.5 epochs after a pre-registered sweep adopted it over 1 and 2.
- Index proxy 0.794 against the 4B's 0.731, teacher questions 0.848 and hand-written hard set 0.859 against 0.766 (55 vs 49 of 64).
- JevBench public items: standard 0.958 and hard 0.775, with hard-tier ECE 0.071 against the 4B's 0.054.
- Runs the JevK5 runtime with CUDA graphs: about 16 ms per short decision on an otherwise idle H100, with roughly 19 GB needed for bf16.
- Leads the 4B by 8-10 points on questions with more than 16 options and scores 0.698 on bev-decision's test sample.
Watch out
Choosing 1.5 epochs was one seed per setting and each index-proxy source has only about 40 rows, so the sweep numbers are noisy. The 9B ties rather than beats the 4B on JevBench's hard tier (0.775 against 0.784, 7 fixed and 8 broken) and is 2-3x slower while needing about 19 GB, with worse hard-tier calibration.
Questions about this entry
Short answers drawn from the linked project or page.
The data is unchanged from v0.3. Training ran 1.5 epochs after a pre-registered sweep: adopt the epoch count with the best mean of index proxy, teacher accuracy and hard set only if it beat 1 epoch by at least 0.02 and nothing dropped by more than 0.03.
Index proxy 0.794 against 0.731, held-out teacher questions 0.848 against 0.834, and the hand-written hard set 0.859 against 0.766 (55 against 49 of 64). It also leads by 8-10 points on questions with more than 16 options.
JevBench's hard tier: 0.784 against the 9B's 0.775, with 7 items fixed and 8 broken, and hard-tier ECE 0.054 against 0.071. The 4B also needs about 9 GB instead of roughly 19 GB and is around 20% faster.
About 16 ms per short decision on an otherwise idle H100 through the JevK5 runtime 0.3.0 or later with CUDA graphs, against roughly 13 ms for the 4B. Latency on hard 1-4k-token documents is reported separately on the 4B card.
jevk5_config.json carries temperature 1.316 for every pass and knockout_temperature 1.05 for questions with more than 16 options, which need runtime 0.3.0 or later for the grouped readout.
Full runs with their own learning-rate schedules, same data, seed and dev set: the 9B at 1 epoch scored 0.762 index proxy, 1.5 scored 0.794 and 2 scored 0.784, so 1.5 was adopted; the 4B stayed at 1 epoch because 2 epochs fell below the +0.02 bar.
0.698 on 2,500 hashed test-split rows (v0.3's 9B 0.700, the v0.3 4B 0.663) with ECE 0.033. By question type: choice 0.732, yes/no 0.774 and score 0.482. Dropping 34 overlapping documents changes accuracy by 0.002.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
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 NewsHacker News: A one-token logprob wrapper that also classifies webcam frames.
allanrbo151 pointsA single function Jev-like wrapper for LLMs, including vision modelsRead the thread on Hacker NewsRelated terms
Glossary definitions related to this entry.
More Jev resources like this
alibiserikbay's Apache-2.0 Jev alternative on Qwen3.5-4B with a merged rank-16 LoRA: one forward pass returns a probability per option, 0.784 on JevBench's public hard tier at about 13 ms per decision.
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
GGUF builds of the JevK5 family for llama.cpp: Q8_0 is effectively lossless (229 of 231 answers match the 4B v0.3 bf16), and every published file clears the card's 95% agreement gate.
Jev Models & Reimplementations#community#open-models#local
Community
chaoliangUNSW's v1 Jev-Style decision model in GGUF from 1.3 GB: calibrated choice, bool and score answers on Qwen3.5-2B-Base, 82.3% on five held-out task families with ECE 0.017.
Jev Models & Reimplementations#community#open-models#local
Community
An Apple Silicon MLX bf16 build of the v1 Jev-Style decision model: 77 ms decisions on an M1 Max with calibrated choice, bool and score probabilities, Apache-2.0 weights.
Jev Models & Reimplementations#community#open-models#local
Community
Mithril's open, Jev-shaped typed-decision model on DeBERTa-v3-large: choice up to 255 options, score and noul from one forward pass, 0.854 in-domain with ECE 0.022, trained on public gold labels only.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
Training recipes, looped architectures, and language ports: what the open side of the typed-decision class shipped this month, and how to pick by runtime and calibration instead of the leaderboard.
Jev Models & Reimplementations#open-models#training#local
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Interface, size, license, claims — in that order. A selection guide across the system-one catalog, from a 706K form scorer to 27B Jev students.
Jev Practices & Patterns#open-models#evaluation#benchmarks
Read article
From the Jev community
Posts from builders shipping with Jev right now.