OPEN
JAD-S1-7B-A1B v0.3 decision model
JAD-S1 v0.3 is a LLaDA-MoE decision adapter with public weights. Read its option limits, reported JevBench results and abstention weaknesses.
By DreamBlooms · Hugging Face package release: 5 Nov 2026
A LoRA decision adapter on LLaDA-MoE-7B-A1B-Instruct. One masked forward pass returns a probability distribution over option letters, with no generated answer. The base weights must be downloaded separately. Use its bundled inference script for the recorded scores; it repairs a rotary-embedding compatibility issue.
Quick answers
What is JAD-S1-7B-A1B v0.3 by DreamBlooms?
JAD-S1-7B-A1B v0.3: A LoRA decision adapter on LLaDA-MoE-7B-A1B-Instruct.
Who makes JAD-S1-7B-A1B v0.3?
Facts
- Base
- LLaDA-MoE-7B-A1B-Instruct; separate base download
- Licence
- Apache-2.0 adapter, code and base
- Limits
- 2–26 letter-labelled options, six recommended; 4,096 tokens recommended, up to 8,192 supported
- Access
- Public LoRA adapter and inference.py
Claims and sources
- The maker reports 195 of 231 public JevBench v1.2.2 tasks correct (84.4%), versus 155 for the base using the same script. Published competitor rows use different serving protocols. On 66 evidence-silent VitaminC cases, it selects “not enough info” for only 16. Calibration fitted on English decision data needs checking on other tasks. vendor reportedsource
Links
Source: Model card and files →
Related projects
No related projects are listed in the registry.