RSI-Jev v7.0-VL 4B decision model
RSI-Jev v7.0-VL 4B returns typed decisions on text and images. Read its early exits, public weights, reported scores and serving limits.
By Shanghua Gao · Hugging Face package release: 5 Nov 2026
A Qwen3.5-4B decision fine-tune with text exits at layers 16, 20 and 32. It reads a document and up to four images, then returns option probabilities without generating text. Image questions always use all 32 layers. Its own serving package supplies the option echo and bidirectional attention required by the weights.
Quick answers
What is RSI-Jev v7.0-VL 4B by Shanghua Gao?
RSI-Jev v7.0-VL 4B: A Qwen3.5-4B decision fine-tune with text exits at layers 16, 20 and 32.
Who makes RSI-Jev v7.0-VL 4B?
RSI-Jev v7.0-VL 4B: Shanghua Gao
Are RSI-Jev v7.0-VL 4B's weights open?
RSI-Jev v7.0-VL 4B: Public weights with the RSI-Jev runtime; no causal prefix cache Licence: Apache-2.0
Facts
- Size
- 4.69B parameters; about 9.7 GB packaged weights
- Inputs
- Text and up to four images; choice, noul and score
- Licence
- Apache-2.0
- Limits
- 32,768 tokens; up to 5,120 options; overlength requests get HTTP 422
- Access
- Public weights with the RSI-Jev runtime; no causal prefix cache
Claims and sources
- The maker reports 60.63 on the public Decision Index 0.3, with 29.4 ms median and 236 ms p95 on one RTX PRO 6000 Blackwell at default effort. This is a public-suite run, not the full private index. Training used one seed and unreleased data. Ambiguous KoBBQ questions and some medicine and customer-support tasks regress against v6.1-VL. vendor reportedsource
Links
Source: Model card and files →
Related projects
No related projects are listed in the registry.