Xing-VL1.0-Decision
Quadrant scores
See the full quadrantScored with the public quadrant rubric (maturity × capability; bubble = adoption). Revise as evidence lands.
- Maturity4.5/10
Apache-2.0 bf16 weights on Qwen3.8-27B with a 255-way readout; custom Python module and a Decision Index engine plugin, ~60 GB GPU. Third-party GGUF (24 Oct). Small team, no hosted API or SLA.
- Capability7.8/10
Decision Index 0.4 Vision board (22 Oct) Full 73.3, #1 with no tie (d3-turbo v1.1 72.46); ECE 0.030 measured; 325 ms per row. Text + up to 4 images, 1–255 options, choice / yes-no / score. No text-board row.
- Adoption1/100
492 HF downloads and 3 likes (24 Oct); third-party GGUF quants. No integrations or production users found.
Vendor claims
- Decision Index edition 0.3, public vision suite (11 benchmarks, 12,210 rows): public vision index 77.78, every row answered[Vendor claim — not independently verified]
publicado (OK do Fabricio 24/10 BRT). Scores are calculated from scratch.
What it is
Xing-VL1.0-Decision (Hugging Face, Apache-2.0) is a 27B multimodal typed decision model from RBDash-Team on Qwen3.8-27B. Given a state (text and up to four images), one question and 1 to 255 options, it returns a probability for every option in a single forward pass through a 255-way readout that replaces the LM head; one fitted temperature (0.9766). It does not generate text. The card says “1.0” is a naming reset: it is the same checkpoint listed before as RBDash-Decision-v0.2, not retrained or re-scored.
Board result
| Board | Row | Result |
|---|---|---|
| Decision Index 0.4, Vision (data of 22 Oct, 20:10 UTC) | RBDash-Decision v0.2 | Full 73.3, #1 of 55 with no tie (vLLM-SR d3-turbo v1.1 72.46, Amos 1 Large 72.4); public 77.7, private 68.91; ECE 0.030; 325 ms per row |
The board caveat: at most 4 images per question; larger questions count as failures. There is no Jev reference row on the Vision board and no text-board row for this model. Boards are never combined.
Limits
- Vision-first; no text-board evidence.
- About 60 GB of GPU memory in bf16; needs its own module (
rbdash_decision.py), cannot load withAutoModelForCausalLM. The card warns that other prompt wording can change predictions.
Scores (from scratch, 24 Oct 2026)
- Capability 7.8 = types 10 × 0.30 + calibration 6 × 0.25 + latency 6 × 0.25 + accuracy 9 × 0.20 = 3.00 + 1.50 + 1.50 + 1.80 = 7.80. Types 10: choice / yes-no / score. Calibration 6: one temperature plus a board-measured ECE of 0.030. Latency 6: 325 ms per row (above 0.25 s). Accuracy 9: #1 with no tie on a Decision Index board.
- Alternative (accuracy 8 → 7.6): if the Vision board without a Jev reference does not count for the “#1 with no tie” exception.
- Maturity 4.5 = availability 7 × 0.30 + docs 6 × 0.25 + integrations 2 × 0.25 + support 2 × 0.20 = 2.10 + 1.50 + 0.50 + 0.40 = 4.50.
- Adoption 1 = engagement 1 × 0.35 + ecosystem 1 × 0.35 + production 0 × 0.30 = 0.70 → 1.