# Chinese Decision Pointer 4B by quincyqiang / yanqiangmiffy

> Chinese Decision Pointer 4B scores supplied candidates with a pointer head. Read its public-suite results, unsupported request and data terms.

Updated 2026-11-12 (Australia/Adelaide)

## Quick answers

### What is Chinese Decision Pointer 4B by quincyqiang / yanqiangmiffy?

[Chinese Decision Pointer 4B](<https://decisionmodels.ai/models/chinese-decision-pointer>): A Qwen3.5-4B LoRA fine-tune with a pointer head that scores markers beside the supplied candidates.

[Chinese Decision Pointer 4B facts](<https://decisionmodels.ai/models/chinese-decision-pointer>)

[Source](<https://huggingface.co/quincyqiang/chinese-decision-pointer-4b>)

### Who makes Chinese Decision Pointer 4B?

[Chinese Decision Pointer 4B](<https://decisionmodels.ai/models/chinese-decision-pointer>): quincyqiang / yanqiangmiffy

[Chinese Decision Pointer 4B facts](<https://decisionmodels.ai/models/chinese-decision-pointer>)

[Source](<https://huggingface.co/quincyqiang/chinese-decision-pointer-4b>)

By quincyqiang / yanqiangmiffy · Hugging Face weight release: 2026-11-08 · OPEN

A Qwen3.5-4B LoRA fine-tune with a pointer head that scores markers beside the supplied candidates. Multiple questions about one state share a sequence. The pointer readout removes the letter-based 26-option cap, but all input still has to fit the runtime. The model selects among supplied options and has no abstention mechanism.

## Facts

- Base: Qwen/Qwen3.5-4B; merged BF16 LoRA
- Inputs: Text state and candidate questions; Chinese and English
- Readout: 256-dimensional pointer head; fitted temperature 1.538
- Licence: Apache-2.0 weights; source datasets retain their own terms, including non-commercial or research-only terms
- Access: Merged model.safetensors with loading and evaluation code

## Claims and sources

- **VENDOR-REPORTED**: The maker reports public Decision Index 0.3 skill 61.69 and raw 71.01 over 140,178 requests. The saved scores contain 140,177 ok responses and one unsupported response. Candidate-order debiasing averages original and reversed orders, so the reported 64.2 ms median and 147.3 ms mean on an A800 include that recipe. The maker describes 13-word overlap filtering; this is not an independent contamination audit. [source](<https://huggingface.co/datasets/quincyqiang/chinese-decision-index-results/blob/main/runs/chinese-decision-pointer-4b/scores.json>)

## Links

- [Source: Card and weights](<https://huggingface.co/quincyqiang/chinese-decision-pointer-4b>)
- [Source: Scores and request counts](<https://huggingface.co/datasets/quincyqiang/chinese-decision-index-results/blob/main/runs/chinese-decision-pointer-4b/scores.json>)
- [Source: Serving and training code](<https://github.com/yanqiangmiffy/chinese-decision>)

## Related projects

No related projects are listed in the registry.