base_model: Qwen/Qwen3-4B
library_name: peft
license: apache-2.0
pipeline_tag: text-classification
tags:
- lora
- peft
- intent-classification
- quorum
extra_gated_prompt: >-
Quorum-Reader is shared for research and evaluation. Tell us a little about
your use — it helps us prioritise what to build next. Access is granted
automatically.
extra_gated_fields:
Name: text
Email: text
Organization: text
What are you building with this?: text
Quorum-Reader-Qwen3-4B-Adapter
A LoRA adapter for Qwen/Qwen3-4B — the in-context reader of the
Quorum 24-shot intent-classification ensemble.
Given 24 retrieved examples in the prompt (no weight update), it scores every candidate label in a single
forward pass. Combined (geometric mean) with a frozen BAAI/bge-large-en-v1.5 nearest-neighbour over the same
24 examples, the ensemble reaches 0.932 on Banking77 (full 3,080 test) — above the closed API Jev's 0.924
at Jev's own 24-shot protocol. In our study a further configuration (per-class descriptions + a
calibration-selected combination) reached 0.938 and significantly beat Jev on a paired McNemar test
(p = 0.0014).
Training
LoRA (r = 64) on Qwen/Qwen3-4B, fine-tuned on a mixture of public intent-classification datasets with
Banking77 held out. Single-forward answer-token option scoring.
Intended use & limitations
- Use as the reader in the Quorum 24-shot pipeline (see the repo). Requires the base model + PEFT to load.
- Contamination: the training mixture includes CLINC150, HWU64 and SNIPS — do not use this adapter to judge zero-shot generalization on those datasets.
- Not a general-purpose chat model; not a claim of open zero-shot state-of-the-art.
Usage
# https://github.com/DataAgent-Lab/Quorum
python scripts/eval_24shot.py --dataset banking77 --adapter DataAgent/Quorum-Reader-Qwen3-4B-Adapter
License
Adapter weights: Apache-2.0. The base model and datasets retain their own licenses.