File size: 3,838 Bytes
2834fbd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d2b4cd8
2834fbd
 
 
 
 
 
 
 
 
 
 
 
 
 
d2b4cd8
2834fbd
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
---

license: apache-2.0
base_model: Qwen/Qwen3-0.6B
tags:
  - medical
  - mcq
  - classification
  - radiology
  - qwen3
  - medmcq
  - stravoris
pipeline_tag: text-generation
language:
  - en
---


# MedMCQ — Radiology Topic Classifier (Qwen3-0.6B)

A small fine-tuned Qwen3 model that takes a **Radiology** medical multiple-choice question (MCQ) and predicts which **topic** within Radiology the question is about.

This is a **per-subject topic classifier** — the second hop in the [MedMCQ three-hop pipeline](#the-medmcq-pipeline). It assumes the input MCQ has already been routed to Radiology by the [subject classifier](https://huggingface.co/stravoris/medmcq-subject-classifier-qwen3-0.6b).

## The MedMCQ pipeline

The MedMCQ project explores small, specialized models for medical MCQs. Instead of using one large model for everything, it splits the task across three hops:

1. **Subject routing** — the [subject classifier](https://huggingface.co/stravoris/medmcq-subject-classifier-qwen3-0.6b) picks the medical subject.
2. **Topic classification** — *this model.* Given a Radiology MCQ, pick the topic within Radiology.
3. **Answer generation** — the [Radiology generator](https://huggingface.co/stravoris/medmcq-radiology-qwen3-1.7b) produces the answer / new MCQs in Radiology.

Each hop is a separate, narrow model. They are all published under the [MedMCQ Medical Models](https://huggingface.co/collections/stravoris/medmcq-medical-models) collection.

## Quick start

```python

from transformers import AutoTokenizer, AutoModelForCausalLM



repo = "stravoris/medmcq-radiology-classifier-qwen3-0.6b"

tokenizer = AutoTokenizer.from_pretrained(repo)

model = AutoModelForCausalLM.from_pretrained(repo)



prompt = """Classify the following Radiology MCQ by topic.



Question: <a Radiology MCQ here>

A) <option A>

B) <option B>

C) <option C>

D) <option D>



Topic:"""



inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=16, do_sample=False)

print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

```

## Prompt format

The model expects MCQs in this exact form:

```

Classify the following Radiology MCQ by topic.



Question: <question text>

A) <option A>

B) <option B>

C) <option C>

D) <option D>



Topic:

```

The model completes the prompt with one of the topic labels seen during training. The full topic list is in the [dataset card](https://huggingface.co/datasets/stravoris/medical-mcq-dataset).

## What this model is not

This is a **sample model for demonstration**. It is not a production-grade medical AI system:

- It has not been formally evaluated against board-level benchmarks.
- It should not be used to make clinical decisions or provide medical advice.
- It is narrow: it only understands Radiology MCQs and assumes the upstream subject router did its job.
- It was trained on a curated educational dataset and inherits any biases or gaps in that data.

The MedMCQ project exists to explore small-model pipeline architectures for medical reasoning, not to ship a medical product.

## Training data

Trained on the Radiology subset of the [Stravoris Medical MCQ dataset](https://huggingface.co/datasets/stravoris/medical-mcq-dataset) — educational Radiology MCQs labeled by topic.

## Base model

Fine-tuned from [`Qwen/Qwen3-0.6B`](https://huggingface.co/Qwen/Qwen3-0.6B). The base model's license and usage terms also apply.

## License

Apache 2.0. See [LICENSE](https://www.apache.org/licenses/LICENSE-2.0).

## Collection

Part of the [MedMCQ Medical Models](https://huggingface.co/collections/stravoris/medmcq-medical-models) collection — all 31 models that make up the MedMCQ three-hop pipeline.