Model Card for Model ID
Model Details
Model Description
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- Developed by: Yap Yung Yoong
- Model type: Causal Language Model (decoder-only Transformer)
- Language(s) (NLP): English
- Base model: Qwen/Qwen3-0.6B
- License: Apache-2.0 (inherits base model + dataset license constraints)
- Finetuned from model: Qwen/Qwen3-0.6B
- Training method: Supervised Fine-Tuning (SFT) using TRL
SFTTrainer
Model Sources [optional]
Uses
Direct Use
- Medical Q&A style assistance (educational / research)
- Producing structured answers (e.g., differential diagnosis, suggested tests, general management)
- Demonstrations of instruction-tuning / CoT-style supervision
Out-of-Scope Use
- Not for real-world clinical diagnosis or treatment decisions
- Not for emergency/urgent medical advice
- Not designed for legal, financial, or safety-critical decisions
- Not robust against adversarial prompts or hallucination-sensitive tasks
Bias, Risks, and Limitations
- The model may produce hallucinated medical facts, diagnoses, or treatment plans.
- The model may reflect biases present in the dataset (English-centric, guideline assumptions, population coverage).
- The model may generate confident outputs even when uncertain.
- If trained with explicit
<think> reasoning fields, it may over-generate reasoning text or follow a rigid format.
Recommendations
- Use only for educational/research purposes.
- For product use: add guardrails (triage warnings, refusal for emergencies, citation requirements, retrieval, and human review).
- Evaluate on held-out sets and real user queries; do not assume training loss implies safety or correctness.
Training Details
Training Data
Dataset: FreedomIntelligence/medical-o1-reasoning-SFT (subset: en)
Split used: train[0:2000] (example; update if different)
Fields used: Question, Complex_CoT, Response
Formatting: Data was converted into a single text field using an instruction template and appending EOS token.
Training Hyperparameters
- Trainer: TRL SFTTrainer
- Optimizer: (e.g., AdamW / paged_adamw_32bit)
- Learning rate: (e.g., 2e-4)
- LoRA / PEFT: YES
Speeds, Sizes, Times [optional]
Training steps: 1000
Training loss: 1.6495
Results
Training loss: 1.6495 (1000 steps)
Environmental Impact
Carbon emissions can be estimated using the
Machine Learning Impact calculator presented in
Lacoste et al. (2019).
- Hardware Type: Tesla T4
- Hours used: 3
- Cloud Provider: AWS
- Compute Region: Singapore
- Carbon Emitted: 0.09
Citation [optional]
Qwen3 base model: Qwen/Qwen3-0.6B
Dataset: FreedomIntelligence/medical-o1-reasoning-SFT
