102 lines
2.6 KiB
Markdown
102 lines
2.6 KiB
Markdown
|
---
|
|||
|
|
library_name: transformers
|
|||
|
|
tags:
|
|||
|
|
- sft
|
|||
|
|
- unsloth
|
|||
|
|
- trl
|
|||
|
|
- medical
|
|||
|
|
license: apache-2.0
|
|||
|
|
datasets:
|
|||
|
|
- medalpaca/medical_meadow_medical_flashcards
|
|||
|
|
language:
|
|||
|
|
- en
|
|||
|
|
base_model:
|
|||
|
|
- unsloth/Qwen3-0.6B
|
|||
|
|
pipeline_tag: text-generation
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
# Model Card for Medical-QA
|
|||
|
|
|
|||
|
|
|
|||
|
|
## Model Details
|
|||
|
|
|
|||
|
|
This model is a fine-tuned version of Qwen3-0.6B on a 34K medical Q&A dataset derived from the Anki Medical Curriculum flashcards.
|
|||
|
|
It is designed to assist with medical education and exam preparation, offering concise and contextually relevant answers to short medical questions.
|
|||
|
|
|
|||
|
|
- **Base Model:** Qwen3-0.6B
|
|||
|
|
- **Fine-tuned on:** 34,000 question-answer pairs
|
|||
|
|
- **Domain:** Medicine & Medical Education
|
|||
|
|
- **Languages:** English
|
|||
|
|
- **License:** MIT
|
|||
|
|
|
|||
|
|
|
|||
|
|
## Uses
|
|||
|
|
|
|||
|
|
### Direct Use
|
|||
|
|
|
|||
|
|
- Primary use case: Medical Q&A for students, exam preparation, and knowledge review.
|
|||
|
|
- Suitable for interactive learning assistants or educational chatbots.
|
|||
|
|
- Not intended for real-world clinical decision-making or replacing professional medical advice.
|
|||
|
|
|
|||
|
|
|
|||
|
|
## Bias, Risks, and Limitations
|
|||
|
|
|
|||
|
|
- The model’s knowledge is constrained to the dataset scope (flashcard-style Q&A).
|
|||
|
|
- Responses are short and exam-style rather than detailed clinical explanations.
|
|||
|
|
- Should not be relied upon for actual patient care, treatment decisions, or emergency use.
|
|||
|
|
|
|||
|
|
## How to Get Started with the Model
|
|||
|
|
|
|||
|
|
Use the code below to get started with the model.
|
|||
|
|
```python
|
|||
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM
|
|||
|
|
|
|||
|
|
tokenizer = AutoTokenizer.from_pretrained("khazarai/Medical-QA")
|
|||
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|||
|
|
"khazarai/Medical-QA",
|
|||
|
|
device_map={"": 0}
|
|||
|
|
)
|
|||
|
|
|
|||
|
|
system = "Answer this question truthfully"
|
|||
|
|
|
|||
|
|
question = """
|
|||
|
|
What can β-blockers cause or exacerbate due to excessive AV nodal inhibition?
|
|||
|
|
"""
|
|||
|
|
|
|||
|
|
messages = [
|
|||
|
|
{"role" : "system", "content" : system},
|
|||
|
|
{"role" : "user", "content" : question}
|
|||
|
|
]
|
|||
|
|
text = tokenizer.apply_chat_template(
|
|||
|
|
messages,
|
|||
|
|
tokenize = False,
|
|||
|
|
add_generation_prompt = True,
|
|||
|
|
enable_thinking = False,
|
|||
|
|
)
|
|||
|
|
|
|||
|
|
from transformers import TextStreamer
|
|||
|
|
_ = model.generate(
|
|||
|
|
**tokenizer(text, return_tensors = "pt").to("cuda"),
|
|||
|
|
max_new_tokens = 512,
|
|||
|
|
temperature = 0.7,
|
|||
|
|
top_p = 0.8,
|
|||
|
|
top_k = 20,
|
|||
|
|
streamer = TextStreamer(tokenizer, skip_prompt = True),
|
|||
|
|
)
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
|
|||
|
|
## Training Details
|
|||
|
|
|
|||
|
|
### Training Data
|
|||
|
|
|
|||
|
|
The dataset is based on Anki Medical Curriculum flashcards, created and updated by medical students.
|
|||
|
|
These flashcards cover the entire medical curriculum, including but not limited to:
|
|||
|
|
|
|||
|
|
- Anatomy
|
|||
|
|
- Physiology
|
|||
|
|
- Pathology
|
|||
|
|
- Pharmacology
|
|||
|
|
- Clinical knowledge and skills
|
|||
|
|
|
|||
|
|
The flashcards typically provide succinct summaries and mnemonics to support learning and retention.
|