BiMediX/BiMediX-Ara

Model Card for BiMediX-Bilingual

Model Details

• Name: BiMediX
• Version: 1.0
• Type: Bilingual Medical Mixture of Experts Large Language Model (LLM)
• Languages: Arabic
• Model Architecture: Mixtral-8x7B-Instruct-v0.1
• Training Data: BiMed1.3M-Arabic, an arabic dataset with diverse medical interactions.

Intended Use

• Primary Use: Medical interactions in both English and Arabic.
• Capabilities: MCQA, closed QA and chats.

Getting Started

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "BiMediX/BiMediX-Ara"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

text = "مرحبًا بيميديكس! لقد كنت أعاني من التعب المتزايد في الأسبوع الماضي."
inputs = tokenizer(text, return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=500)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training Procedure

• Dataset: BiMed1.3M-Arabic.
• QLoRA Adaptation: Implements a low-rank adaptation technique, incorporating learnable low-rank adapter weights into the experts and the routing network. This results in training about 4% of the original parameters.
• Training Resources: The model underwent training on the Arabic corpus.

Model Performance

Model	CKG	CBio	CMed	MedGen	ProMed	Ana	MedMCQA	MedQA	PubmedQA	AVG
Jais-30B	52.1	50.7	40.5	49.0	39.3	43.0	37.0	28.8	74.6	46.1
BiMediX (Arabic)	60.0	54.9	55.5	58.0	58.1	49.6	46.0	40.2	76.6	55.4
BiMediX (Bilingual)	63.8	57.6	52.6	64.0	52.9	50.4	49.1	47.3	78.4	56.5

Safety and Ethical Considerations

• Potential issues: hallucinations, toxicity, stereotypes.
• Usage: Research purposes only.

Accessibility

• Availability: BiMediX GitHub Repository.
• arxiv.org/abs/2402.13253

Authors

Sara Pieri, Sahal Shaji Mullappilly, Fahad Shahbaz Khan, Rao Muhammad Anwer Salman Khan, Timothy Baldwin, Hisham Cholakkal
Mohamed Bin Zayed University of Artificial Intelligence (MBZUAI)