PharmaQA‑1.2B is a merged, instruction-tuned pharmacology and pharmacy domain language model based on Liquid AI LFM2-1.2B. It was fine-tuned using the MIRIAD-4.4M dataset for research and educational Q&A in pharmacology, therapeutics, and drug mechanisms. This model is not intended for clinical or diagnostic use.

🧪 Model Details

Property	Value
Base Model	Liquid AI `LFM2-1.2B`
Fine-tuning Method	LoRA using Unsloth
Parameters Trained	~9M (0.78% of total)
Dataset Used	MIRIAD-4.4M (subset of 50,000 examples)
Epochs	1
Final Format	Merged (LoRA + base)
Model Size	1.2 Billion
License	ODC-BY v1.0 dataset license, non-commercial educational use only
Author	Mohamed Yasser

⚠️ Disclaimer

This model is not intended for medical diagnosis, treatment planning, or patient care. It was trained on synthetic Q&A pairs from peer-reviewed literature via MIRIAD and is for educational and academic research only.

MIRIAD includes a cautionary note that aligns with OpenAI’s usage policies:

Do not use this dataset or models trained on it for actual medical diagnosis, decision-making, or any application involving real-world patients.

Example Usage

python

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, TextStreamer

model_name = "yasserrmd/PharmaQA-1.2B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")
model.eval()

# Example pharmacy-related question
question = "What is the mechanism of action of metformin?"

# Format as chat message
messages = [{"role": "user", "content": f"Q: {question} A:"}]

# Tokenize with chat template
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt",
    tokenize=True,
    return_dict=True,
).to(model.device)

# Clean input if necessary
if "token_type_ids" in inputs:
    del inputs["token_type_ids"]

# Generate the answer
with torch.no_grad():
    output_ids = model.generate(
        **inputs,
        max_new_tokens=128,
        temperature=0.3,
        min_p=0.15,
        repetition_penalty=1.05
    )

# Decode the response
response = tokenizer.decode(output_ids[0], skip_special_tokens=True).strip()
answer = response.split("A:")[-1].strip()

print("💊 Question:", question)
print("🧠 Answer:", answer)

Performance Insights

From manual analysis of 50 unseen pharmacology questions:

✅ No hallucinations observed
✅ High alignment with biomedical terms (e.g., dihydrofolate reductase, QT prolongation)
✅ Long-form answers are clinically descriptive and accurate for education
⚠️ Short answers are concise but can lack therapeutic context

License

Model: Educational and research use only
Dataset: MIRIAD (ODC-BY v1.0)

Acknowledgements

MIRIAD Team (for the dataset)
Unsloth team (for fast & efficient LoRA)
Hugging Face + Liquid AI for open model access

Let me know if you'd like a Markdown file for this (README.md) or want help preparing the Hugging Face push commands.