It was trained to adopt the persona of an empathetic and professional physician, capable of analyzing patient symptoms and structuring responses in a clinical format ("Assessment", "Plan", etc.).

💻 Hardware & Training

Hardware: Trained on 3x NVIDIA H100 80GB GPUs.
Base Model: Liquid LFM 1.2B (Lightweight, efficient architecture).
Dataset: UCSD Medical Dialogue (English).
Method: QLoRA / LoRA fine-tuning via trl and peft.

⚠️ Medical Disclaimer

This model is an AI research artifact and IS NOT a substitute for professional medical advice, diagnosis, or treatment. It has been trained on historical medical dialogues but can hallucinate or provide incorrect information. Always consult a qualified healthcare provider for personal medical needs.

🚀 Quick Usage

python

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# 1. Load Base Model
base_model = AutoModelForCausalLM.from_pretrained(
    "LiquidAI/LFM2.5-1.2B-Instruct",
    device_map="auto",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True
)

# 2. Load Adapter
adapter_id = "5ivatej/Liquid-LFM-1.2B-Medical-Doctor" 
model = PeftModel.from_pretrained(base_model, adapter_id)
model = model.merge_and_unload()

# 3. Run Inference
tokenizer = AutoTokenizer.from_pretrained("LiquidAI/LFM2.5-1.2B-Instruct")
messages = [{"role": "user", "content": "I have a headache."}]
input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt").to(model.device)

output = model.generate(input_ids, max_new_tokens=256)
print(tokenizer.decode(output[0], skip_special_tokens=True))