medassist-qa

Medassist-QA – LoRA Fine-tuned Medical Assistant

Medassist-QA is a multi-turn fine-tuned model designed to answer healthcare and medical-related queries with high relevance.
It was built as part of a real-world Generative AI Engineer project by Raghuramreddy Thirumalareddy.

---

<details> <summary><strong>Intended Use & Limitations</strong></summary>

Intended Use

• Educational and research purposes in healthcare Q&A
• Demo of fine-tuning with LoRA/PEFT for domain-specific assistants

Limitations

• ❌ Not a replacement for professional medical advice
• ❌ Should not be used for emergency or clinical decision-making

</details>

---

<details> <summary><strong>Real-World Use Cases</strong></summary>

• Patient self-assistance chatbots
• Quick orientation for clinics & hospitals
• Healthcare FAQ for institutions
• Medical record summarization pipelines

</details>

---

<details> <summary><strong>Model Training Details</strong></summary>

• Base Model: tiiuae/falcon-7b-instruct
• Technique: LoRA using PEFT (Hugging Face)
• Dataset: Healthcare FAQs + symptom-based dialogues (curated & synthetic)
• Epochs: 3 | LoRA Rank: 8
• Hardware: Trained on local RTX 4060 (8GB VRAM)

</details>

---

<details> <summary><strong>Quick Start Code</strong></summary>

from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

# Load model & tokenizer
model = AutoModelForCausalLM.from_pretrained("RaghuramReddyT/medassist-qa")
tokenizer = AutoTokenizer.from_pretrained("RaghuramReddyT/medassist-qa")

# Create pipeline
qa = pipeline("text-generation", model=model, tokenizer=tokenizer)

# Example query
print(qa("What are common symptoms of dengue fever?", max_new_tokens=100))

</details> <details> <summary><strong>Included Files</strong></summary>

adapter_model.safetensors – Fine-tuned LoRA weights

adapter_config.json – LoRA adapter setup

tokenizer_config.json, vocab.json, merges.txt – Tokenizer files

</details> <details> <summary><strong>Credits</strong></summary>

Author: Raghuramreddy Thirumalareddy

Part of my Generative AI Engineer project's

Training support: Hugging Face PEFT

</details>