You can deploy this model and start earning money today!

swnometal666/
finnish-healthcare-terminology-v3

Name: finnish-healthcare-terminology-v3
Availability: InStock

TurkuNLP/gpt3-finnish-small

adapter

apache-2.0

Views

No views yet

Finnish Healthcare Terminology Model v3

Model Description

This is a fine-tuned Finnish language model specialized for healthcare terminology and medical education. Built on TurkuNLP/gpt3-finnish-small using LoRA (Low-Rank Adaptation).

Training Details

Base Model: TurkuNLP/gpt3-finnish-small
Training Strategy: Ultra-conservative with early stopping
Dataset: 20 high-quality Finnish medical articles
Training Loss: 2.43 (optimal checkpoint)
Training Steps: 10 (early stopped at 16)
Learning Rate: 5e-6
LoRA Configuration: r=6, alpha=12

Intended Use

Educational healthcare content in Finnish
Medical terminology explanations
Health information chatbots
NOT for medical diagnosis or treatment advice

Model Performance

Superior medical terminology usage
Clean, coherent Finnish responses
No formatting artifacts
Outperforms base model in medical contexts

Usage Example

python

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# Load base model and tokenizer
base_model = AutoModelForCausalLM.from_pretrained("TurkuNLP/gpt3-finnish-small")
tokenizer = AutoTokenizer.from_pretrained("TurkuNLP/gpt3-finnish-small")

# Load fine-tuned model
model = PeftModel.from_pretrained(base_model, "swnometal666/finnish-healthcare-terminology-v3")

# Generate response
prompt = "Kysymys: Mitä tarkoittaa diabetes? Vastaus:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100, temperature=0.7)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Limitations

Designed for educational purposes only
Should not replace professional medical advice
May contain inaccuracies - always verify medical information
Optimized for Finnish language only

Training Data

Curated Finnish healthcare articles
Cleaned format without artifacts
Quality-focused selection (20 articles)

License

Apache 2.0