This model is a fine-tuned version of TurkuNLP/gpt3-finnish-medium, specialized for Finnish healthcare policy and administration (not clinical diagnosis).

Model Focus

This model excels at understanding and generating content about:

Hyvinvointialueet (Wellbeing areas) and their administration
Sote-uudistus (Healthcare reform) impacts and implementation
Kela-korvausjärjestelmä (Kela compensation systems)
Hoitotakuu (Treatment guarantee) implementation
Terveydenhuollon digitalisaatio (Healthcare digitalization)
Työterveyshuolto (Occupational health services)
Healthcare financing and administration policies

Performance Improvements

Based on comparative testing against the base model:

Text diversity: +6.5% improvement (0.907 vs 0.851)
Healthcare terminology usage: +25.0% improvement (20 vs 16 terms)
Domain knowledge: Significantly better responses on healthcare policy topics

Training Configuration

Base Model: TurkuNLP/gpt3-finnish-medium (~350M parameters)
Fine-tuning Method: LoRA (Low-Rank Adaptation)
Training Data: 50-60 Finnish healthcare policy articles
Training Steps: 45 steps with improved learning rate schedule
Learning Rate: 2e-5 with 20-step warmup
LoRA Config: r=6, alpha=32, targeting ["q_proj", "k_proj", "v_proj", "o_proj"]

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load the model
tokenizer = AutoTokenizer.from_pretrained("TurkuNLP/gpt3-finnish-medium")
base_model = AutoModelForCausalLM.from_pretrained("TurkuNLP/gpt3-finnish-medium")
model = PeftModel.from_pretrained(base_model, "swnometal666/finnish-healthcare-policy-v5")

# Generate healthcare policy content
prompt = "Kysymys: Miten hyvinvointialueiden rahoitusmalli toimii käytännössä? 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)

Example Outputs

Prompt: "Kysymys: Mitä hyvinvointialueet tarkoittavat suomalaisessa terveydenhuoltojärjestelmässä?"

Model Response: "Hyvinvointialueiden tehtävä on turvata kansalaisten terveys- ja sosiaalipalvelut sekä tuottaa laadukkaita sosiaali- ja terveyspalveluja. Ne vastaavat myös hyvinvointialueen palvelujen järjestämisestä, taloudesta ja rahoituksesta..."

Important Limitations

⚠️ This model is NOT for clinical use!

This model is designed for:

✅ Healthcare policy and administration content
✅ Educational material about Finnish healthcare system
✅ Understanding healthcare reform impacts

This model should NOT be used for:

❌ Medical diagnosis or treatment recommendations
❌ Clinical decision-making
❌ Emergency medical situations
❌ Personal health advice

Always consult qualified healthcare professionals for medical advice.

Training Methodology

Conservative approach: Early stopping at 45 steps to prevent overfitting
Quality over quantity: Focused on 50-60 high-quality policy articles
Learning rate warmup: Gradual LR increase for training stability
Healthcare policy focus: Domain-specific terminology and concepts