Submit a Model for Contract Clause Extractor
Extracts and classifies legal clauses from enterprise contracts
Performance Target
Harmonic mean of precision and recall
Token Reward
1,000 CLAUS per verified DeltaOne
Tokens minted per unit improvement in F1 Score
Register Your Model
After creating your model entry, register your trained model with the Hokusai registry using the SDK.
Install SDK
pip install git+https://github.com/Hokusai-protocol/hokusai-data-pipeline.git#subdirectory=hokusai-ml-platformConfigure API Key
export HOKUSAI_API_KEY="your-hokusai-api-key-here"Note: Use your Hokusai API key, not an MLflow token. The Hokusai API key authenticates both services.
Register Model
Register Contract Clause Extractor with token CLAUS
import os
import mlflow
from hokusai.core import ModelRegistry
# Set up MLflow tracking URI
mlflow.set_tracking_uri("https://registry.hokus.ai/api/mlflow")
# IMPORTANT: Use your Hokusai API key, NOT an MLflow token
# The Hokusai API key authenticates both the registry and MLflow
os.environ["MLFLOW_TRACKING_TOKEN"] = os.getenv("HOKUSAI_API_KEY")
# Initialize registry
registry = ModelRegistry()
# Register your model
with mlflow.start_run() as run:
# Log your model (replace with your actual model)
mlflow.sklearn.log_model(
your_trained_model,
"model",
registered_model_name="Contract Clause Extractor"
)
# Register with Hokusai
model_uri = f"runs:/{run.info.run_id}/model"
registered_model = registry.register_tokenized_model(
model_uri=model_uri,
model_name="Contract Clause Extractor",
token_id="CLAUS",
metric_name="accuracy",
baseline_value=0.92,
additional_tags={"author": "your-name", "version": "1.0"}
)
print(f"✅ Model registered successfully: {registered_model.name}")Trigger Evaluation
Once your model is registered, trigger an evaluation run against the benchmark dataset. The system will automatically measure your model's performance against the target criteria.
Evaluation process:
- Your model runs on the specified evaluation dataset
- Performance is measured using F1 Score
- Results are verified and recorded on-chain
- Token rewards are calculated based on performance improvement
Claim Your Rewards
If your model meets or exceeds the benchmark target, you'll earn token rewards. Tokens are automatically minted and can be claimed from your dashboard.
Evaluation Timeline
Evaluation runs typically complete within 5-30 minutes, depending on model complexity and dataset size. You'll receive notifications when evaluation completes.
Performance Verification
All evaluation results are verified and recorded on-chain to ensure transparency and prevent manipulation. Your model's performance will be publicly visible.
Token Distribution
Tokens are minted automatically when your model achieves performance improvements. The amount is calculated based on the delta between your model's performance and the baseline, multiplied by the tokens-per-delta-one rate.
Support & Documentation
Need help? Visit our model submission guide for detailed instructions, code examples, and troubleshooting tips.