OncoLlama Assets

Assets for OncoLlama: Generating high fidelity synthetic cancer letters, and fine-tuning LLMs for structured data extraction.

Features

This package primarily exposes the OncoLlama schema for runtime output validation. Use it to:

• Validate LLM outputs against the expected OncoLlama structure
• Generate JSON schemas for API contracts and documentation
• Parse and validate extracted oncology data at runtime

Schema Access

Access the Pydantic model and its JSON schema:

from oncollama_assets.schema import OncoLlamaModel

# Get the JSON schema for validation
schema = OncoLlamaModel.model_json_schema()

# Parse and validate deserialised output
data = OncoLlamaModel.model_validate(llm_output)
# Parse and validate deserialised json string
data = OncoLlamaModel.model_validate_json(llm_output)

System Prompts

Load system prompts with the schema automatically injected:

from oncollama_assets.wrapper import OncoLlamaAssets

assets = OncoLlamaAssets()

# Load inference system prompt (default)
system_prompt = assets.load_system_prompt()

# Or specify a different prompt template
system_prompt = assets.load_system_prompt("systemprompt_finetune.md")

Available prompt templates:

• systemprompt_infer.md - For inference (default)
• systemprompt_finetune.md - For fine-tuning
• systemprompt_datagen.md - For data generation

Wrapper Class (Internal Use)

The OncoLlamaAssets wrapper class also provides testing and internal release mechanisms.

Structure

📁 ONCOLLAMA_ASSETS
├── prompts/             # Prompt templates
├── schema.py            # Pydantic model for specifying expected OncoLlama output structure
├── wrapper.py           # Wrapper class for internal testing and release mechanisms

License

This project uses a proprietary license (see LICENSE).