Inference

Use your trained model for predictions. Two options: the convenience method lr.predict() or the OpenAI-compatible API directly.

lr.predict()

Single prediction with a trained model:

response = lr.predict(job.model_id, "Will the Fed cut rates by 25bp in March 2026?")
print(response)

Parameter	Type	Default	Description
model_id	str	—	Your trained model ID
prompt	str	—	The question or prompt text
system_prompt	str	"Answer as a probability between 0 and 1 between <answer></answer> tags."	System message
**kwargs	—	—	Passed to openai.chat.completions.create

Requires pip install openai (uses Lightning Rod's API key and base URL automatically).

OpenAI-compatible API

Use the OpenAI client with Lightning Rod's base URL for full control:

from openai import OpenAI

client = OpenAI(
    api_key=api_key,
    base_url="https://api.lightningrod.ai/api/public/v1/openai",
)

response = client.chat.completions.create(
    model=job.model_id,
    messages=[
        {"role": "system", "content": "Answer as a probability between 0 and 1 between <answer></answer> tags."},
        {"role": "user", "content": "Will the Fed cut rates by 25bp in March 2026?"},
    ],
)
print(response.choices[0].message.content)

Model availability

Trained model checkpoints are available for 7 days after training completes. For long-term hosting, contact [email protected].