​

Running Collinear Guard And Collinear Guard Nano

​

Parameters

model_name: Allowed values are collinear_guard and collinear_guard_classifier

nano_model_type: This is required only when model_name is collinear_guard_classifier. Allowed values are prompt, response, and refusal.

scoring_criteria: This field is optional. Only use it if custom scoring criteria is required

• This should be an array of two elements when using collinear_guard_classifier. The possible values for score are 0 and 1.
• This should be an array of five elements when using collinear_guard. The possible values for score are 1,2,3,4,5.

conversation: This is the conversation prefix in OpenAI format. It should be a list of dictionaries with keys role and content.

response: This is the response from the model. It should be in OpenAI format.

​

Response

The response will be a JSON object with the following fields:

​

Collinear Guard Nano

judgement : The safety judgement of the model output. The possible values are

• 1: Which means the output is safe/compliant
• 0: Which means the output is unsafe/refusal

​

Collinear Guard

judgement : The safety judgement of the model output. The possible values are

• 1: Very unsafe
• 2: Unsafe
• 3: Neutral
• 4: Safe
• 5: Very safe

extra: The exact judge output with rationale