🦴 Anatomy of Feedback Functions¶

The Feedback class contains the starting point for feedback function specification and evaluation.

Example

# Context relevance between question and each context chunk.
f_context_relevance = (
    Feedback(
        provider.context_relevance_with_cot_reasons,
        name="Context Relevance"
    )
    .on(Select.RecordCalls.retrieve.args.query)
    .on(Select.RecordCalls.retrieve.rets)
    .aggregate(numpy.mean)
)

The components of this specifications are:

Feedback Providers¶

The provider is the back-end on which a given feedback function is run. Multiple underlying models are available througheach provider, such as GPT-4 or Llama-2. In many, but not all cases, the feedback implementation is shared cross providers (such as with LLM-based evaluations).

Feedback implementations¶

OpenAI.context_relevance is an example of a feedback function implementation.

Feedback implementations are simple callables that can be run on any arguments matching their signatures. In the example, the implementation has the following signature:

Example

def context_relevance(self, prompt: str, context: str) -> float:

That is, context_relevance is a plain python method that accepts the prompt and context, both strings, and produces a float (assumed to be between 0.0 and 1.0).

Read more about feedback implementations

Feedback constructor¶

The line Feedback(openai.relevance) constructs a Feedback object with a feedback implementation.

Argument specification¶

The next line, on_input_output, specifies how the context_relevance arguments are to be determined from an app record or app definition. The general form of this specification is done using on but several shorthands are provided. For example, on_input_output states that the first two argument to context_relevance (prompt and context) are to be the main app input and the main output, respectively.

Read more about argument specification and selector shortcuts.

Aggregation specification¶

The last line aggregate(numpy.mean) specifies how feedback outputs are to be aggregated. This only applies to cases where the argument specification names more than one value for an input. The second specification, for statement was of this type. The input to aggregate must be a method which can be imported globally. This requirement is further elaborated in the next section. This function is called on the float results of feedback function evaluations to produce a single float. The default is numpy.mean.