[python][dspy] instrument module

[mikeldking]

Instrument module so that the forward passes are captured

[dosubot]

🤖

To instrument a module in dspy so that the forward passes are captured, you can use the DSPyInstrumentor class from the openinference.instrumentation.dspy module. This class provides the instrument() method which you can call to instrument your dspy module.

Here is a basic example of how you can use it:

from openinference.instrumentation.dspy import DSPyInstrumentor

# ...

DSPyInstrumentor().instrument()

In the provided code, the DSPyInstrumentor().instrument() line is responsible for instrumenting the dspy module. After this line, any forward passes in the dspy module will be captured and traced.

The DSPyInstrumentor().instrument() method in the OpenInference codebase is used to instrument or add tracing to the DSPy (Data Science Python) code. This is done by wrapping certain function calls with wrappers that start and end spans, which are units of work within a trace.

The method does the following:

1. It first checks if a tracer_provider is provided in the arguments. If not, it gets the default tracer_provider using trace_api.get_tracer_provider(). A tracer is then obtained using this tracer_provider.

2. It instruments the basic_request method of all subclasses of the LM (Language Model) class from the dsp.modules.lm module. This is done by wrapping the basic_request method with the _LMBasicRequestWrapper which starts a span, calls the original method, and then ends the span.

3. It also instruments the forward method of the Predict class from the dspy module. This is done by wrapping the forward method with the _PredictForwardWrapper which also starts a span, calls the original method, and then ends the span.

The _LMBasicRequestWrapper and _PredictForwardWrapper classes are defined in the same file. They both inherit from the _WithTracer class, which provides them with a tracer object. These wrappers are used to start a span before the wrapped function is called and end the span after the function call. They also handle exceptions by recording them in the span and re-raising them. They also set certain attributes in the span such as the input and output values and their MIME types.

You can find more details in the OpenInference repository.

Sources

• python/instrumentation/openinference-instrumentation-dspy/examples/dspy_basic_qa.py
• python/instrumentation/openinference-instrumentation-dspy/src/openinference/instrumentation/dspy/init.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.