How to Intermediate · 3 min read

Fix DSPy optimization not improving

Q: Fix DSPy optimization not improving

To fix dspy optimization not improving, ensure you configure the LM instance with the correct model and API key, use the dspy.configure() method properly, and run training or prediction with the right signature and input. Also, verify your training loop calls qa(question="...") instead of qa.predict(), as the latter is deprecated and can cause no optimization progress.

Quick answer

To fix dspy optimization not improving, ensure you configure the LM instance with the correct model and API key, use the dspy.configure() method properly, and run training or prediction with the right signature and input. Also, verify your training loop calls qa(question="...") instead of qa.predict(), as the latter is deprecated and can cause no optimization progress.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install dspy openai>=1.0

Setup

Install dspy and openai packages, and set your OpenAI API key as an environment variable.

Run pip install dspy openai
Set environment variable OPENAI_API_KEY with your API key

bash

pip install dspy openai

Step by step

Use the correct dspy pattern for optimization by configuring the LM, defining a signature, and invoking prediction properly.

python

import os
import dspy
from openai import OpenAI

# Initialize LM with OpenAI GPT-4o-mini
lm = dspy.LM("openai/gpt-4o-mini", api_key=os.environ["OPENAI_API_KEY"])
dspy.configure(lm=lm)

# Define a signature for QA task
class QA(dspy.Signature):
    question: str = dspy.InputField()
    answer: str = dspy.OutputField()

# Create a prediction module
qa = dspy.Predict(QA)

# Run prediction with correct call (not qa.predict)
result = qa(question="What is retrieval-augmented generation?")
print("Answer:", result.answer)

output

Answer: Retrieval-augmented generation (RAG) is a technique that combines retrieval of relevant documents with generative models to produce more accurate and context-aware responses.

Common variations

You can use different models by changing the LM initialization string, or run asynchronous calls with dspy if supported. Avoid deprecated methods like qa.predict() which do not trigger optimization.

python

import asyncio

async def async_example():
    result = await qa.acall(question="Explain chain-of-thought prompting.")
    print("Async answer:", result.answer)

asyncio.run(async_example())

output

Async answer: Chain-of-thought prompting is a technique where the model is guided to reason step-by-step, improving complex problem solving.

Troubleshooting

If optimization does not improve, confirm you are using qa(question="...") instead of qa.predict().
Check your API key is valid and environment variable is set.
Ensure dspy.configure(lm=lm) is called before using the signature.
Verify your model string matches a supported model like openai/gpt-4o-mini.

✅

Key Takeaways

Always configure dspy with a valid LM instance before use.
Use qa(question="...") for prediction to enable optimization, not qa.predict().
Check environment variables and model names to avoid silent failures.
Async calls require await and acall method usage.
Deprecated methods can cause no optimization progress; update your code accordingly.

Verified 2026-04 · openai/gpt-4o-mini

Verify ↗