dspy-two-step-adapter

name

dspy-two-step-adapter

description

Use when working with reasoning models (o1, o3, o3-mini, DeepSeek-R1, Claude extended thinking) that reject system prompts or ignore formatting instructions. Common scenarios - using o1 or o3 with DSPy, getting structured output from reasoning models, two-phase prompting where a reasoning model generates freely then an extraction model parses the output, or fixing format errors from thinking models. Related - dspy-adapters, dspy-lm, ai-switching-models. Also used for dspy.TwoStepAdapter, o1 with DSPy, o3-mini DSPy, reasoning model in DSPy, DeepSeek-R1 DSPy, extended thinking DSPy, thinking model ignores format, o1 ignores my DSPy format, TwoStepAdapter setup, two-phase prompting, extraction LM, reasoning model structured output, extraction_model TwoStepAdapter, model pairing for reasoning models.

Use Reasoning Models with dspy.TwoStepAdapter

Guide the user through configuring DSPy to work with reasoning models (o1, o3, o3-mini, DeepSeek-R1, Claude with extended thinking) that need special handling for structured output.

Why reasoning models need TwoStepAdapter

Reasoning models (o1, o3, DeepSeek-R1, Claude extended thinking) behave differently from standard chat models:

• They reject or ignore system prompts (o1/o3 strip them)
• They ignore formatting instructions (the model "thinks" and produces free-form output)
• They cannot follow ChatAdapter's field delimiters ([[ ## field_name ## ]])

TwoStepAdapter solves this with a two-phase approach:

1. Phase 1 (main LM): The reasoning model generates freely -- no formatting constraints
2. Phase 2 (extraction LM): A fast, cheap model parses the reasoning output into structured fields

When to use TwoStepAdapter

Use TwoStepAdapter when...	Use ChatAdapter (default) when...