Flexible handling of dspy.ToolCalls
parsing
#8563
Open
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We have observed that LMs can mess up the deeply nested type, especially when it understands the semantics behind the scene, specifically
dspy.ToolCalls
here. We are expecting the format{"tool_calls": [{"name": "foo", "args": {...}}, {"name": "bar", "args": {...}}]}
, but in reality there are a few cases seen from the same LM:{"name": foo, "args": {...}}
, which is just a dict of information of one function calling. This behavior is pretty common, it exists not just in those small models on ollama, but also gpt-4o-mini, gpt-4o, even gemini-2.5.[{"name": "foo", "args": {...}}, {"name": "bar", "args": {...}}]
. The outertool_calls
key is lost.Since this is a stable pattern, it makes sense to have the default type handle this parsing.