GH-3657: Fix DeepSeek tool call content null issue #3817
+35
−3
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Outdated suggestions cannot be applied.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.
✨ Summary
This PR resolves the issue where
DeepSeek
chat completions with tool calls return an assistant message withcontent = null
, leading to schema validation errors or downstream failures (e.g., insglang
or FastAPI with pydantic).To address this, we explicitly set the assistant message content to
"__tool_call__"
when tool calls are present but content isnull
.🛠 Changes
DeepSeekChatModel
, updatedbuildGeneration(...)
logic:content == null && toolCalls != empty
, setcontent = "__tool_call__"
to maintain compatibility.content
to be a non-null string.AssistantMessage#getText()
logic, keeping it backward compatible for other models.💡 Context
DeepSeek sometimes returns:
This results in downstream consumers (e.g.,
sglang
, FastAPI with pydantic) throwing errors:By ensuring a placeholder content (
"__tool_call__"
), we:🧪 Tests & Verification
ChatClientResponseTests.whenAssistantMessageHasOnlyToolCalls_thenContentIsToolCallMarker
getText()
returns"__tool_call__"
when expectedGeneration.getOutput().getText()
./mvnw clean verify
)-Ddisable.checks=false
)git log --oneline --graph
confirmed)main
📘 Documentation
🔗 Closes #3657
Please review and let me know if you’d like any refinements.
Happy to iterate!