Fix message content field handling for tool calls and multimodal input#4029
Merged
lvhan028 merged 1 commit intoInternLM:mainfrom Oct 20, 2025
Merged
Fix message content field handling for tool calls and multimodal input#4029lvhan028 merged 1 commit intoInternLM:mainfrom
lvhan028 merged 1 commit intoInternLM:mainfrom
Conversation
This was referenced Oct 11, 2025
Closed
lvhan028
reviewed
Oct 13, 2025
lvhan028
reviewed
Oct 13, 2025
Contributor
Author
|
Thanks @lvhan028 for the excellent review! GPT-5-Codex is really helpful at walking through code! |
This commit fixes two issues with message content processing: 1. **Missing content field in tool call responses**: When assistant messages contain tool_calls, the content field may be missing or None, causing KeyError when checking isinstance(msg['content'], list). 2. **Incomplete multimodal text merging**: Previous implementation only extracted the first text block. Now correctly merges all text blocks with newline separator. Changes: - Add _merge_message_content() module-level function in async_engine.py - Handle 4 cases: missing content, None content, string content, list content - Convert None/missing content to empty string '' (prevents Jinja2 errors) - Use '\n'.join() to merge text blocks (matches vLLM behavior) - Preserve all message fields (e.g., tool_calls, name, etc.) - Add comprehensive unit tests (19 test cases in test_content_merge.py) Implementation based on vLLM's content normalization logic: https://github.com/vllm-project/vllm/blob/main/vllm/entrypoints/chat_utils.py 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
lvhan028
approved these changes
Oct 20, 2025
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
This commit fixes two issues with message content processing:
Missing content field in tool call responses: When assistant messages contain tool_calls, the content field may be missing or None, causing KeyError when checking isinstance(msg['content'], list).
Incomplete multimodal text merging: Previous implementation only extracted the first text block. Now correctly merges all text blocks with newline separator.
Changes:
Implementation based on vLLM's content normalization logic: https://github.com/vllm-project/vllm/blob/main/vllm/entrypoints/chat_utils.py
🤖 Generated with Claude Code