Feat: Patch Anthropic system message

[arcaputo3]

• Enable OpenAI-style system messages with arrays for Anthropic
• Allow prompt caching at the system message level for Anthropic
• Remove Databricks assertion as they now mimic OpenAI broadly
• Enable multimodality without response model and autodetect_images=False

---

Important

Enhance Anthropic system message handling with OpenAI-style support, prompt caching, and multimodality adjustments, while removing Databricks-specific assertion.

• Behavior:
  • Enable OpenAI-style system messages with arrays for Anthropic in process_response.py.
  • Allow prompt caching at the system message level for Anthropic in process_response.py.
  • Remove Databricks assertion in client.py as they now mimic OpenAI broadly.
  • Enable multimodality without response model and autodetect_images=False in process_response.py.
• Utilities:
  • Add combine_system_messages() and extract_system_messages() in utils.py to handle system messages.
• Tests:
  • Add test_system.py to test Anthropic system message handling.
  • Add tests for combine_system_messages() and extract_system_messages() in test_utils.py.

^{This description was created by for 0b6d0d1. It will automatically update as commits are pushed.}

[ellipsis-dev]

👍 Looks good to me! Reviewed everything up to 0b6d0d1 in 23 seconds

More details

• Looked at 531 lines of code in 5 files
• Skipped 0 files when reviewing.
• Skipped posting 4 drafted comments based on config settings.

1. instructor/process_response.py:335

• Draft comment:
  The logic for handling system messages is duplicated in handle_anthropic_tools and handle_anthropic_json. Consider refactoring to avoid duplication.
• Reason this comment was not posted:
  Confidence changes required: 50%
  In instructor/process_response.py, the handle_anthropic_tools and handle_anthropic_json functions both use extract_system_messages and combine_system_messages. The logic for handling system messages is duplicated. Consider refactoring to avoid code duplication.

2. instructor/process_response.py:352

• Draft comment:
  The logic for handling system messages is duplicated in handle_anthropic_tools and handle_anthropic_json. Consider refactoring to avoid duplication.
• Reason this comment was not posted:
  Confidence changes required: 50%
  In instructor/process_response.py, the handle_anthropic_tools and handle_anthropic_json functions both use extract_system_messages and combine_system_messages. The logic for handling system messages is duplicated. Consider refactoring to avoid code duplication.

3. tests/llm/test_anthropic/test_system.py:37

• Draft comment:
  Assertions should include error messages for better debugging. This applies to all assertions in this file and in tests/test_utils.py.
• Reason this comment was not posted:
  Confidence changes required: 80%
  The assertion statements in the test files lack error messages, which is against the rule that assertions should always have an error message.

4. tests/llm/test_anthropic/test_system.py:1

• Draft comment:
  Ensure this new test file is added to the mkdocs.yml for documentation purposes.
• Reason this comment was not posted:
  Confidence changes required: 80%
  The new test file should be added to the mkdocs.yml for documentation purposes.

Workflow ID: wflow_z14QsaO8RsmT88IQ

---

You can customize Ellipsis with 👍 / 👎 feedback, review rules, user-specific overrides, quiet mode, and more.

[jxnl]

have you tested this all locally?

[arcaputo3]

have you tested this all locally?

yes I tried to make sure that this isn't a breaking change for standard system messages by merging them if they exist. tests are passing locally

[ivanleomk]

Forked this in #1130 and made changes there.

[arcaputo3]

@ivanleomk thanks for the cleanup!