Better boolean

[andreasjansson]

Important

Improved boolean and JSON handling, enhanced input generation for model fuzzing, and added tests for image output matching.

• Behavior:
  • Updated boolean() in ai.py to use a more specific system prompt and added thinking=True to call().
  • Enhanced json_object() in ai.py to handle JSON outputs more robustly, including stripping code block markers.
  • Modified call() in ai.py to support a thinking parameter and updated model version.
  • Improved output_matches_prompt() in match_outputs.py to better handle media description matching.
  • Added normalize_suffix() in match_outputs.py to standardize file extensions.
  • Updated make_fuzz_inputs() in predict.py to generate diverse inputs while respecting schema constraints.
• Tests:
  • Added test_non_matching_images.py to test image similarity and matching.
  • Updated test_output_matches_prompt.py to include more comprehensive image prompt matching tests.
  • Enhanced test_predict.py to test various scenarios for make_fuzz_inputs() including handling of null values and fixed inputs.
• Misc:
  • Renamed make_predict_inputs() to make_fuzz_inputs() across multiple files.
  • Updated predict.py to include a new function make_fuzz_system_prompt() for generating system prompts for fuzzing.

^{This description was created by for e8b7c0d. You can customize this summary. It will automatically update as commits are pushed.}