fix(evals): miscellaneous fixes and ergonomic improvements

[ehutt]

Changes Include:

• dataframe eval methods return json instead of string in score/details columns
• adds a default tqdm progress bar formatter to dataframe evals + flag to hide
• moves LLM to a top level import since it is commonly used with Evaluators
• updates notebooks in tutorials accordingly with the new import statements

Bug fix:

• runningbind_evaluator on an LLMEvaluator returns an error due to attempting to deep copy the LLM object which can't be pickled. Now the method does a shallow copy of the evaluator and only deep copies the input_mapping, which is the only property we don't want the original and copy to share.

---

Note

Switch dataframe eval outputs to dicts, add default/hideable tqdm progress bar, expose LLM at top level, and fix bind_evaluator copying; update notebooks/tests accordingly.

• Evals 2.0 (Core):
  • Return JSON-serializable dicts (not strings) in {score}_score and {evaluator}_execution_details columns for dataframe evals.
  • Add progress bar controls: default formatter via default_tqdm_progress_bar_formatter(...) and hide_tqdm_bar flag for sync/async dataframe evals.
  • Fix bind_evaluator(...): use shallow copy of evaluator (avoids deepcopying LLM with locks) and deep copy only the input_mapping.
• API/Exports:
  • Expose LLM at top-level import (from phoenix.evals import LLM).
• Utils:
  • Enhance to_annotation_dataframe to parse dict-backed scores; export default_tqdm_progress_bar_formatter.
• Docs/Tutorials/Examples:
  • Update imports to use top-level LLM; minor notebook cleanup.
• Tests:
  • Adjust expectations to handle dict outputs for scores and execution details; add progress bar coverage.

^{Written by Cursor Bugbot for commit 6d7c3f0. This will update automatically on new commits. Configure here.}

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB