Bugfix/IC Ops issues

[david-rfai]

Changes

• Fixed the following bugs -
  • [BUG][RAG][v0.12.6][RUN_EVALS] Run Evals progress table does not display knobs for STOPPED, DELETED / CLONED runs after completion #101
  • [BUG][RAG][v0.12.6][RUN_EVALS] Printing of the returned dict/df of run_evals does not include STOPPED runs as tuples #100
  • [BUG][RAG][v0.12.6][IC-OPS] Clone config knobs not changed through IC-OPs Clone #99
  • Clone’s dict knob is displayed in a different format than parent

Testing

• Ran E2E of FIQA notebook

Screenshots (if applicable)

---

Note

Compute final metrics for all non-deleted/non-failed runs, clone pipelines using original user param dicts, and make the progress display more robust; tutorials switch tokenizer padding to right.

• Evals/Scheduler:
  • Compute final metrics for pipelines with statuses COMPLETED, STOPPED, and ONGOING; skip DELETED and FAILED in _compute_final_metrics_for_pipelines.
  • Preserve and reorder cumulative metrics with added pipeline info; update progress display with final metrics.
• Interactive Control (Clone):
  • For vllm and openai, source parent configs from _user_params (e.g., model_config, sampling_params, client/rate limits) before applying JSON edits.
  • Extract sampling params for display from _user_params.
• Progress Display:
  • Add defensive checks (skip missing pipeline data), default missing model to -, and ensure pandas shows all columns in start/stop/_render.
• Experiment API:
  • Update run_evals return docstring to clarify included/excluded run statuses.
• Tutorial Notebooks:
  • Change tokenizer padding_side to right across DPO/GRPO notebooks and minor copy updates.

^{Written by Cursor Bugbot for commit ac051f2. This will update automatically on new commits. Configure here.}