Feature/results to df

[humaira-rf]

Changes

• updated notebooks to display dataframe from the results dict
• Added other experimentation knobs to the results dict

Testing

• Tested on all three notebooks with downsampled data and just 2 configs

Screenshots

a. hyrbid - gsm8k:

b. Fully openai - scifact:(couldn't screenshot the full df)

c. Fully Local: fiqa

---

Note

Adds pipeline metadata to final metrics with consistent ordering and updates tutorials to display results as a DataFrame.

• Evals Controller (rapidfireai/evals/scheduling/controller.py):
  • Final metrics pipeline: _compute_final_metrics_for_pipelines now accepts optional pipeline_id_to_info and injects pipeline metadata (e.g., model_name, search_type, rag_k, top_n, chunk_size, chunk_overlap, sampling_params, prompt_manager_k, model_config).
  • Ordering and output: Reorders cumulative metrics to run_id, model_name, hyperparams, Samples Processed, then remaining metrics; returns ordered_metrics and uses it for progress display.
  • Integration: Builds pipeline_id_to_info from pipeline_info and passes it to final-metrics computation.
• Tutorial Notebooks:
  • Replace sample printouts with conversion of results into a pandas DataFrame (results_df) in rf-tutorial-gsm8k-fewshot.ipynb, rf-tutorial-rag-fiqa.ipynb, and rf-tutorial-scifact-full-evaluation.ipynb.
  • Minor notebook metadata/formatting tweaks (ids, kernelspec).

^{Written by Cursor Bugbot for commit efdc979. This will update automatically on new commits. Configure here.}

[arun-rfai]

I'd suggest leaving the dict return type of run_evals as is. Conver the dict to dataframe in the notebook cell itself, limiting it only metrics columns and the config knobs akin to the second table printed in run_evals.

Also, pipeline_id -> run_id.

[arun-rfai]

LGTM