Added new example for Anthropic HH / Bias EK-FAC score computation for Pythia models (base and SFTed)

[sevendaystoglory]

This folder shows one minimal, self-contained example of how to use kronfluence to:

1. Fit EKFAC influence factors on a large-language model (the 410 M parameter Pythia model) using a subset of 10 k Anthropic-HH training samples.
2. Compute pairwise influence scores between that training set and the "Stereotypical Bias" evaluation set.

The goal of the example is to be copy-paste simple: after installing the requirements, a single command will produce both the factors and the influence scores.

Layout

AnthropicHH-Bias/
 ├─ fit_all_factors.py         # Step-1 · compute EKFAC factors
 ├─ compute_pairwise_scores.py # Step-2 · compute pairwise scores
 ├─ task.py                    # Loss + measurement definitions
 ├─ utils.py                   # Helper functions (dataset loading, metrics…)
 ├─ SFT_Trainer_Lora.py        # Optional LoRA fine-tuning script
 ├─ requirements.txt           # Extra runtime deps (torch + transformers …)
 └─ README.md                 

[pomonam]

Thank you! This looks great! Would it be easy to fix some ruff / lining issues? I'm happy to merge this.