Rerun Shanehsazzadeh notebook with ESM-2 650M

[matsen]

Rerun Shanehsazzadeh notebook with ESM-2 650M

Problem

The paper states that ESM-2 650M was used for all benchmarks, but shanehsazzadeh.ipynb currently uses ESM-2 3B:

esm_wrapper = esm2_wrapper_of_size("3B")  # line 107

This was flagged by an external reviewer comparing our reported correlations.

Fix

1. Change esm2_wrapper_of_size("3B") to esm2_wrapper_of_size("650M") in notebooks/dasm_paper/shanehsazzadeh.ipynb
2. Rerun the notebook to regenerate cached results in _ignore/
3. Update correlation values if they change
4. Verify Table 2 (shane correlations) in the paper matches the new results

Notes

• Koenig notebook already uses 650M on main branch
• The 3B vs 650M comparison for Koenig is documented in PR Upgrade ESM model from 650M to 3B #2 and Table S1 of the paper