2026-05-29 | Maxwell

Goal: Train Vaswani Encoder Decoder and on various depths and test generalization

Summary: Seq2Seq and GPT-2 with RoPE can generalize to unseen frames within the same depth but not different depths

Work sessions

In	Out	Task
09:00	17:00	Train models

Goals

1. [P0] Document discussion from Weekly meeting

   a. Draw a flowchart of how grammar is generated

   b. Summary of Grammar, frames

   c. What types of models were trained/datasets

   1c. Interesting generalization
   

   e. Snippet images of attention matrices

2. [P1] Plan out tasks for next week from follow ups 05/28 Depth and Attention Summary

Meetings

• None