PSSA: A non-transformer language model written from scratch in Rust

Automated excerpt

The training throughput corpus switch, and whether ablating the memory bank changes the loss. any checkpoint and corpus it finds in the working directory. hours. Fit a checkpoint on a text corpus and write a . pssa file. PSSA layer, forward pass, plastic learning, consolidation, and . pssa serialization.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.