Skip to content

Tags: lucidrains/self-reasoning-tokens-pytorch

Tags

0.0.4

Toggle 0.0.4's commit message
allow also for scaling the grads near and far for queries, keys, values

0.0.3

Toggle 0.0.3's commit message
allow also for scaling the grads near and far for queries, keys, values

0.0.2

Toggle 0.0.2's commit message
offer a naive unoptimized attention w/ stop graddable queries, keys, …

…values, and fix the self reasoning transformer to only stop grad keys and values for the reasoning tokens