Research Engineering & Debugging · Problem 4 of 8
Write a cached-vs-full-forward equivalence test for incremental decoding.
Implement the function/class skeleton in the editor. Any correct approach is accepted.
import torch
def test_cache_equals_full(model, input_ids, rtol=0.0001, atol=1e-05):
raise NotImplementedErrorReady when you are
Submit your solution and a structured review appears here — verdict, score, and concrete feedback. Any correct approach passes.
Write a cached-vs-full-forward equivalence test for incremental decoding.
Implement the function/class skeleton in the editor. Any correct approach is accepted.