transformers
perf: avoid recomputing rotary_emb for each layer in some Google and ModernBERT models
#45555
Merged

Commits
  • perf: avoid redundant rotary_emb calls for `position_embeddings`
    casinca committed 181 days ago
  • perf: avoid redundant rotary_emb calls for `position_embeddings` dict
    casinca committed 180 days ago
Loading