transformers
[DiffusionGemma] Support gradient checkpointing
#46572
Merged

Commits
  • Support gradient checkpointing in DiffusionGemma
    kashif committed 93 days ago
  • Test that checkpointing reproduces the exact loss and gradients
    kashif committed 93 days ago
  • Merge branch 'main' into diffusion-gemma-gradient-checkpointing
    kashif committed 93 days ago
  • Merge branch 'main' into diffusion-gemma-gradient-checkpointing
    kashif committed 82 days ago
  • update use_cache test now that the decoder ignores it instead of raising
    kashif committed 82 days ago
  • Merge branch 'main' into diffusion-gemma-gradient-checkpointing
    kashif committed 80 days ago
  • Merge remote-tracking branch 'upstream/main' into diffusion-gemma-gradient-checkpointing
    kashif committed 78 days ago
  • use a single read-only kv cache under gradient checkpointing
    kashif committed 78 days ago
  • Merge remote-tracking branch 'upstream/main' into diffusion-gemma-gradient-checkpointing
    kashif committed 78 days ago
  • Merge remote-tracking branch 'kashif/diffusion-gemma-gradient-checkpointing' into diffusion-gemma-gradient-checkpointing
    kashif committed 78 days ago
  • drop the unused encoder kv return and guard the decoder cache
    kashif committed 78 days ago
  • Merge branch 'main' into diffusion-gemma-gradient-checkpointing
    kashif committed 74 days ago
  • Merge branch 'main' into diffusion-gemma-gradient-checkpointing
    kashif committed 73 days ago
  • reuse BaseModelOutputWithPast and drop the unrelated mask cast
    kashif committed 73 days ago
  • Merge branch 'main' into diffusion-gemma-gradient-checkpointing
    kashif committed 71 days ago
  • Merge branch 'main' into diffusion-gemma-gradient-checkpointing
    kashif committed 65 days ago
Loading