[None][chore] DO NOT REVIEW, test PR#12829

Draft

longlee0622 wants to merge 28 commits intoNVIDIA:mainfrom

longlee0622:transformers-5.3.0

Commits on Apr 10, 2026

Upgrade transformers 5.3.0 dependency
longlee0622
committed
fix(auto_deploy): fallback SDPA mask patch when executorch helper removed
longlee0622
committed
fix: import AutoModelForImageTextToText for Transformers v5
longlee0622
committed
fix: shim get_parameter_device/dtype for Transformers v5
longlee0622
committed
fix: pass exist_ok to AutoConfig.register for Transformers 5.3
longlee0622
committed
fix: replace removed load_sharded_checkpoint for Transformers 5.3
longlee0622
committed
format code
longlee0622
committed
test: replace HybridCache with StaticCache helper for Transformers v5
longlee0622
committed
fix: map Transformers v5 rope_scaling type default to RoPE none
longlee0622
committed
fix: read HF rope_theta from rope_parameters for Transformers v5
longlee0622
committed
test: compat helpers for DynamicCache legacy API (Transformers v5)
longlee0622
committed
fix(auto_deploy): patch BambaModel when _update_causal_mask is absent
longlee0622
committed
fix: add SlidingWindowCache compatibility shim for Transformers v5
longlee0622
committed
test: update test_gpt_attention rope config for Transformers v5
longlee0622
committed
fix: map rope_type "default" to rope_gpt_neox in PositionEmbeddingType
longlee0622
committed
test: fix test_gpt_attention_IFB for Transformers v5
longlee0622
committed
fix: use getattr for pad_token_id in MllamaConfig for Transformers v5
longlee0622
committed
style: apply pre-commit formatting fixes
longlee0622
committed
fix: prevent duplicate 'disable' kwarg in DisabledTqdm
longlee0622
committed
fix: handle rope_scaling key changes in Qwen models for Transformers v5
longlee0622
committed
fix: use getattr for tie_word_embeddings for Transformers v5
longlee0622
committed
fix: support per-layer-type RoPE config (Gemma3) for Transformers v5
longlee0622
committed
test: fix test_gpt_attention HF attention calls for Transformers v5
longlee0622
committed
fix: pass return_dict=False in apply_chat_template for Transformers v5
longlee0622
committed
fix: pass return_dict=False in chat example for Transformers v5
longlee0622
committed
fix: add return_dict=False to remaining apply_chat_template calls
longlee0622
committed
fix: replace batch_encode_plus removed in Transformers v5
longlee0622
committed
test: add rope_type key for scaled RoPE configs in attention tests
longlee0622
committed