transformers

225254c1 - Preventing initialization of siglip's lecun_normal_, default_flax_embed_init in ZeRO3 (#43574)

Preventing initialization of siglip's lecun_normal_, default_flax_embed_init in ZeRO3 (#43574) * Prevent redundant initialization in lecun_normal_ and default_flax_embed_init * apply style * fix check_repository_consistency * lecun_normal_ & default_flax_embed init > initialization.py * Update src/transformers/initialization.py Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com> * Rename `_variance_scaling_` to `_variance_scaling` for consistency and update references * Refactor initialization calls in `Phi4MultimodalVisionPreTrainedModel`: remove redundant `init.` prefix for clarity * Fix initialization calls in `Phi4MultimodalVisionPreTrainedModel`: update to use `init` namespace for clarity * Add test for SigLIP model initialization with DeepSpeed ZeRO-3 * Update tests/deepspeed/test_deepspeed.py Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com> * Update tests/deepspeed/test_deepspeed.py Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com> * Apply suggestion from @vasqu Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com> * fix: update embedding initialization function to use the correct suffix * add test for variance scaling initialization with DeepSpeed ZeRO-3 in SigLIP models * small nits --------- Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com> Co-authored-by: vasqu <antonprogamer@gmail.com>