Timesfm 2.5 (#41763)
* init
* initial modules
* initial model
* fix config
* compare
* rope embedding
* normalization
* fix scale
* add _scale_query
* revin
* get rid of magic numbers
* use configs
* scaling=1
* fix Timesfm2P5MLP
* normalization
* wip
* update for v5
* updated docs
* Delete convert_timesfm_2p5_orignal_to_hf.py
* fix copyright
* fix test
* fix import
* added MODEL_FOR_TIME_SERIES_PREDICTION_MAPPING_NAMES
* fix CI
* fix docstring
* add Timesfm2P5Model to check_repo.py
* download from hub and fix docs
* use transformers weights
* add back test_eager_matches_sdpa_inference
* add flash and flex attn support
* fix CI
* remove unused helper
* fix auto_docstring
* fix slow tests
* faster tests
* fix doc
* fix docs
* formatting
* add dates
* require_flash_attn
* fix from review
* added Gemma2DecoderLayer
* fixes from reviewer
* fix repo
* fix year
* fix formatting
* Update docs/source/en/model_doc/timesfm_2p5.md
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
* Update src/transformers/models/timesfm_2p5/modular_timesfm_2p5.py
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
* fixes from review
* use llama attnetion and rope
* use config and inherit from timesfm
* add comment and fix docstring
* remove docstring
* remove unused decorator
* remove attn_weights
* parent model and inherit it
* kwargs type
* remove return_forecast_on_context
* simplify
* removed unneeded test
* fix reason
* fix CI
* fix repo
* fix-repo
* fix-repo
* fix CI
* fix float
* remove true only configs
* Update tests/models/timesfm_2p5/test_modeling_timesfm_2p5.py
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
* Update src/transformers/models/timesfm_2p5/modular_timesfm_2p5.py
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
* intial broken rename
* rename classes and config
* fix test
* fix formatting
* update timesfm 2.5 release date
* remove query_pre_attn_scalar and rope_theta
* small fixups
* fix run_slow test on gpu
* lets add a revision
* fix parallel test
---------
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
Co-authored-by: vasqu <antonprogamer@gmail.com>
