Fix a regression in encoder-decoder generation cache initialization (#46111)
* Fix a regression in encoder-decoder generation cache initialization
Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
* move the fix to modeling part
Signed-off-by: root <root@pvc747634.jf.intel.com>
* update code
Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
* Override `_prepare_cache_for_generation`
Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
* add related fast test
Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
* revert masking_utils.py change
Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
---------
Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
Signed-off-by: root <root@pvc747634.jf.intel.com>
Co-authored-by: root <root@pvc747634.jf.intel.com>