You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardexpand all lines: src/common/transformations/include/transformations/fp16_compression/mark_decompression_convert_constant_folding.hpp
+2-2
Original file line number
Diff line number
Diff line change
@@ -62,10 +62,10 @@ class ov::pass::KeepConstantsPrecisionAndAddConverts : public MatcherPass {
62
62
63
63
/**
64
64
* @ingroup ov_transformation_common_api
65
-
* @brief Prevents ConstantFolding for f16/bf16 Const + Convert_To_FP32 to keep original FW float Constants.
65
+
* @brief Prevents ConstantFolding for low precision Const + Convert_To_FP32 to keep original FW float Constants.
66
66
* Original precision should be kept as long as possible, this prevents redundant conversions and saves memory.
67
67
* E.g. if original FW model was already compressed no need to upcast during CF, store intermediate f32 consts and
68
-
* then again compress them to f16 during save_model.
68
+
* then again compress them to low precision during save_model.
69
69
*/
70
70
classov::pass::MarkCompressedFloatConstants : public MatcherPass {
Copy file name to clipboardexpand all lines: src/common/transformations/src/transformations/fp16_compression/mark_decompression_convert_constant_folding.cpp
0 commit comments