Skip to content

Commit 17f4aca

Browse files
jorickertljfitz
andauthored
Make documentation clearer
Co-authored-by: Liam Fitzpatrick <[email protected]>
1 parent 538a52b commit 17f4aca

File tree

1 file changed

+2
-1
lines changed

1 file changed

+2
-1
lines changed

mlir/include/mlir/Dialect/Quant/IR/QuantTypes.h

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -237,7 +237,8 @@ class AnyQuantizedType
237237
/// number of elements. For example, a single BFP16 element requires 16 bits to
238238
/// be represented, but a block of 8 BFP16 elements can be packed into 9 bits
239239
/// per element on average (72 bits total). The storage type for
240-
/// BlockFloatQuantizedType is the "packed" type, so for BFP16 i9.
240+
/// an element from BlockFloatQuantizedType is the "packed" type
241+
/// divided by the number of packed elements, so for BFP16 i9.
241242
/// -- As accessing properties like min/max storage values and integral width
242243
/// depend on the block size, these methods are overridden to return errors.
243244
/// - The expressed type is not stored yet, this may change if there is a use

0 commit comments

Comments
 (0)