diff --git a/source/elements/oneDNN/source/data_model/data_types/int8.rst b/source/elements/oneDNN/source/data_model/data_types/int8.rst index 0b08821cbd..09773bb12b 100644 --- a/source/elements/oneDNN/source/data_model/data_types/int8.rst +++ b/source/elements/oneDNN/source/data_model/data_types/int8.rst @@ -19,7 +19,7 @@ acceptable) accuracy. Workflow ******** -oneDNN support symmetric and assymetric quantization models. +oneDNN support symmetric and asymmetric quantization models. .. _int8-quantization-label: @@ -27,7 +27,7 @@ Quantization Model ================== For each int8 tensor, the oneDNN library allows to specify scaling -factors and zero-points (also refered to as quantization +factors and zero-points (also referred to as quantization parameters), and assumes the following mathematical relationship: .. math:: diff --git a/source/elements/oneDNN/source/graph/programming_model.rst b/source/elements/oneDNN/source/graph/programming_model.rst index 45c9514415..7ff22eb93f 100644 --- a/source/elements/oneDNN/source/graph/programming_model.rst +++ b/source/elements/oneDNN/source/graph/programming_model.rst @@ -177,14 +177,14 @@ tensors with the same id should be identical at the graph construction time. Once the graph is fully described, |finalize| should be called. This prevents any other operation from being added, and allows to call |get_partitions| in order to get the set of partitions for that -graph. The graph doesn’t hold any meaning to the user after +graph. The graph does not hold any meaning to the user after partitioning and should freed by the user. All the OPs added to the graph will be contained in one of the returned partitions. If an OP is not supported by the oneDNN Graph API implementation, the corresponding partition will be marked as “not supported”. Users can check the supporting status of a partition via the |is_supported|. Partitions -should not form cyclic dependence within the graph. If user doesn’t pass a +should not form cyclic dependence within the graph. If user does not pass a complete graph, it is the user's responsibility to detect any dependence cycle between the partitions and operations not passing to oneDNN Graph implementation. diff --git a/source/elements/oneDPL/source/parallel_api.rst b/source/elements/oneDPL/source/parallel_api.rst index 6a78560e8c..b3f96f3595 100644 --- a/source/elements/oneDPL/source/parallel_api.rst +++ b/source/elements/oneDPL/source/parallel_api.rst @@ -11,7 +11,7 @@ All those algorithms work with *C++ Standard aligned execution policies* and wit execution policies*. Additionally, oneDPL provides wrapper functions for `SYCL`_ buffers, special iterators, and -a set of non-standard parallel algortithms. +a set of non-standard parallel algorithms. C++ Standard aligned execution policies +++++++++++++++++++++++++++++++++++++++ diff --git a/source/elements/oneMKL/source/domains/lapack/ormtr.rst b/source/elements/oneMKL/source/domains/lapack/ormtr.rst index c5595437eb..babd3b380a 100644 --- a/source/elements/oneMKL/source/domains/lapack/ormtr.rst +++ b/source/elements/oneMKL/source/domains/lapack/ormtr.rst @@ -100,7 +100,7 @@ lda The leading dimension of ``a`` :math:`(\max(1, r) \le \text{lda})`. tau - The buffer ``tau`` as returned bya :ref:`onemkl_lapack_sytrd`. The + The buffer ``tau`` as returned by a :ref:`onemkl_lapack_sytrd`. The dimension of ``tau`` must be at least :math:`\max(1, r-1)`. c diff --git a/source/spelling_wordlist.txt b/source/spelling_wordlist.txt index 236e9b27e4..f37d64b6c0 100644 --- a/source/spelling_wordlist.txt +++ b/source/spelling_wordlist.txt @@ -811,6 +811,7 @@ zgebrd ≥ fi fl -parmeterizes +parameterizes QNaN decrementing +deallocating