#13929: Update document and sweep test file

.github/workflows/ttnn-run-sweeps.yaml

@@ -232,7 +232,7 @@ on:
           - eltwise.binary.ne.ne_scalar_pytorch2
           - eltwise.binary.hypot.hypot
           - eltwise.binary.xlogy.xlogy
-          - eltwise.binary_backward.ldexp_bw
+          - eltwise.binary_backward.ldexp_bw.ldexp_bw
           - eltwise.binary_backward.logaddexp_bw
           - eltwise.binary_backward.logaddexp2_bw
           - eltwise.binary_backward.addalpha_bw.addalpha_bw

tests/sweep_framework/sweeps/eltwise/binary_backward/ldexp_bw.py → tests/sweep_framework/sweeps/eltwise/binary_backward/ldexp_bw/ldexp_bw.py

@@ -14,9 +14,6 @@
 from tests.ttnn.utils_for_testing import check_with_pcc, start_measuring_time, stop_measuring_time
 from models.utility_functions import torch_random
 
-# Override the default timeout in seconds for hang detection.
-TIMEOUT = 30
-
 random.seed(0)
 
 
@@ -25,7 +22,7 @@
 # Each suite has a key name (in this case "suite_1" and "suite_2") which will associate the test vectors to this specific suite of inputs.
 # Developers can create their own generator functions and pass them to the parameters as inputs.
 parameters = {
-    "nightly": {
+    "ldx_exp": {
         "input_shape": gen_shapes([1, 1, 1, 1], [6, 12, 256, 256], [1, 1, 1, 1], 16)
         + gen_shapes([1, 1, 1], [12, 256, 256], [1, 1, 1], 16)
         + gen_shapes([1, 1], [256, 256], [1, 1], 16),
@@ -74,13 +71,13 @@ def run(
     )
 
     torch_input_tensor_a = gen_func_with_cast_tt(
-        partial(torch_random, low=-90, high=90, dtype=torch.float32), input_a_dtype
+        partial(torch_random, low=-80, high=80, dtype=torch.float32), input_a_dtype
     )(input_shape)
     torch_input_tensor_a.requires_grad = True
     torch_input_tensor_a.retain_grad()
 
     torch_input_tensor_b = gen_func_with_cast_tt(
-        partial(torch_random, low=-90, high=90, dtype=torch.float32), input_b_dtype
+        partial(torch_random, low=-80, high=80, dtype=torch.float32), input_b_dtype
     )(input_shape)
     torch_input_tensor_b.requires_grad = True
     torch_input_tensor_b.retain_grad()

ttnn/cpp/ttnn/operations/eltwise/binary_backward/binary_backward_pybind.hpp

@@ -22,7 +22,7 @@ namespace binary_backward {
 namespace detail {
 
 template <typename binary_backward_operation_t>
-void bind_binary_backward_ops(py::module& module, const binary_backward_operation_t& operation, const std::string_view description, const std::string_view supported_dtype = "BFLOAT16") {
+void bind_binary_backward_ops(py::module& module, const binary_backward_operation_t& operation, const std::string_view description, const std::string_view supported_dtype = "BFLOAT16", const std::string_view note = "") {
     auto doc = fmt::format(
         R"doc(
         {2}
@@ -53,6 +53,8 @@ void bind_binary_backward_ops(py::module& module, const binary_backward_operatio
 
             bfloat8_b/bfloat4_b is only supported on TILE_LAYOUT
 
+            {4}
+
         Example:
             >>> grad_tensor = ttnn.to_device(ttnn.from_torch(torch.tensor((1, 2), dtype=torch.bfloat16)), device=device)
             >>> tensor1 = ttnn.to_device(ttnn.from_torch(torch.tensor((1, 2), dtype=torch.bfloat16)), device=device)
@@ -65,7 +67,8 @@ void bind_binary_backward_ops(py::module& module, const binary_backward_operatio
         operation.base_name(),
         operation.python_fully_qualified_name(),
         description,
-        supported_dtype);
+        supported_dtype,
+        note);
 
     bind_registered_operation(
         module,
@@ -1110,7 +1113,7 @@ void py_module(py::module& module) {
         module,
         ttnn::ldexp_bw,
         R"doc(Performs backward operations for ldexp of :attr:`input_tensor_a` and :attr:`input_tensor_b` with given :attr:`grad_tensor`.)doc",
-        R"doc(BFLOAT16)doc");
+        R"doc(BFLOAT16)doc", R"doc(Recommended input range : [-80, 80]. Performance of the PCC may degrade if the input falls outside this range.)doc");
 
 
     detail::bind_binary_backward_ops(