tilelang.utils.tensor¶

The profiler and convert to torch utils

類別¶

TensorSupplyType

Create a collection of name/value pairs.

函式¶

`is_float8_dtype`(dtype)
`fp8_remove_negative_zeros_`(tensor)
`map_torch_type`(intype)
`get_tensor_supply`([supply_type])
`torch_assert_close`(tensor_a, tensor_b[, rtol, atol, ...])	Custom function to assert that two tensors are "close enough," allowing a specified

Module Contents¶

tilelang.utils.tensor.is_float8_dtype(dtype)¶

參數:: dtype (torch.dtype)
回傳型別:: bool

tilelang.utils.tensor.fp8_remove_negative_zeros_(tensor)¶

參數:: tensor (torch.Tensor)

class tilelang.utils.tensor.TensorSupplyType(*args, **kwds)¶

Bases: enum.Enum

Create a collection of name/value pairs.

Example enumeration:

>>> class Color(Enum):
...     RED = 1
...     BLUE = 2
...     GREEN = 3

Access them by:

attribute access:
```
>>> Color.RED
<Color.RED: 1>
```
value lookup:
```
>>> Color(1)
<Color.RED: 1>
```
name lookup:
```
>>> Color['RED']
<Color.RED: 1>
```

Enumerations can be iterated over, and know how many members they have:

>>> len(Color)
3

>>> list(Color)
[<Color.RED: 1>, <Color.BLUE: 2>, <Color.GREEN: 3>]

Methods can be added to enumerations, and members can have their own attributes -- see the documentation for details.

Integer = 1¶

Uniform = 2¶

Normal = 3¶

Randn = 4¶

Zero = 5¶

One = 6¶

Auto = 7¶

tilelang.utils.tensor.map_torch_type(intype)¶

回傳型別:: torch.dtype

tilelang.utils.tensor.get_tensor_supply(supply_type=TensorSupplyType.Integer)¶

參數:: supply_type (TensorSupplyType)

tilelang.utils.tensor.torch_assert_close(tensor_a, tensor_b, rtol=0.01, atol=0.001, max_mismatched_ratio=0.001, verbose=False, equal_nan=True, check_device=True, check_dtype=True, check_layout=True, check_stride=False, base_name='LHS', ref_name='RHS')¶

Custom function to assert that two tensors are "close enough," allowing a specified percentage of mismatched elements.

Parameters:¶

tensor_atorch.Tensor: The first tensor to compare.
tensor_btorch.Tensor: The second tensor to compare.
rtolfloat, optional: Relative tolerance for comparison. Default is 1e-2.
atolfloat, optional: Absolute tolerance for comparison. Default is 1e-3.
max_mismatched_ratiofloat, optional: Maximum ratio of mismatched elements allowed (relative to the total number of elements). Default is 0.001 (0.1% of total elements).

Raises:¶

AssertionError:: If the ratio of mismatched elements exceeds max_mismatched_ratio.

參數:

verbose (bool)
equal_nan (bool)
check_device (bool)
check_dtype (bool)
check_layout (bool)
check_stride (bool)
base_name (str)
ref_name (str)