ttnn.bias_gelu_bw

ttnn.bias_gelu_bw(grad_tensor: ttnn.Tensor, input_tensor_a: ttnn.Tensor, input_tensor_b: ttnn.Tensor or Number, *, approximate: string = none, memory_config: ttnn.MemoryConfig | None = None) → List of ttnn.Tensor

Performs backward operations for bias_gelu on input_tensor_a and input_tensor_b or input_tensor and bias, with given grad_tensor using given approximate mode. approximate mode can be ‘none’, ‘tanh’.

Parameters:

  • grad_tensor (ttnn.Tensor) – the input gradient tensor.

  • input_tensor_a (ttnn.Tensor) – the input tensor.

  • input_tensor_b (ttnn.Tensor or Number) – the input tensor.

Keyword Arguments:

  • approximate (string) – Approximation type. Defaults to none.

  • memory_config (ttnn.MemoryConfig, optional) – Memory configuration for the operation. Defaults to None.

Returns:

List of ttnn.Tensor – the output tensor.

Note

Supported dtypes, layouts, and ranks:

Dtypes

Layouts

Ranks

BFLOAT16

TILE

2, 3, 4

bfloat8_b/bfloat4_b is only supported on TILE_LAYOUT

For more details about BFLOAT8_B, refer to the BFLOAT8_B limitations.

Example

>>> grad_tensor = ttnn.from_torch(torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16), layout=ttnn.TILE_LAYOUT, device=device)
>>> tensor1 = ttnn.from_torch(torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16, requires_grad=True), layout=ttnn.TILE_LAYOUT, device=device)
>>> tensor2 = ttnn.from_torch(torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16, requires_grad=True), layout=ttnn.TILE_LAYOUT, device=device)
>>> approximate = "none"
>>> output = ttnn.bias_gelu_bw(grad_tensor, tensor1, tensor2, approximate)