[Feature Request] Support bfp8_b 1d tensor in ttnn #15540
Assignees
Labels
Comments
hschoi4448
added
feature-request
hschoi4448
added a commit
that referenced
this issue
Nov 28, 2024
5 tasks
hschoi4448
added a commit
that referenced
this issue
Nov 28, 2024
### Ticket #15540 ### Problem description Currently, when trying to create a bfp8_b type 1D tensor using from_torch in ttnn, a runtime error occurs. However, it seems that removing the 1D runtime error doesn't cause any issues. ### What's changed I will remove the code that raises the runtime error. ```Python # SPDX-FileCopyrightText: © 2023 Tenstorrent Inc. # SPDX-License-Identifier: Apache-2.0 import pytest import torch import ttnn torch.set_printoptions(threshold=1000000, linewidth=100000000, sci_mode=False) @pytest.mark.parametrize("ttnn_dtype, torch_dtype", [ [ttnn.DataType.BFLOAT16, torch.bfloat16], [ttnn.DataType.BFLOAT8_B, torch.bfloat16], ],) def test_support_1d(device, ttnn_dtype, torch_dtype): torch_x = torch.rand(20, dtype=torch_dtype) ttnn_x_bf16 = ttnn.from_torch(torch_x, dtype=ttnn.DataType.BFLOAT16, layout=ttnn.TILE_LAYOUT) ttnn_x = ttnn.from_torch(torch_x, dtype=ttnn_dtype, layout=ttnn.TILE_LAYOUT) torch_y = ttnn.to_torch(ttnn_x).to(dtype=torch_dtype) print("ttnn_dtype", ttnn_dtype) print("torch_dtype", torch_dtype) print("torch_x\n", torch_x) print("ttnn_x_bf16\n", ttnn_x_bf16) print("ttnn_x\n", ttnn_x) print("torch_y\n", torch_y, type(torch_y)) ``` I have confirmed that the 1D `bfp8_b` tensor is created without any issues, but I couldn't find an appropriate location to add this test. ### Checklist - [ ] Post commit CI passes - [ ] Blackhole Post commit (if applicable) - [ ] Model regression CI testing passes (if applicable) - [ ] Device performance regression CI testing passes (if applicable) - [ ] New/Existing tests provide coverage for changes
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Is your feature request related to a problem? Please describe.
Currently, when trying to create a
bfp8_btype 1D tensor usingfrom_torchinttnn, a runtime error occurs.However, it seems that removing the 1D runtime error doesn't cause any issues.
Describe the solution you'd like
I will remove the code that raises the runtime error.
The text was updated successfully, but these errors were encountered: