riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-08-02 11:11:05 +06:00
Code Issues Actions 3 Packages Projects Releases Wiki Activity

Fix the seamless_m4t cannot work on Gaudi (#38363)
Some checks are pending
Self-hosted runner (benchmark) / Benchmark (aws-g5-4xlarge-cache) (push) Waiting to run
Details
Build documentation / build (push) Waiting to run
Details
New model PR merged notification / Notify new model (push) Waiting to run
Details
Slow tests on important models (on Push - A10) / Get all modified files (push) Waiting to run
Details
Slow tests on important models (on Push - A10) / Slow & FA2 tests (push) Blocked by required conditions
Details
Self-hosted runner (push-caller) / Check if setup was changed (push) Waiting to run
Details
Self-hosted runner (push-caller) / build-docker-containers (push) Blocked by required conditions
Details
Self-hosted runner (push-caller) / Trigger Push CI (push) Blocked by required conditions
Details
Secret Leaks / trufflehog (push) Waiting to run
Details
Update Transformers metadata / build_and_package (push) Waiting to run
Details

Browse Source 
* Fix the seamless_m4t cannot work on Gaudi

Signed-off-by: yuanwu <yuan.wu@intel.com>

* Refine the patch

Signed-off-by: yuanwu <yuan.wu@intel.com>

* Fix seamless_m4t_v2 crash

Signed-off-by: yuanwu <yuan.wu@intel.com>

* Use the patched_gather

Signed-off-by: yuanwu <yuan.wu@intel.com>

* Remove debug logs

Signed-off-by: yuanwu <yuan.wu@intel.com>

* Remove useless modifications

Signed-off-by: yuanwu <yuan.wu@intel.com>

* Add hpu check

Signed-off-by: yuanwu <yuan.wu@intel.com>

* Add comments

Signed-off-by: yuanwu <yuan.wu@intel.com>

---------

Signed-off-by: yuanwu <yuan.wu@intel.com>
Co-authored-by: Ilyas Moutawwakil <57442720+IlyasMoutawwakil@users.noreply.github.com>
This commit is contained in:
Yuan Wu 2025-06-25 18:40:01 +08:00 committed by GitHub
parent 7503cb9113
commit de98fb25a3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 22 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
src/transformers/utils/import_utils.py
View File

@ -851,6 +851,28 @@ def is_torch_hpu_available():
torch.Tensor.masked_fill_ = patched_masked_fill_
# We patch torch.gather for int64 tensors to avoid a bug on Gaudi
# Graph compile failed with synStatus 26 [Generic failure]
# This can be removed once bug is fixed but for now we need it.
original_gather = torch.Tensor.gather
def patched_gather(input: torch.Tensor, dim: int, index: torch.LongTensor) -> torch.Tensor:
if input.dtype == torch.int64 and input.device.type == "hpu":
logger.warning_once(
"torch.gather is not supported for int64 tensors on Gaudi. "
"This operation will be performed patched_gather using indexing."
)
idx = [torch.arange(size, device=input.device, dtype=input.dtype) for size in input.shape]
idx[dim] = index
idx = tuple(idx)
output = input[idx]
return output
else:
return original_gather(input, dim, index)
torch.Tensor.gather = patched_gather
# IlyasMoutawwakil: we patch torch.compile to use the HPU backend by default
# https://github.com/huggingface/transformers/pull/38790#discussion_r2157043944
# This is necessary for cases where torch.compile is used as a decorator (defaulting to inductor)