Mercykid-bash 29fb27d3bb BugFix: Fix moe_load accumulation error in ACL graph mode (#6182) ...

This PR fixes the numerical error in moe_load accumulation under ACL
graph mode on NPU: using += for NPU tensors in graph mode does not throw
errors but leads to incorrect values, so we replace it with the in-place
add_() method to ensure accurate calculation.

Signed-off-by: Mercykid-bash <ruanche0218@gmail.com>

2026-01-26 17:18:46 +08:00

..

__init__.py

[Refactor] [MoE] Rename moe-related classes & files (#3646)

2025-10-25 11:22:03 +08:00

comm_utils.py

[Refactor] [MoE] Rename moe-related classes & files (#3646)

2025-10-25 11:22:03 +08:00

experts_selector.py

[Kernel] Add moe_gating_top_k operator support for Ascend NPU (#5579)

2026-01-07 21:42:31 +08:00

fused_moe.py

BugFix: Fix moe_load accumulation error in ACL graph mode (#6182)

2026-01-26 17:18:46 +08:00

moe_comm_method.py

[BugFix] Fix input parameter bug of dispatch_gmm_combine_decode[RFC: issue 5476] (#5932)

2026-01-21 09:26:40 +08:00

moe_mlp.py

[refactor] Remove unnecessary attributes from set_ascend_forward_context (#5204)

2025-12-23 08:49:52 +08:00

prepare_finalize.py

[Feature] Support fine-grained shared expert overlap (#5482)

2026-01-17 11:53:22 +08:00

token_dispatcher.py

[BugFix] Fix input parameter bug of dispatch_gmm_combine_decode[RFC: issue 5476] (#5932)

2026-01-21 09:26:40 +08:00