From 277b34289aca77437ddb378962cf5d3f3d8160c2 Mon Sep 17 00:00:00 2001 From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com> Date: Wed, 8 Jul 2026 23:25:07 +0000 Subject: [PATCH 1/2] Initial plan From 5eda23ababc32e47ac04570e62ca19a8227640d3 Mon Sep 17 00:00:00 2001 From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com> Date: Wed, 8 Jul 2026 23:29:27 +0000 Subject: [PATCH 2/2] Fix SDPA GQA assertion failure when q/kv heads are equal (torch-nightly) --- onnxscript/function_libs/torch_lib/ops/nn.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/onnxscript/function_libs/torch_lib/ops/nn.py b/onnxscript/function_libs/torch_lib/ops/nn.py index b1686f7418..a37c48f259 100644 --- a/onnxscript/function_libs/torch_lib/ops/nn.py +++ b/onnxscript/function_libs/torch_lib/ops/nn.py @@ -1835,7 +1835,7 @@ def aten_scaled_dot_product_attention( if is_causal: attn_mask = _causal_attention_mask(query, key) - if enable_gqa: + if enable_gqa and query.shape[1] != key.shape[1]: key, value = _attention_repeat_kv_for_group_query(query, key, value) if attn_mask is None: