From 67b9b0e7f6ce45d929a4411907d3c48ec719e81c Mon Sep 17 00:00:00 2001 From: helanfxz <126638465+helanfxz@users.noreply.github.com> Date: Wed, 22 Jul 2026 10:55:44 +0800 Subject: [PATCH] llama-arch: fix DeepSeek4 APE tensor op (#25945) --- src/llama-arch.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp index 89fa097fe..e5a9a29c7 100644 --- a/src/llama-arch.cpp +++ b/src/llama-arch.cpp @@ -666,7 +666,7 @@ static const std::map LLM_TENSOR_INFOS = { {LLM_TENSOR_HC_FFN_SCALE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}}, {LLM_TENSOR_ATTN_COMPRESSOR_WKV, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}}, {LLM_TENSOR_ATTN_COMPRESSOR_WGATE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}}, - {LLM_TENSOR_ATTN_COMPRESSOR_APE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}}, + {LLM_TENSOR_ATTN_COMPRESSOR_APE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_GET_ROWS}}, {LLM_TENSOR_ATTN_COMPRESSOR_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}}, {LLM_TENSOR_ATTN_K_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}}, {LLM_TENSOR_ATTN_V_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}}, @@ -833,7 +833,7 @@ static const std::map LLM_TENSOR_INFOS = { {LLM_TENSOR_INDEXER_ATTN_Q_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}}, {LLM_TENSOR_INDEXER_COMPRESSOR_WKV, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}}, {LLM_TENSOR_INDEXER_COMPRESSOR_WGATE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}}, - {LLM_TENSOR_INDEXER_COMPRESSOR_APE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}}, + {LLM_TENSOR_INDEXER_COMPRESSOR_APE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_GET_ROWS}}, {LLM_TENSOR_INDEXER_COMPRESSOR_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}}, {LLM_TENSOR_FFN_GATE_TID2EID, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_GET_ROWS}}, {LLM_TENSOR_NEXTN_PROJ_PRE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},