diff --git a/ggml b/ggml
index 4bf5f600..4e86b56f 160000
--- a/ggml
+++ b/ggml
@@ -1 +1 @@
-Subproject commit 4bf5f6000653b7881d00963cd6ddb665ccd62a8d
+Subproject commit 4e86b56f1658203d7f403a3ab8b5da852d7f6cd0
diff --git a/src/model/common/ggml_block.hpp b/src/model/common/ggml_block.hpp
index 991b88f2..560f49f8 100644
--- a/src/model/common/ggml_block.hpp
+++ b/src/model/common/ggml_block.hpp
@@ -214,7 +214,7 @@ public:
             if (ctx->backend != nullptr) {
                 ggml_tensor* fp8_matmul = ggml_mul_mat(ctx->ggml_ctx, w, x);
                 if (force_prec_f32) {
-                    ggml_mul_mat_set_prec(fp8_matmul, GGML_PREC_F32);
+                    ggml_prec_set_acc(fp8_matmul, GGML_PREC_F32);
                 }
                 supports_fp8_matmul = ggml_backend_supports_op(ctx->backend, fp8_matmul);
             }
diff --git a/src/model/te/llm.hpp b/src/model/te/llm.hpp
index 20521cc7..5adebe61 100644
--- a/src/model/te/llm.hpp
+++ b/src/model/te/llm.hpp
@@ -1717,7 +1717,7 @@ namespace LLM {
                 v_attn      = ggml_reshape_3d(ctx->ggml_ctx, v_attn, n_token, head_dim, num_kv_heads * N);
 
                 auto kq = ggml_mul_mat(ctx->ggml_ctx, k, q);
-                ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
+                ggml_prec_set_acc(kq, GGML_PREC_F32);
                 kq = ggml_scale_inplace(ctx->ggml_ctx, kq, 1.0f / std::sqrt(static_cast<float>(head_dim)));
                 if (attention_mask != nullptr) {
                     kq = ggml_add_inplace(ctx->ggml_ctx, kq, attention_mask);