diff --git a/src/core/ggml_compat.cpp b/src/core/ggml_compat.cpp
index 061b33706..f6e8e7fc3 100644
--- a/src/core/ggml_compat.cpp
+++ b/src/core/ggml_compat.cpp
@@ -151,17 +151,6 @@ bool ggml_graph_cpy_use_counts(const ggml_cgraph* src, ggml_cgraph* dst) {
     }
     return true;
 }
-
-ggml_status ggml_backend_graph_compute_async_range(ggml_backend_t backend, ggml_cgraph* graph, int first, int last) {
-    if (!backend || !graph || first < 0 || last < first || last > graph->n_nodes) {
-        return GGML_STATUS_FAILED;
-    }
-    if (first == last) {
-        return GGML_STATUS_SUCCESS;
-    }
-    auto view = ggml_graph_view(graph, first, last);
-    return ggml_backend_graph_compute_async(backend, &view);
-}
 #endif
 
 #ifdef SD_USE_UPSTREAM_GGML
diff --git a/src/core/ggml_compat.h b/src/core/ggml_compat.h
index 9f64262fb..9a367a8bc 100644
--- a/src/core/ggml_compat.h
+++ b/src/core/ggml_compat.h
@@ -29,13 +29,6 @@ bool ggml_graph_cpy_mapped(const struct ggml_cgraph* src, struct ggml_cgraph* ds
 // This preserves external consumers when constructing an inference graph segment.
 bool ggml_graph_cpy_use_counts(const struct ggml_cgraph* src, struct ggml_cgraph* dst);
 
-// Executes nodes [first, last) with the original graph's use counts. No graph ownership
-// is transferred. Tensor storage must already be allocated; invalid bounds fail.
-enum ggml_status ggml_backend_graph_compute_async_range(ggml_backend_t backend,
-                                                        struct ggml_cgraph* graph,
-                                                        int first,
-                                                        int last);
-
 #else
 #error "Patched ggml requires ggml-graph.h; select SD_USE_UPSTREAM_GGML for upstream compatibility"
 #endif
diff --git a/src/core/ggml_extend_backend.cpp b/src/core/ggml_extend_backend.cpp
index 70f2ae2ec..fa4a3efcb 100644
--- a/src/core/ggml_extend_backend.cpp
+++ b/src/core/ggml_extend_backend.cpp
@@ -638,50 +638,6 @@ void sd_backend_clear_buffer_cache(ggml_backend_t backend) {
     }
 }
 
-ggml_status sd_backend_graph_compute_with_eval_callback(ggml_backend_t backend,
-                                                        ggml_cgraph* gf,
-                                                        sd_graph_eval_callback_t callback_eval,
-                                                        void* callback_eval_user_data) {
-    if (callback_eval == nullptr) {
-        return ggml_backend_graph_compute(backend, gf);
-    }
-
-    ggml_status status = GGML_STATUS_SUCCESS;
-    const int n_nodes  = ggml_graph_n_nodes(gf);
-    bool stopped       = false;
-
-    for (int j0 = 0; j0 < n_nodes; ++j0) {
-        ggml_tensor* t = ggml_graph_node(gf, j0);
-        bool need      = callback_eval(t, true, callback_eval_user_data);
-        int j1         = j0;
-
-        while (!need && j1 < n_nodes - 1) {
-            t    = ggml_graph_node(gf, ++j1);
-            need = callback_eval(t, true, callback_eval_user_data);
-        }
-
-        status = ggml_backend_graph_compute_async_range(backend, gf, j0, j1 + 1);
-        if (status != GGML_STATUS_SUCCESS) {
-            break;
-        }
-
-        ggml_backend_synchronize(backend);
-
-        if (need && !callback_eval(t, false, callback_eval_user_data)) {
-            stopped = true;
-            break;
-        }
-
-        j0 = j1;
-    }
-
-    ggml_backend_synchronize(backend);
-    if (stopped && status == GGML_STATUS_SUCCESS) {
-        status = GGML_STATUS_ABORTED;
-    }
-    return status;
-}
-
 const char* sd_get_system_info() {
     static std::string cache_info = []() -> std::string {
         ggml_backend_load_all_once();
diff --git a/src/core/ggml_extend_backend.h b/src/core/ggml_extend_backend.h
index 037ef0d86..a535ab984 100644
--- a/src/core/ggml_extend_backend.h
+++ b/src/core/ggml_extend_backend.h
@@ -99,10 +99,6 @@ bool sd_backend_cpu_set_n_threads(ggml_backend_t backend_cpu, int n_threads);
 void sd_backend_set_buffer_cache_enabled(ggml_backend_t backend, bool enabled);
 void sd_backend_clear_buffer_cache(ggml_backend_t backend);
 
-ggml_status sd_backend_graph_compute_with_eval_callback(ggml_backend_t backend,
-                                                        ggml_cgraph* gf,
-                                                        sd_graph_eval_callback_t callback_eval,
-                                                        void* callback_eval_user_data);
 std::string sd_backend_resolve_name(const std::string& name);
 bool sd_parse_backend_assignment(const std::string& spec, SDBackendAssignment* assignment, std::string* error);
 bool sd_parse_flash_attention_assignment(const std::string& spec, SDBackendAssignment* assignment, std::string* error);