llama : skip unused tensors

ggml-ci

Author: ggerganov

src/llama-model.cpp

@@ -1424,6 +1424,14 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
                 throw std::runtime_error(format("missing tensor info mapping for %s", tn.str().c_str()));
             }
 
+            // skip unused tensors
+            if (info.op == GGML_OP_NONE) {
+                LLAMA_LOG_WARN("model has unused tensor %s -- ignoring\n", tn.str().c_str());
+                ml.n_created++;
+
+                return nullptr;
+            }
+
             // tensors with "bias" suffix are always used with GGML_OP_ADD
             ggml_op op;
             bool bias = tn.suffix != nullptr && strcmp(tn.suffix, "bias") == 0;
@@ -1460,15 +1468,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
                     GGML_ABORT("invalid layer %d for tensor %s", info.layer, tn.str().c_str());
             }
 
-            ggml_backend_buffer_type_t buft;
-
-            if (op == GGML_OP_NONE) {
-                LLAMA_LOG_WARN("tensor %s has no operation assigned, using host buffer\n", tn.str().c_str());
-                auto * cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
-                buft = ggml_backend_dev_buffer_type(cpu_dev);
-            } else {
-                buft = select_weight_buft(hparams, t_meta, op, *buft_list);
-            }
+            ggml_backend_buffer_type_t buft = select_weight_buft(hparams, t_meta, op, *buft_list);
             if (!buft) {
                 throw std::runtime_error(format("failed to find a compatible buffer type for tensor %s", tn.str().c_str()));
             }