feat: throttle model loading progress updates (#782)

Some terminals have slow display latency, so frequent output
during model loading can actually slow down the process.

Also, since tensor loading times can vary a lot, the progress
display now shows the average across past iterations instead
of just the last one.

.github/workflows/build.yml

@@ -163,7 +163,7 @@ jobs:
           - build: "avx512"
             defines: "-DGGML_NATIVE=OFF -DGGML_AVX512=ON -DGGML_AVX=ON -DGGML_AVX2=ON -DSD_BUILD_SHARED_LIBS=ON"
           - build: "cuda12"
-            defines: "-DSD_CUDA=ON -DSD_BUILD_SHARED_LIBS=ON -DCMAKE_CUDA_ARCHITECTURES=90;89;80;75"
+            defines: "-DSD_CUDA=ON -DSD_BUILD_SHARED_LIBS=ON -DCMAKE_CUDA_ARCHITECTURES=90;89;86;80;75"
           # - build: "rocm5.5"
           #   defines: '-G Ninja -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DSD_HIPBLAS=ON -DCMAKE_BUILD_TYPE=Release -DAMDGPU_TARGETS="gfx1100;gfx1102;gfx1030" -DSD_BUILD_SHARED_LIBS=ON'
           - build: 'vulkan'

model.cpp

@@ -1942,8 +1942,11 @@ bool ModelLoader::load_tensors(on_new_tensor_cb_t on_new_tensor_cb, ggml_backend
             return true;
         };
         int tensor_count = 0;
-        int64_t t1       = ggml_time_ms();
-        bool partial     = false;
+        int64_t t0       = ggml_time_ms();
+        int64_t t1       = t0;
+        bool partial     = true;
+        int tensor_max   = (int)processed_tensor_storages.size();
+        pretty_progress(0, tensor_max, 0.0f);
         for (auto& tensor_storage : processed_tensor_storages) {
             if (tensor_storage.file_index != file_index) {
                 ++tensor_count;
@@ -2046,21 +2049,29 @@ bool ModelLoader::load_tensors(on_new_tensor_cb_t on_new_tensor_cb, ggml_backend
                     ggml_backend_tensor_set(dst_tensor, convert_buffer.data(), 0, ggml_nbytes(dst_tensor));
                 }
             }
-            size_t tensor_max = processed_tensor_storages.size();
-            int64_t t2        = ggml_time_ms();
-            pretty_progress(++tensor_count, tensor_max, (t2 - t1) / 1000.0f);
-            t1      = t2;
-            partial = tensor_count != tensor_max;
+            ++tensor_count;
+            int64_t t2 = ggml_time_ms();
+            if ((t2 - t1) >= 200) {
+                t1 = t2;
+                pretty_progress(tensor_count, tensor_max, (t1 - t0) / (1000.0f * tensor_count));
+                partial = tensor_count != tensor_max;
+            }
+        }
+
+        if (partial) {
+            if (tensor_count >= 1) {
+                t1 = ggml_time_ms();
+                pretty_progress(tensor_count, tensor_max, (t1 - t0) / (1000.0f * tensor_count));
+            }
+            if (tensor_count < tensor_max) {
+                printf("\n");
+            }
         }
 
         if (zip != NULL) {
             zip_close(zip);
         }
 
-        if (partial) {
-            printf("\n");
-        }
-
         if (!success) {
             break;
         }