show available RAM, fixed SD vae tiling noise

Author: LostRuins

koboldcpp.py

@@ -1213,10 +1213,6 @@ def fetch_gpu_properties(testCL,testCU,testVK):
             MaxMemory[0] = max(lowestclmem,MaxMemory[0])
         except Exception:
             pass
-    if MaxMemory[0]>0:
-        print(f"Detected Free GPU Memory: {int(MaxMemory[0]/1024/1024)} MB (Set GPU layers manually if incorrect)")
-    else:
-        print("Unable to determine GPU Memory")
     return
 
 def auto_set_backend_cli():
@@ -6672,6 +6668,17 @@ def kcpp_main_process(launch_args, g_memory=None, gui_launcher=False):
         args.threads = get_default_threads()
         print(f"Auto Set Threads: {args.threads}")
 
+    if MaxMemory[0]>0:
+        print(f"Detected Available GPU Memory: {int(MaxMemory[0]/1024/1024)} MB")
+    else:
+        print("Unable to determine GPU Memory")
+    try:
+        import psutil
+        vmem = psutil.virtual_memory()
+        print(f"Detected Available RAM: {int(vmem.available/1024/1024)} MB")
+    except Exception:
+        print("Unable to determine available RAM")
+
     init_library() # Note: if blas does not exist and is enabled, program will crash.
     print("==========")
     time.sleep(1)

otherarch/sdcpp/ggml_extend.hpp

@@ -604,6 +604,7 @@ typedef std::function<void(ggml_tensor*, ggml_tensor*, bool)> on_tile_process;
 
 // Tiling
 __STATIC_INLINE__ void sd_tiling(ggml_tensor* input, ggml_tensor* output, const int scale, const int tile_size, const float tile_overlap_factor, on_tile_process on_processing) {
+    output = ggml_set_f32(output, 0);
     int input_width   = (int)input->ne[0];
     int input_height  = (int)input->ne[1];
     int output_width  = (int)output->ne[0];