Merge branch 'comfyanonymous:master' into feature/preview-latent

comfy/model_management.py

@@ -1,6 +1,7 @@
 import psutil
 from enum import Enum
 from comfy.cli_args import args
+import torch
 
 class VRAMState(Enum):
     CPU = 0
@@ -33,28 +34,67 @@ if args.directml is not None:
     lowvram_available = False #TODO: need to find a way to get free memory in directml before this can be enabled by default.
 
 try:
-    import torch
-    if directml_enabled:
-        pass #TODO
-    else:
-        try:
-            import intel_extension_for_pytorch as ipex
-            if torch.xpu.is_available():
-                xpu_available = True
-                total_vram = torch.xpu.get_device_properties(torch.xpu.current_device()).total_memory / (1024 * 1024)
-        except:
-            total_vram = torch.cuda.mem_get_info(torch.cuda.current_device())[1] / (1024 * 1024)
-    total_ram = psutil.virtual_memory().total / (1024 * 1024)
-    if not args.normalvram and not args.cpu:
-        if lowvram_available and total_vram <= 4096:
-            print("Trying to enable lowvram mode because your GPU seems to have 4GB or less. If you don't want this use: --normalvram")
-            set_vram_to = VRAMState.LOW_VRAM
-        elif total_vram > total_ram * 1.1 and total_vram > 14336:
-            print("Enabling highvram mode because your GPU has more vram than your computer has ram. If you don't want this use: --normalvram")
-            vram_state = VRAMState.HIGH_VRAM
+    import intel_extension_for_pytorch as ipex
+    if torch.xpu.is_available():
+        xpu_available = True
 except:
     pass
 
+def get_torch_device():
+    global xpu_available
+    global directml_enabled
+    if directml_enabled:
+        global directml_device
+        return directml_device
+    if vram_state == VRAMState.MPS:
+        return torch.device("mps")
+    if vram_state == VRAMState.CPU:
+        return torch.device("cpu")
+    else:
+        if xpu_available:
+            return torch.device("xpu")
+        else:
+            return torch.device(torch.cuda.current_device())
+
+def get_total_memory(dev=None, torch_total_too=False):
+    global xpu_available
+    global directml_enabled
+    if dev is None:
+        dev = get_torch_device()
+
+    if hasattr(dev, 'type') and (dev.type == 'cpu' or dev.type == 'mps'):
+        mem_total = psutil.virtual_memory().total
+        mem_total_torch = mem_total
+    else:
+        if directml_enabled:
+            mem_total = 1024 * 1024 * 1024 #TODO
+            mem_total_torch = mem_total
+        elif xpu_available:
+            mem_total = torch.xpu.get_device_properties(dev).total_memory
+            mem_total_torch = mem_total
+        else:
+            stats = torch.cuda.memory_stats(dev)
+            mem_reserved = stats['reserved_bytes.all.current']
+            _, mem_total_cuda = torch.cuda.mem_get_info(dev)
+            mem_total_torch = mem_reserved
+            mem_total = mem_total_cuda
+
+    if torch_total_too:
+        return (mem_total, mem_total_torch)
+    else:
+        return mem_total
+
+total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
+total_ram = psutil.virtual_memory().total / (1024 * 1024)
+print("Total VRAM {:0.0f} MB, total RAM {:0.0f} MB".format(total_vram, total_ram))
+if not args.normalvram and not args.cpu:
+    if lowvram_available and total_vram <= 4096:
+        print("Trying to enable lowvram mode because your GPU seems to have 4GB or less. If you don't want this use: --normalvram")
+        set_vram_to = VRAMState.LOW_VRAM
+    elif total_vram > total_ram * 1.1 and total_vram > 14336:
+        print("Enabling highvram mode because your GPU has more vram than your computer has ram. If you don't want this use: --normalvram")
+        vram_state = VRAMState.HIGH_VRAM
+
 try:
     OOM_EXCEPTION = torch.cuda.OutOfMemoryError
 except:
@@ -128,29 +168,17 @@ if args.cpu:
 
 print(f"Set vram state to: {vram_state.name}")
 
-def get_torch_device():
-    global xpu_available
-    global directml_enabled
-    if directml_enabled:
-        global directml_device
-        return directml_device
-    if vram_state == VRAMState.MPS:
-        return torch.device("mps")
-    if vram_state == VRAMState.CPU:
-        return torch.device("cpu")
-    else:
-        if xpu_available:
-            return torch.device("xpu")
-        else:
-            return torch.cuda.current_device()
-
 def get_torch_device_name(device):
     if hasattr(device, 'type'):
-        return "{}".format(device.type)
-    return "CUDA {}: {}".format(device, torch.cuda.get_device_name(device))
+        if device.type == "cuda":
+            return "{} {}".format(device, torch.cuda.get_device_name(device))
+        else:
+            return "{}".format(device.type)
+    else:
+        return "CUDA {}: {}".format(device, torch.cuda.get_device_name(device))
 
 try:
-    print("Using device:", get_torch_device_name(get_torch_device()))
+    print("Device:", get_torch_device_name(get_torch_device()))
 except:
     print("Could not pick default device.")
 

folder_paths.py

@@ -1,4 +1,5 @@
 import os
+import time
 
 supported_ckpt_extensions = set(['.ckpt', '.pth', '.safetensors'])
 supported_pt_extensions = set(['.ckpt', '.pt', '.bin', '.pth', '.safetensors'])
@@ -154,7 +155,7 @@ def get_filename_list_(folder_name):
         output_list.update(filter_files_extensions(files, folders[1]))
         output_folders = {**output_folders, **folders_all}
 
-    return (sorted(list(output_list)), output_folders)
+    return (sorted(list(output_list)), output_folders, time.perf_counter())
 
 def cached_filename_list_(folder_name):
     global filename_list_cache
@@ -162,6 +163,8 @@ def cached_filename_list_(folder_name):
     if folder_name not in filename_list_cache:
         return None
     out = filename_list_cache[folder_name]
+    if time.perf_counter() < (out[2] + 0.5):
+        return out
     for x in out[1]:
         time_modified = out[1][x]
         folder = x
@@ -170,8 +173,9 @@ def cached_filename_list_(folder_name):
 
     folders = folder_names_and_paths[folder_name]
     for x in folders[0]:
-        if x not in out[1]:
-            return None
+        if os.path.isdir(x):
+            if x not in out[1]:
+                return None
 
     return out
 

server.py

@@ -23,6 +23,7 @@ except ImportError:
 import mimetypes
 from comfy.cli_args import args
 import comfy.utils
+import comfy.model_management
 
 @web.middleware
 async def cache_control(request: web.Request, handler):
@@ -280,6 +281,27 @@ class PromptServer():
                 return web.Response(status=404)
             return web.json_response(dt["__metadata__"])
 
+        @routes.get("/system_stats")
+        async def get_queue(request):
+            device = comfy.model_management.get_torch_device()
+            device_name = comfy.model_management.get_torch_device_name(device)
+            vram_total, torch_vram_total = comfy.model_management.get_total_memory(device, torch_total_too=True)
+            vram_free, torch_vram_free = comfy.model_management.get_free_memory(device, torch_free_too=True)
+            system_stats = {
+                "devices": [
+                    {
+                        "name": device_name,
+                        "type": device.type,
+                        "index": device.index,
+                        "vram_total": vram_total,
+                        "vram_free": vram_free,
+                        "torch_vram_total": torch_vram_total,
+                        "torch_vram_free": torch_vram_free,
+                    }
+                ]
+            }
+            return web.json_response(system_stats)
+
         @routes.get("/prompt")
         async def get_prompt(request):
             return web.json_response(self.get_queue_info())