Bump ComfyUI version to v0.3.18

Instructions shouldn't recommend to run comfyui with --listen

README.md

@@ -46,7 +46,7 @@ ComfyUI lets you design and execute advanced stable diffusion pipelines using a
 #### [Manual Install](#manual-install-windows-linux)
 Supports all operating systems and GPU types (NVIDIA, AMD, Intel, Apple Silicon, Ascend).
 
-## Examples
+## [Examples](https://comfyanonymous.github.io/ComfyUI_examples/)
 See what ComfyUI can do with the [example workflows](https://comfyanonymous.github.io/ComfyUI_examples/).
 
 
@@ -68,6 +68,7 @@ See what ComfyUI can do with the [example workflows](https://comfyanonymous.gith
    - [LTX-Video](https://comfyanonymous.github.io/ComfyUI_examples/ltxv/)
    - [Hunyuan Video](https://comfyanonymous.github.io/ComfyUI_examples/hunyuan_video/)
    - [Nvidia Cosmos](https://comfyanonymous.github.io/ComfyUI_examples/cosmos/)
+   - [Wan 2.1](https://comfyanonymous.github.io/ComfyUI_examples/wan/)
 - [Stable Audio](https://comfyanonymous.github.io/ComfyUI_examples/audio/)
 - Asynchronous Queue system
 - Many optimizations: Only re-executes the parts of the workflow that changes between executions.
@@ -260,6 +261,13 @@ For models compatible with Ascend Extension for PyTorch (torch_npu). To get star
 3. Next, install the necessary packages for torch-npu by adhering to the platform-specific instructions on the [Installation](https://ascend.github.io/docs/sources/pytorch/install.html#pytorch) page.
 4. Finally, adhere to the [ComfyUI manual installation](#manual-install-windows-linux) guide for Linux. Once all components are installed, you can run ComfyUI as described earlier.
 
+#### Cambricon MLUs
+
+For models compatible with Cambricon Extension for PyTorch (torch_mlu). Here's a step-by-step guide tailored to your platform and installation method:
+
+1. Install the Cambricon CNToolkit by adhering to the platform-specific instructions on the [Installation](https://www.cambricon.com/docs/sdk_1.15.0/cntoolkit_3.7.2/cntoolkit_install_3.7.2/index.html)
+2. Next, install the PyTorch(torch_mlu) following the instructions on the [Installation](https://www.cambricon.com/docs/sdk_1.15.0/cambricon_pytorch_1.17.0/user_guide_1.9/index.html)
+3. Launch ComfyUI by running `python main.py`
 
 # Running
 

comfy/ldm/wan/model.py

@@ -18,7 +18,7 @@ def sinusoidal_embedding_1d(dim, position):
     # preprocess
     assert dim % 2 == 0
     half = dim // 2
-    position = position.type(torch.float64)
+    position = position.type(torch.float32)
 
     # calculation
     sinusoid = torch.outer(
@@ -353,7 +353,7 @@ class WanModel(torch.nn.Module):
 
         # embeddings
         self.patch_embedding = operations.Conv3d(
-            in_dim, dim, kernel_size=patch_size, stride=patch_size, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
+            in_dim, dim, kernel_size=patch_size, stride=patch_size, device=operation_settings.get("device"), dtype=torch.float32)
         self.text_embedding = nn.Sequential(
             operations.Linear(text_dim, dim, device=operation_settings.get("device"), dtype=operation_settings.get("dtype")), nn.GELU(approximate='tanh'),
             operations.Linear(dim, dim, device=operation_settings.get("device"), dtype=operation_settings.get("dtype")))
@@ -411,7 +411,7 @@ class WanModel(torch.nn.Module):
                 List of denoised video tensors with original input shapes [C_out, F, H / 8, W / 8]
         """
         # embeddings
-        x = self.patch_embedding(x)
+        x = self.patch_embedding(x.float()).to(x.dtype)
         grid_sizes = x.shape[2:]
         x = x.flatten(2).transpose(1, 2)
 
@@ -421,7 +421,7 @@ class WanModel(torch.nn.Module):
         e0 = self.time_projection(e).unflatten(1, (6, self.dim))
 
         # context
-        context = self.text_embedding(torch.cat([context, context.new_zeros(context.size(0), self.text_len - context.size(1), context.size(2))], dim=1))
+        context = self.text_embedding(context)
 
         if clip_fea is not None and self.img_emb is not None:
             context_clip = self.img_emb(clip_fea)  # bs x 257 x dim

comfy/model_management.py

@@ -95,6 +95,13 @@ try:
 except:
     npu_available = False
 
+try:
+    import torch_mlu  # noqa: F401
+    _ = torch.mlu.device_count()
+    mlu_available = torch.mlu.is_available()
+except:
+    mlu_available = False
+
 if args.cpu:
     cpu_state = CPUState.CPU
 
@@ -112,6 +119,12 @@ def is_ascend_npu():
         return True
     return False
 
+def is_mlu():
+    global mlu_available
+    if mlu_available:
+        return True
+    return False
+
 def get_torch_device():
     global directml_enabled
     global cpu_state
@@ -127,6 +140,8 @@ def get_torch_device():
             return torch.device("xpu", torch.xpu.current_device())
         elif is_ascend_npu():
             return torch.device("npu", torch.npu.current_device())
+        elif is_mlu():
+            return torch.device("mlu", torch.mlu.current_device())
         else:
             return torch.device(torch.cuda.current_device())
 
@@ -153,6 +168,12 @@ def get_total_memory(dev=None, torch_total_too=False):
             _, mem_total_npu = torch.npu.mem_get_info(dev)
             mem_total_torch = mem_reserved
             mem_total = mem_total_npu
+        elif is_mlu():
+            stats = torch.mlu.memory_stats(dev)
+            mem_reserved = stats['reserved_bytes.all.current']
+            _, mem_total_mlu = torch.mlu.mem_get_info(dev)
+            mem_total_torch = mem_reserved
+            mem_total = mem_total_mlu
         else:
             stats = torch.cuda.memory_stats(dev)
             mem_reserved = stats['reserved_bytes.all.current']
@@ -232,7 +253,7 @@ try:
         if torch_version_numeric[0] >= 2:
             if ENABLE_PYTORCH_ATTENTION == False and args.use_split_cross_attention == False and args.use_quad_cross_attention == False:
                 ENABLE_PYTORCH_ATTENTION = True
-    if is_intel_xpu() or is_ascend_npu():
+    if is_intel_xpu() or is_ascend_npu() or is_mlu():
         if args.use_split_cross_attention == False and args.use_quad_cross_attention == False:
             ENABLE_PYTORCH_ATTENTION = True
 except:
@@ -316,6 +337,8 @@ def get_torch_device_name(device):
         return "{} {}".format(device, torch.xpu.get_device_name(device))
     elif is_ascend_npu():
         return "{} {}".format(device, torch.npu.get_device_name(device))
+    elif is_mlu():
+        return "{} {}".format(device, torch.mlu.get_device_name(device))
     else:
         return "CUDA {}: {}".format(device, torch.cuda.get_device_name(device))
 
@@ -905,6 +928,8 @@ def xformers_enabled():
         return False
     if is_ascend_npu():
         return False
+    if is_mlu():
+        return False
     if directml_enabled:
         return False
     return XFORMERS_IS_AVAILABLE
@@ -936,6 +961,8 @@ def pytorch_attention_flash_attention():
             return True
         if is_ascend_npu():
             return True
+        if is_mlu():
+            return True
         if is_amd():
             return True #if you have pytorch attention enabled on AMD it probably supports at least mem efficient attention
     return False
@@ -984,6 +1011,13 @@ def get_free_memory(dev=None, torch_free_too=False):
             mem_free_npu, _ = torch.npu.mem_get_info(dev)
             mem_free_torch = mem_reserved - mem_active
             mem_free_total = mem_free_npu + mem_free_torch
+        elif is_mlu():
+            stats = torch.mlu.memory_stats(dev)
+            mem_active = stats['active_bytes.all.current']
+            mem_reserved = stats['reserved_bytes.all.current']
+            mem_free_mlu, _ = torch.mlu.mem_get_info(dev)
+            mem_free_torch = mem_reserved - mem_active
+            mem_free_total = mem_free_mlu + mem_free_torch
         else:
             stats = torch.cuda.memory_stats(dev)
             mem_active = stats['active_bytes.all.current']
@@ -1053,6 +1087,9 @@ def should_use_fp16(device=None, model_params=0, prioritize_performance=True, ma
     if is_ascend_npu():
         return True
 
+    if is_mlu():
+        return True
+
     if torch.version.hip:
         return True
 
@@ -1121,6 +1158,11 @@ def should_use_bf16(device=None, model_params=0, prioritize_performance=True, ma
             return False
 
     props = torch.cuda.get_device_properties(device)
+
+    if is_mlu():
+        if props.major > 3:
+            return True
+
     if props.major >= 8:
         return True
 

comfy/text_encoders/wan.py

@@ -11,7 +11,7 @@ class UMT5XXlModel(sd1_clip.SDClipModel):
 class UMT5XXlTokenizer(sd1_clip.SDTokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         tokenizer = tokenizer_data.get("spiece_model", None)
-        super().__init__(tokenizer, pad_with_end=False, embedding_size=4096, embedding_key='umt5xxl', tokenizer_class=SPieceTokenizer, has_start_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, pad_token=0)
+        super().__init__(tokenizer, pad_with_end=False, embedding_size=4096, embedding_key='umt5xxl', tokenizer_class=SPieceTokenizer, has_start_token=False, pad_to_max_length=False, max_length=99999999, min_length=512, pad_token=0)
 
     def state_dict(self):
         return {"spiece_model": self.tokenizer.serialize_model()}

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.3.16"
+__version__ = "0.3.18"

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.3.16"
+version = "0.3.18"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.9"