Merge pull request #9 from EnragedAntelope/claude/modelopt-comfyui-integration-guide-011CUyFEovRwnw5Ti462aaMV

Claude/modelopt comfyui integration guide 011 c uy f eov rwnw5 ti462aa mv

nodes/quantizer.py

@@ -134,11 +134,21 @@ class ModelOptQuantizeUNet:
             print(f"\nIntrospecting model architecture...")
             model_info = introspect_diffusion_model(diffusion_model)
             print(f"  Architecture: {model_info['architecture']}")
+            print(f"  Parameters: {model_info['param_count_billions']:.2f}B ({model_info['param_count']:,})")
             print(f"  Y dimension: {model_info['y_dim']} ({model_info.get('y_dim_source', 'not detected')})")
             print(f"  Context dimension: {model_info['context_dim']}")
             print(f"  Latent format: {model_info['latent_channels']}x{model_info['latent_spatial']}x{model_info['latent_spatial']}")
             print(f"  Has y parameter: {model_info['has_y_param']}")
 
+            # Debug: Print found attributes
+            if model_info.get('important_attributes_found'):
+                found = [k for k, v in model_info['important_attributes_found'].items() if v]
+                print(f"  Found attributes: {', '.join(found) if found else 'None'}")
+
+            # Debug: Print detection events
+            if model_info.get('detected_attributes'):
+                print(f"  Detection events: {', '.join(model_info['detected_attributes'])}")
+
             # Parse skip layers
             skip_layer_list = []
             if skip_layers:
@@ -172,6 +182,13 @@ class ModelOptQuantizeUNet:
             print(f"  Using base config: {precision.upper()}")
             print(f"  Algorithm: {quant_cfg.get('algorithm', 'default')}")
 
+            # Debug: Print config details to troubleshoot "Inserted 0 quantizers"
+            if "quant_cfg" in quant_cfg:
+                print(f"  Config keys: {list(quant_cfg['quant_cfg'].keys())}")
+                # Check if there are any wildcard patterns
+                wildcards = [k for k in quant_cfg['quant_cfg'].keys() if '*' in k]
+                print(f"  Wildcard patterns: {wildcards}")
+
             # Customize config to skip certain layers if specified
             if skip_layer_list:
                 import copy

nodes/utils.py

@@ -297,8 +297,19 @@ def introspect_diffusion_model(model):
         "has_y_param": False,
         "model_channels": None,
         "num_heads": None,
+        "detected_attributes": [],  # Track what we actually found
     }
 
+    # DEBUG: List all top-level attributes to help troubleshoot
+    all_attrs = [attr for attr in dir(model) if not attr.startswith('_')]
+    info["all_model_attributes"] = all_attrs
+
+    # Check for specific important attributes
+    important_attrs = ['adm_in_channels', 'y_dim', 'label_emb', 'time_embed', 'model_channels',
+                       'context_dim', 'in_channels', 'conv_in', 'num_heads', 'num_res_blocks']
+    found_attrs = {attr: hasattr(model, attr) for attr in important_attrs}
+    info["important_attributes_found"] = found_attrs
+
     # Detect y dimension (pooled embeddings) from various attributes
     y_dim_candidates = []
 
@@ -405,4 +416,70 @@ def introspect_diffusion_model(model):
     elif info["latent_channels"] == 4:
         info["latent_spatial"] = 128  # SDXL/SD1.5 use 4 channels
 
+    # Count parameters to help identify model
+    param_count = sum(p.numel() for p in model.parameters())
+    info["param_count"] = param_count
+    info["param_count_billions"] = param_count / 1e9
+
+    # Use parameter count as additional signal
+    # SDXL: ~2.5-2.6B parameters
+    # SD1.5: ~900M parameters
+    # SD3: ~2B parameters
+    if param_count > 2e9 and info["y_dim"] is None:
+        # Likely SDXL but failed to detect y_dim
+        # Try to infer y_dim = 2816 for SDXL
+        info["y_dim"] = 2816
+        info["y_dim_source"] = "inferred_from_param_count_SDXL"
+        info["has_y_param"] = True
+        info["architecture"] = "SDXL-like (inferred from 2.5B params)"
+
+    # FINAL CHECK: Test forward pass to definitively determine if y is needed
+    # This is the most reliable method
+    if info["y_dim"] is None or not info["has_y_param"]:
+        print(f"  Attempting test forward pass to detect y parameter requirement...")
+        try:
+            with torch.no_grad():
+                # Create dummy inputs
+                device = next(model.parameters()).device
+                dtype = next(model.parameters()).dtype
+
+                test_latent = torch.randn(1, info["latent_channels"], 64, 64, device=device, dtype=dtype)
+                test_timestep = torch.tensor([999.0], device=device, dtype=dtype)
+                test_context = torch.randn(1, 77, info["context_dim"], device=device, dtype=dtype)
+
+                # Try without y first
+                try:
+                    _ = model(test_latent, test_timestep, test_context)
+                    # Success without y - doesn't need it
+                    info["has_y_param"] = False
+                    info["detected_attributes"].append("test_forward_pass_no_y_SUCCESS")
+                except TypeError as e:
+                    error_str = str(e)
+                    if "must specify y" in error_str or "class-conditional" in error_str:
+                        # Needs y parameter!
+                        info["has_y_param"] = True
+                        # Try to infer y_dim
+                        if info["y_dim"] is None:
+                            # Try common SDXL value
+                            info["y_dim"] = 2816
+                            info["y_dim_source"] = "inferred_from_forward_pass_error"
+                        info["detected_attributes"].append("test_forward_pass_NEEDS_Y")
+                    elif "unexpected keyword argument" in error_str:
+                        # Model uses different signature
+                        info["detected_attributes"].append(f"test_forward_pass_different_signature: {error_str}")
+                except Exception as e:
+                    info["detected_attributes"].append(f"test_forward_pass_failed: {str(e)[:100]}")
+        except Exception as e:
+            info["detected_attributes"].append(f"test_forward_pass_setup_failed: {str(e)[:100]}")
+
+    # Final architecture determination
+    if info["y_dim"] == 2816 or info["param_count_billions"] > 2.0:
+        info["architecture"] = "SDXL-like"
+    elif info["y_dim"] == 1280:
+        info["architecture"] = "SDXL-compact"
+    elif not info["has_y_param"] and info["context_dim"] == 768:
+        info["architecture"] = "SD1.5-like"
+    elif info["context_dim"] == 4096:
+        info["architecture"] = "SD3-like"
+
     return info

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "comfy-modelopt"
-version = "0.2.0"
+version = "0.2.1"
 description = "NVIDIA ModelOpt integration for ComfyUI - Quantize and optimize diffusion models with INT8/FP8/INT4"
 readme = "README.md"
 requires-python = ">=3.10,<3.13"
@@ -9,18 +9,6 @@ authors = [
     {name = "ComfyUI ModelOpt Contributors"}
 ]
 keywords = ["comfyui", "nvidia", "modelopt", "tensorrt", "quantization", "stable-diffusion", "optimization"]
-classifiers = [
-    "Development Status :: 4 - Beta",
-    "Intended Audience :: Developers",
-    "Intended Audience :: Science/Research",
-    "License :: OSI Approved :: MIT License",
-    "Programming Language :: Python :: 3",
-    "Programming Language :: Python :: 3.10",
-    "Programming Language :: Python :: 3.11",
-    "Programming Language :: Python :: 3.12",
-    "Topic :: Scientific/Engineering :: Artificial Intelligence",
-    "Topic :: Software Development :: Libraries :: Python Modules",
-]
 
 dependencies = [
     "torch>=2.0.0",
@@ -73,8 +61,8 @@ target-version = ['py310', 'py311', 'py312']
 # ComfyUI Manager metadata
 display_name = "NVIDIA ModelOpt"
 description = "Quantize and optimize diffusion models with NVIDIA ModelOpt (INT8/FP8/INT4)"
-author = "ComfyUI ModelOpt Contributors"
-version = "0.2.0"
+author = "enragedantelope"
+version = "0.2.1"
 node_list = [
     "ModelOptUNetLoader",
     "ModelOptQuantizeUNet",