chore: run ruff check --fix

.github/workflows/lint.yml

@@ -47,7 +47,7 @@ jobs:
       - name: Install linting tools
         run: |
           pip install --upgrade pip
-          pip install black mypy
+          pip install black mypy ruff
 
       - name: Check Python formatting with black
         run: black --check python/ tests/ examples/
@@ -56,3 +56,7 @@ jobs:
       - name: Run mypy type checking
         run: mypy python/
         continue-on-error: true
+
+      - name: Run ruff
+        run: ruff
+        continue-on-error: true

examples/compare_quantization_sizes.py

@@ -7,7 +7,6 @@
 """
 
 import tempfile
-import os
 import numpy as np
 from pathlib import Path
 
@@ -122,9 +121,9 @@ def compare_sizes(size_multiplier=1):
     print("=" * 80)
     print("QUANTIZATION FILE SIZE COMPARISON")
     print("=" * 80)
-    print(f"\nModel Configuration:")
+    print("\nModel Configuration:")
     print(f"  Base size: {64 * size_multiplier} x {64 * size_multiplier}")
-    print(f"  Layers: 2 fully-connected layers (matmul + relu + matmul)")
+    print("  Layers: 2 fully-connected layers (matmul + relu + matmul)")
     print(f"  Total parameters: ~{2 * (64 * size_multiplier) ** 2:,}")
     print()
 
@@ -226,7 +225,7 @@ def compare_sizes(size_multiplier=1):
 
     if size_multiplier > 1:
         print(f"\n[INFO] Using size multiplier: {size_multiplier}x")
-        print(f"[INFO] This will create larger models for more realistic comparison\n")
+        print("[INFO] This will create larger models for more realistic comparison\n")
 
     results = compare_sizes(size_multiplier)
 

examples/kv_cache_device_tensors.py

@@ -133,7 +133,7 @@ def run_with_device_tensors(ctx, graph, num_steps=50):
         past_key_device.write(np.zeros(kv_shape, dtype=np.float32))
         past_value_device.write(np.zeros(kv_shape, dtype=np.float32))
 
-        print(f"  Device tensors created:")
+        print("  Device tensors created:")
         print(f"    KV cache: {past_key_device}")
 
         start = time.time()
@@ -177,7 +177,7 @@ def run_with_device_tensors(ctx, graph, num_steps=50):
 
     except Exception as e:
         print(f"  Device tensor execution failed: {e}")
-        print(f"  Note: This may be expected if device tensors are not fully supported yet")
+        print("  Note: This may be expected if device tensors are not fully supported yet")
         return None
 
 
@@ -198,17 +198,17 @@ def main():
     # Build decoder graph
     print("Building decoder graph...")
     graph = build_decoder_graph(ctx)
-    print(f"  Graph built successfully")
+    print("  Graph built successfully")
     print()
 
     # Configuration
     num_steps = 50
 
-    print(f"Configuration:")
-    print(f"  Batch size: 1")
-    print(f"  Sequence length: 128")
-    print(f"  Hidden dim: 256")
-    print(f"  Num heads: 4")
+    print("Configuration:")
+    print("  Batch size: 1")
+    print("  Sequence length: 128")
+    print("  Hidden dim: 256")
+    print("  Num heads: 4")
     print(f"  Decode steps: {num_steps}")
     print()
     print("-" * 70)
@@ -240,7 +240,7 @@ def main():
         print("Note: Current implementation does host round-trips for both modes.")
         print("True zero-copy execution will show larger speedups on GPU/NPU.")
     else:
-        print(f"Device tensors: Not available")
+        print("Device tensors: Not available")
     print()
     print("This demo shows the API for device tensors.")
     print("Full zero-copy execution coming soon!")

examples/minilm_embeddings.py

@@ -337,16 +337,16 @@ def _load_webnn_model(self):
             print(f"[INFO] Downloading model from Hugging Face Hub: {self.model_id}")
             hub = webnn.Hub()
             model_files = hub.download_model(self.model_id)
-            print(f"[OK] Model downloaded")
+            print("[OK] Model downloaded")
 
         # Load graph from files
-        print(f"[INFO] Loading graph...")
+        print("[INFO] Loading graph...")
         graph = webnn.MLGraph.load(
             model_files['graph'],
             manifest_path=model_files['manifest'],
             weights_path=model_files['weights']
         )
-        print(f"[OK] Graph loaded successfully")
+        print("[OK] Graph loaded successfully")
         print(f"    - Operand count: {graph.operand_count}")
         print(f"    - Operation count: {graph.operation_count}")
         print(f"    - Inputs: {graph.get_input_names()}")
@@ -677,7 +677,7 @@ def main():
     except Exception as e:
         error_msg = str(e)
         if "404" in error_msg or "Entry Not Found" in error_msg:
-            print(f"[WARNING] Model not found on Hugging Face Hub")
+            print("[WARNING] Model not found on Hugging Face Hub")
             print(
                 "[INFO] The model may not be uploaded yet to: tarekziade/all-MiniLM-L6-v2-webnn"
             )

examples/mobilenetv2_from_hub.py

@@ -35,9 +35,9 @@ def load_imagenet_labels():
         # Download from GitHub if not found
         import urllib.request
         url = "https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt"
-        print(f"   [DOWNLOAD] ImageNet labels from GitHub...")
+        print("   [DOWNLOAD] ImageNet labels from GitHub...")
         urllib.request.urlretrieve(url, labels_file)
-        print(f"   [OK] Labels downloaded")
+        print("   [OK] Labels downloaded")
 
     with open(labels_file) as f:
         return [line.strip() for line in f]

examples/mobilenetv2_real.py

@@ -17,7 +17,6 @@
 import time
 import argparse
 from pathlib import Path
-import urllib.request
 
 try:
     import numpy as np

examples/save_load_example.py

@@ -66,7 +66,7 @@ def main():
     save_path = "example_graph.webnn"
     print(f"\n5. Saving graph to '{save_path}'...")
     graph.save(save_path)
-    print(f"   Graph saved successfully")
+    print("   Graph saved successfully")
 
     # Check file size
     file_size = os.path.getsize(save_path)
@@ -114,7 +114,7 @@ def main():
             print(f"... ({len(content) - 500} more characters)")
 
     # Cleanup
-    print(f"\n11. Cleaning up...")
+    print("\n11. Cleaning up...")
     os.remove(save_path)
     print(f"   Removed '{save_path}'")
 

examples/test_quantization_roundtrip.py

@@ -45,45 +45,45 @@ def test_quantization_roundtrip(quant_dtype="int8", quant_level_name="int8"):
 
     # Quantize the input
     quantized = builder.quantize_linear(x, scale, zero_point)
-    print(f"  Created quantize_linear node")
-    print(f"    Input dtype: float32")
+    print("  Created quantize_linear node")
+    print("    Input dtype: float32")
     print(f"    Output dtype: {quant_dtype}")
 
     # Dequantize for computation
     output = builder.dequantize_linear(quantized, scale, zero_point)
-    print(f"  Created dequantize_linear node")
+    print("  Created dequantize_linear node")
     print(f"    Input dtype: {quant_dtype}")
-    print(f"    Output dtype: float32")
+    print("    Output dtype: float32")
 
     # Build graph
     graph = builder.build({"output": output})
-    print(f"  Graph built successfully")
+    print("  Graph built successfully")
 
     # Step 2: Save with quantized=True marker
     with tempfile.TemporaryDirectory() as tmpdir:
         model_path = Path(tmpdir) / "quantized_model.webnn"
         print(f"\n[STEP 2] Saving quantized graph to {model_path}...")
         graph.save(str(model_path), quantized=True)
-        print(f"  Model saved with quantized=True marker")
+        print("  Model saved with quantized=True marker")
 
         # Verify file was created
         if not model_path.exists():
-            print(f"  [ERROR] Model file not found!")
+            print("  [ERROR] Model file not found!")
             return False
 
         # Read the file to verify quantized marker (JSON format uses "quantized": true)
-        print(f"\n[STEP 3] Verifying saved file contains quantized marker...")
+        print("\n[STEP 3] Verifying saved file contains quantized marker...")
         with open(model_path, 'r') as f:
             content = f.read()
             if '"quantized": true' in content:
-                print(f"  [OK] JSON quantized field found: \"quantized\": true")
+                print("  [OK] JSON quantized field found: \"quantized\": true")
             else:
-                print(f"  [ERROR] Quantized marker NOT found in saved file")
+                print("  [ERROR] Quantized marker NOT found in saved file")
                 print(f"  First 500 chars: {content[:500]}")
                 return False
 
         # Step 4: Run inference to verify it works
-        print(f"\n[STEP 4] Running inference...")
+        print("\n[STEP 4] Running inference...")
         test_input = np.array([[1.0, 2.0, 3.0, 4.0],
                                 [5.0, 6.0, 7.0, 8.0],
                                 [9.0, 10.0, 11.0, 12.0],
@@ -104,11 +104,11 @@ def test_quantization_roundtrip(quant_dtype="int8", quant_level_name="int8"):
         # So y should be close to x but with quantization error
 
         if not np.allclose(output_tensor, test_input, rtol=0.1):
-            print(f"  [WARNING] Output differs from input more than expected")
+            print("  [WARNING] Output differs from input more than expected")
         else:
-            print(f"  [OK] Output matches input (within quantization error)")
+            print("  [OK] Output matches input (within quantization error)")
 
-    print(f"\n[STEP 5] COMPLETE")
+    print("\n[STEP 5] COMPLETE")
     print("=" * 80)
     print()
     return True

examples/text_generation_enhanced.py

@@ -19,7 +19,6 @@
     python text_generation_enhanced.py --weights model.json --use-kv-cache --tokens 100
 """
 
-import sys
 import argparse
 import json
 from pathlib import Path
@@ -187,8 +186,8 @@ def build_graph_with_kv(self, builder, x_input, cached_k, cached_v):
         ln_beta = builder.constant(self.ln_beta, list(self.ln_beta.shape), "float32")
 
         # Compute Q, K, V for new token
-        Q = builder.matmul(x_input, W_q)  # (1, d_model)
-        K = builder.matmul(x_input, W_k)  # (1, d_model)
+        builder.matmul(x_input, W_q)  # (1, d_model)
+        builder.matmul(x_input, W_k)  # (1, d_model)
         V = builder.matmul(x_input, W_v)  # (1, d_model)
 
         # Simplified attention: use mean pooling

examples/text_generation_gpt.py

@@ -14,7 +14,6 @@
 using pre-trained or initialized weights.
 """
 
-import sys
 import argparse
 import json
 from pathlib import Path
@@ -133,8 +132,8 @@ def build_graph(self, builder, x_input):
 
         # Simplified attention: use last token's representation
         # Q, K, V = x @ W_q, x @ W_k, x @ W_v
-        Q = builder.matmul(x_input, W_q)  # (seq_len, d_model)
-        K = builder.matmul(x_input, W_k)  # (seq_len, d_model)
+        builder.matmul(x_input, W_q)  # (seq_len, d_model)
+        builder.matmul(x_input, W_k)  # (seq_len, d_model)
         V = builder.matmul(x_input, W_v)  # (seq_len, d_model)
 
         # Compute attention scores for last position
@@ -252,7 +251,7 @@ def main():
     print("Next-Token Generation with Attention (WebNN)")
     print("=" * 70)
     print(f"Backend: {backend_name}")
-    print(f"Model: vocab=128 (ASCII), d_model=64, max_seq=32")
+    print("Model: vocab=128 (ASCII), d_model=64, max_seq=32")
     print()
 
     # Create WebNN context

examples/train_simple_demo.py

@@ -18,7 +18,6 @@
 import argparse
 import json
 import numpy as np
-from pathlib import Path
 
 
 # Character-level vocabulary

examples/train_text_model.py

@@ -72,8 +72,8 @@ def compute_loss(model, X_batch, y_batch):
 
         # Forward pass (simplified in NumPy)
         # Q, K, V
-        Q = x_embedded @ model.W_q
-        K = x_embedded @ model.W_k
+        x_embedded @ model.W_q
+        x_embedded @ model.W_k
         V = x_embedded @ model.W_v
 
         # Mean pooling approximation
@@ -114,7 +114,7 @@ def train_step(model, X_batch, y_batch, lr=0.05):
     Perform one training step with simple gradient descent.
     Improved to focus on most important parameters.
     """
-    batch_size = X_batch.shape[0]
+    X_batch.shape[0]
     epsilon = 0.0001
 
     # Get original loss

python/webnn/__init__.pyi

@@ -1,6 +1,6 @@
 """Type stubs for webnn package"""
 
-from typing import Any, Dict, List, Optional, Union
+from typing import Any, Dict, List, Optional
 import numpy as np
 import numpy.typing as npt
 

python/webnn/hub.py

@@ -4,7 +4,6 @@
 Downloads WebNN model files (graph, weights, manifest) from Hugging Face Hub.
 """
 
-import os
 from pathlib import Path
 from urllib.request import urlretrieve
 from urllib.error import HTTPError

tests/test_coreml_basic.py

@@ -20,9 +20,7 @@
 import argparse
 import os
 import sys
-from pathlib import Path
 
-import numpy as np
 
 try:
     import webnn

tests/test_integration.py

@@ -17,7 +17,6 @@
 import argparse
 import os
 import sys
-from pathlib import Path
 
 import numpy as np
 

tests/test_performance.py

@@ -21,7 +21,6 @@
 from typing import Dict, List, Tuple
 
 # Import conftest fixtures
-from conftest import ml, context, builder
 
 
 def measure_inference_time(context, graph, inputs: Dict, num_runs: int = 5) -> Tuple[float, List[float]]:
@@ -41,7 +40,7 @@ def measure_inference_time(context, graph, inputs: Dict, num_runs: int = 5) -> T
 
     for i in range(num_runs):
         start = time.time()
-        result = context.compute(graph, inputs)
+        context.compute(graph, inputs)
         elapsed = (time.time() - start) * 1000
         times.append(elapsed)
 
@@ -107,7 +106,7 @@ def test_performance_simple_onnx_cpu(ml):
     warm_avg = sum(warm_times) / len(warm_times)
 
     print(f"\n{'='*60}")
-    print(f"ONNX CPU - Simple Model (10 layers)")
+    print("ONNX CPU - Simple Model (10 layers)")
     print(f"{'='*60}")
     print(f"Cold start: {cold_time:.2f}ms")
     print(f"Warm avg:   {warm_avg:.2f}ms")
@@ -135,7 +134,7 @@ def test_performance_simple_coreml(ml):
     warm_avg = sum(warm_times) / len(warm_times)
 
     print(f"\n{'='*60}")
-    print(f"CoreML - Simple Model (10 layers)")
+    print("CoreML - Simple Model (10 layers)")
     print(f"{'='*60}")
     print(f"Cold start: {cold_time:.2f}ms")
     print(f"Warm avg:   {warm_avg:.2f}ms")
@@ -160,7 +159,7 @@ def test_performance_complex_onnx_cpu(ml):
     warm_avg = sum(warm_times) / len(warm_times)
 
     print(f"\n{'='*60}")
-    print(f"ONNX CPU - Complex Model (200 operations)")
+    print("ONNX CPU - Complex Model (200 operations)")
     print(f"{'='*60}")
     print(f"Cold start: {cold_time:.2f}ms")
     print(f"Warm avg:   {warm_avg:.2f}ms")
@@ -190,7 +189,7 @@ def test_performance_complex_coreml(ml):
     speedup = cold_time / warm_avg
 
     print(f"\n{'='*60}")
-    print(f"CoreML - Complex Model (200 operations)")
+    print("CoreML - Complex Model (200 operations)")
     print(f"{'='*60}")
     print(f"Cold start: {cold_time:.2f}ms")
     print(f"Run 2:      {warm_times[0]:.2f}ms")