Deploy from GitHub - 2026-01-19 04:19:46

.gitattributes

@@ -1,35 +0,0 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -1,39 +0,0 @@
-__pycache__/
-*.pyc
-*.pyo
-*.pyd
-.Python
-*.so
-*.egg
-*.egg-info/
-dist/
-build/
-
-# Model weights (downloaded at runtime via GitHub releases)
-models/*.pth
-models/*.pt
-
-# Test outputs
-test_outputs/
-*.jpg
-*.png
-!examples/*.jpg
-!examples/*.png
-
-# IDE
-.vscode/
-.idea/
-*.swp
-*.swo
-
-# OS
-.DS_Store
-Thumbs.db
-
-# Gradio
-gradio_cached_examples/
-flagged/
-
-# BUT allow pre-compiled kernels in prebuilt/
-!kernels/prebuilt/*.so
-!kernels/prebuilt/*.pyd

compile_kernels.py

@@ -73,11 +73,12 @@ try:
     print("Compiling... (this may take 1-2 minutes)")
 
     # Compile the kernel
+    # Note: PyTorch 2.x requires cpp_sources even if empty (bindings are in CUDA)
     module = load_inline(
         name='fused_instance_norm',
+        cpp_sources=[],  # Empty since bindings are in the .cu file
         cuda_sources=[cuda_source],
         extra_cuda_cflags=extra_cuda_cflags,
-        with_pybind11=True,
         verbose=False
     )
 

kernels/__init__.py

@@ -2,7 +2,7 @@
 StyleForge CUDA Kernels Package
 Custom CUDA kernels for accelerated neural style transfer.
 
-For ZeroGPU: Pre-compiled kernels are loaded from prebuilt/.
+For ZeroGPU/HuggingFace: Pre-compiled kernels are downloaded from HF dataset.
 For local: Kernels are JIT-compiled if prebuilt not available.
 """
 
@@ -15,11 +15,59 @@ _CUDA_KERNELS_AVAILABLE = False
 _FusedInstanceNorm2d = None
 _KERNELS_COMPILED = False
 
-# Check if running on ZeroGPU
-_ZERO_GPU = os.environ.get('SPACE_ID', '').startswith('hf.co') or os.environ.get('ZERO_GPU') == '1'
+# Check if running on ZeroGPU/HuggingFace
+_ZERO_GPU = os.environ.get('SPACE_ID', '') or os.environ.get('ZERO_GPU') == '1'
 
 # Path to pre-compiled kernels
 _PREBUILT_PATH = Path(__file__).parent / "prebuilt"
+_PREBUILT_PATH.mkdir(exist_ok=True)
+
+# HuggingFace dataset for prebuilt kernels
+_KERNEL_DATASET = "oliau/styleforge-kernels"  # You'll need to create this dataset
+
+
+def _download_kernels_from_dataset():
+    """Download pre-compiled kernels from HuggingFace dataset."""
+    try:
+        from huggingface_hub import hf_hub_download, HfFileSystem
+        fs = HfFileSystem()
+
+        # List all .so and .pyd files in the dataset
+        kernel_files = []
+        try:
+            files = fs.ls(f"datasets/{_KERNEL_DATASET}")
+            for f in files:
+                if f['name'].endswith(('.so', '.pyd')):
+                    kernel_files.append(Path(f['name']).name)
+        except Exception:
+            # Dataset might not exist yet
+            return False
+
+        if not kernel_files:
+            return False
+
+        # Download each kernel file
+        for kernel_file in kernel_files:
+            try:
+                local_path = hf_hub_download(
+                    repo_id=_KERNEL_DATASET,
+                    filename=kernel_file,
+                    repo_type="dataset",
+                    local_dir=str(_PREBUILT_PATH.parent),
+                    local_dir_use_symlinks=False
+                )
+                print(f"Downloaded kernel: {kernel_file}")
+            except Exception as e:
+                print(f"Failed to download {kernel_file}: {e}")
+                continue
+
+        return True
+    except ImportError:
+        print("huggingface_hub not available, skipping kernel download")
+        return False
+    except Exception as e:
+        print(f"Failed to download kernels from dataset: {e}")
+        return False
 
 
 def check_cuda_kernels():
@@ -46,6 +94,7 @@ def get_fused_instance_norm(num_features, **kwargs):
 def load_prebuilt_kernels():
     """
     Try to load pre-compiled CUDA kernels from prebuilt/ directory.
+    On HuggingFace, downloads from dataset if local files not found.
 
     Returns True if successful, False otherwise.
     """
@@ -54,10 +103,18 @@ def load_prebuilt_kernels():
     if _KERNELS_COMPILED:
         return _CUDA_KERNELS_AVAILABLE
 
-    # Check if prebuilt kernels exist
+    # Check if prebuilt kernels exist locally
     prebuilt_files = list(_PREBUILT_PATH.glob("*.so")) + list(_PREBUILT_PATH.glob("*.pyd"))
+
+    # On HuggingFace Spaces, try downloading from dataset if not found locally
+    if not prebuilt_files and _ZERO_GPU:
+        print("No local pre-compiled kernels found. Trying HuggingFace dataset...")
+        if _download_kernels_from_dataset():
+            # Check again after download
+            prebuilt_files = list(_PREBUILT_PATH.glob("*.so")) + list(_PREBUILT_PATH.glob("*.pyd"))
+
     if not prebuilt_files:
-        print("No pre-compiled kernels found in prebuilt/")
+        print("No pre-compiled kernels found")
         return False
 
     try:

kernels/prebuilt/README.md

@@ -1,64 +0,0 @@
-# Pre-Compiled CUDA Kernels
-
-This directory contains pre-compiled CUDA kernels for use on Hugging Face Spaces.
-
-## How to Compile Kernels Locally
-
-To compile the CUDA kernels locally and upload them here:
-
-### 1. Compile Locally
-
-Run this script from the `huggingface-space` directory:
-
-```bash
-python compile_kernels.py
-```
-
-Or compile manually:
-
-```bash
-cd huggingface-space
-python -c "
-from kernels.cuda_build import compile_inline
-from pathlib import Path
-
-cuda_source = (Path('kernels') / 'instance_norm.cu').read_text()
-module = compile_inline(
-    name='fused_instance_norm',
-    cuda_source=cuda_source,
-    functions=['forward'],
-    build_directory=Path('build'),
-    verbose=True
-)
-print('Compiled successfully!')
-print(f'Module location: {module.__file__}')
-"
-```
-
-### 2. Copy Compiled File
-
-After compilation, copy the compiled `.so` file to this directory:
-
-```bash
-# Find the compiled file (usually in build/)
-find build/ -name "*.so" -exec cp {} kernels/prebuilt/ \;
-```
-
-### 3. Commit and Push
-
-```bash
-git add kernels/prebuilt/
-git commit -m "Add pre-compiled CUDA kernels"
-git push
-```
-
-## Notes
-
-- The compiled kernels are architecture-specific (e.g., `sm_70`, `sm_75`, `sm_86`)
-- Hugging Face Spaces typically use Tesla T4 (sm_75) or A100 (sm_80)
-- For maximum compatibility, compile with multiple compute capabilities
-
-## Current Status
-
-No pre-compiled kernels found. The app will use PyTorch's InstanceNorm2d fallback,
-which is still GPU-accelerated but not as fast as custom fused kernels.

upload_kernels_to_dataset.py

@@ -0,0 +1,159 @@
+#!/usr/bin/env python3
+"""
+Upload pre-compiled CUDA kernels to Hugging Face Dataset.
+
+This avoids git push issues with binary files on Hugging Face Spaces.
+
+Usage:
+    python upload_kernels_to_dataset.py
+
+The kernels will be uploaded to: huggingface.co/datasets/oliau/styleforge-kernels
+"""
+
+import sys
+from pathlib import Path
+
+try:
+    from huggingface_hub import HfApi, login
+except ImportError:
+    print("ERROR: huggingface_hub not installed.")
+    print("Install with: pip install huggingface_hub")
+    sys.exit(1)
+
+
+def upload_kernels():
+    """Upload pre-compiled kernels to Hugging Face dataset."""
+
+    # Configuration
+    DATASET_ID = "oliau/styleforge-kernels"
+    PREBUILT_DIR = Path("kernels/prebuilt")
+
+    print("=" * 60)
+    print("StyleForge Kernel Uploader")
+    print("=" * 60)
+    print()
+
+    # Check if prebuilt directory exists
+    if not PREBUILT_DIR.exists():
+        print(f"ERROR: Prebuilt directory not found: {PREBUILT_DIR}")
+        print("Run compile_kernels.py first to generate the kernels.")
+        sys.exit(1)
+
+    # Find all kernel files
+    kernel_files = list(PREBUILT_DIR.glob("*.so")) + list(PREBUILT_DIR.glob("*.pyd"))
+
+    if not kernel_files:
+        print(f"ERROR: No kernel files found in {PREBUILT_DIR}")
+        print("Expected .so or .pyd files.")
+        sys.exit(1)
+
+    print(f"Found {len(kernel_files)} kernel file(s):")
+    for f in kernel_files:
+        size_kb = f.stat().st_size / 1024
+        print(f"  - {f.name} ({size_kb:.1f} KB)")
+    print()
+
+    # Initialize HF API
+    api = HfApi()
+
+    # Check if user is logged in
+    try:
+        whoami = api.whoami()
+        print(f"Logged in as: {whoami.get('name', whoami.get('user', 'unknown'))}")
+    except Exception:
+        print("Not logged in to Hugging Face.")
+        print("Please run: huggingface-cli login")
+        print("Or set HF_TOKEN environment variable.")
+        sys.exit(1)
+
+    print()
+    print(f"Uploading to dataset: {DATASET_ID}")
+    print()
+
+    # Create dataset if it doesn't exist
+    try:
+        repo_info = api.repo_info(DATASET_ID, repo_type="dataset")
+        print(f"Dataset exists: {DATASET_ID}")
+    except Exception:
+        print(f"Creating new dataset: {DATASET_ID}")
+        api.create_repo(
+            repo_id=DATASET_ID.split("/")[1],
+            repo_type="dataset",
+            public=True,
+            exist_ok=True
+        )
+        print(f"Dataset created: {DATASET_ID}")
+
+    # Create README for the dataset
+    readme_content = """---
+title: StyleForge CUDA Kernels
+license: mit
+tags:
+  - cuda
+  - neural-style-transfer
+  - styleforge
+---
+
+# StyleForge Pre-compiled CUDA Kernels
+
+This repository contains pre-compiled CUDA kernels for the StyleForge neural style transfer project.
+
+## Files
+
+"""
+    for f in kernel_files:
+        readme_content += f"- `{f.name}`\n"
+
+    readme_content += """
+## Usage
+
+These kernels are automatically downloaded by StyleForge when running on Hugging Face Spaces.
+
+## Compilation
+
+Kernels are compiled for multiple GPU architectures:
+- sm_70 (V100)
+- sm_75 (T4)
+- sm_80 (A100)
+
+For local compilation, see `compile_kernels.py` in the main repository.
+"""
+
+    # Upload files
+    print("Uploading files...")
+
+    # Upload README
+    api.upload_file(
+        path_or_fileobj=readme_content.encode(),
+        path_in_repo="README.md",
+        repo_id=DATASET_ID,
+        repo_type="dataset",
+        commit_message="Add dataset README"
+    )
+    print("  Uploaded: README.md")
+
+    # Upload kernel files
+    for kernel_file in kernel_files:
+        print(f"  Uploading {kernel_file.name}...", end=" ", flush=True)
+        api.upload_file(
+            path_or_fileobj=str(kernel_file),
+            path_in_repo=kernel_file.name,
+            repo_id=DATASET_ID,
+            repo_type="dataset",
+            commit_message=f"Add {kernel_file.name}"
+        )
+        print("✓")
+
+    print()
+    print("=" * 60)
+    print("Upload complete!")
+    print("=" * 60)
+    print()
+    print(f"Dataset URL: https://huggingface.co/datasets/{DATASET_ID}")
+    print()
+    print("The kernels will be automatically downloaded by StyleForge")
+    print("when running on Hugging Face Spaces.")
+
+
+if __name__ == "__main__":
+    upload_kernels()