Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,85 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
---
|
4 |
+
# LTX-Q8-Kernels Pre-built Wheel
|
5 |
+
|
6 |
+
This repository contains pre-built wheel files for the LTX-Q8-Kernels to avoid the need for building on SM89 instances with Python 3.11.
|
7 |
+
|
8 |
+
## Quick Install
|
9 |
+
|
10 |
+
Instead of building the kernels from source (which requires SM89 compute capability), you can directly install the pre-built wheel:
|
11 |
+
|
12 |
+
```bash
|
13 |
+
pip install https://huggingface.co/ModelsLab/LTX-Q8-Kernels-build-ada/resolve/main/q8_kernels-0.0.5-cp311-cp311-linux_x86_64.whl
|
14 |
+
```
|
15 |
+
|
16 |
+
## Requirements
|
17 |
+
|
18 |
+
- **Python**: 3.11
|
19 |
+
- **Platform**: Linux x86_64
|
20 |
+
- **CUDA**: Compatible CUDA installation
|
21 |
+
- **Architecture**: Built for Ada Lovelace (RTX 40 series) and compatible GPUs
|
22 |
+
|
23 |
+
## Usage
|
24 |
+
|
25 |
+
After installation, you can import and use the kernels in your LTX-Video projects:
|
26 |
+
|
27 |
+
```python
|
28 |
+
import q8_kernels
|
29 |
+
# Your LTX-Video code here
|
30 |
+
```
|
31 |
+
|
32 |
+
## Benefits
|
33 |
+
|
34 |
+
- **No SM89 Build Requirements**: Skip the complex build process that requires specific GPU compute capabilities
|
35 |
+
- **Faster Setup**: Direct installation without compilation time
|
36 |
+
- **Pre-optimized**: Built with optimal settings for Ada Lovelace architecture
|
37 |
+
|
38 |
+
## Compatibility
|
39 |
+
|
40 |
+
This wheel is specifically built for:
|
41 |
+
- **Python 3.11** (`cp311-cp311`)
|
42 |
+
- **Linux x86_64** systems
|
43 |
+
- **Ada Lovelace GPU architecture** (RTX 4090, RTX 4080, etc.)
|
44 |
+
|
45 |
+
## Alternative Installation Methods
|
46 |
+
|
47 |
+
### From Hugging Face Hub
|
48 |
+
|
49 |
+
```bash
|
50 |
+
# Using huggingface_hub
|
51 |
+
pip install huggingface_hub
|
52 |
+
python -c "from huggingface_hub import hf_hub_download; hf_hub_download(repo_id='ModelsLab/LTX-Q8-Kernels-build-ada', filename='q8_kernels-0.0.5-cp311-cp311-linux_x86_64.whl', local_dir='.')"
|
53 |
+
pip install q8_kernels-0.0.5-cp311-cp311-linux_x86_64.whl
|
54 |
+
```
|
55 |
+
|
56 |
+
### Download and Install
|
57 |
+
|
58 |
+
```bash
|
59 |
+
wget https://huggingface.co/ModelsLab/LTX-Q8-Kernels-build-ada/resolve/main/q8_kernels-0.0.5-cp311-cp311-linux_x86_64.whl
|
60 |
+
pip install q8_kernels-0.0.5-cp311-cp311-linux_x86_64.whl
|
61 |
+
```
|
62 |
+
|
63 |
+
## Troubleshooting
|
64 |
+
|
65 |
+
If you encounter issues:
|
66 |
+
|
67 |
+
1. **Verify Python version**: `python --version` should show 3.11.x
|
68 |
+
2. **Check platform**: This wheel only works on Linux x86_64
|
69 |
+
3. **CUDA compatibility**: Ensure you have a compatible CUDA installation
|
70 |
+
4. **GPU support**: Verify your GPU supports the required compute capability
|
71 |
+
|
72 |
+
## Building from Source (Alternative)
|
73 |
+
|
74 |
+
If you need to build from source or for different architectures, refer to the main [LTX-Video repository](https://github.com/Lightricks/LTX-Video) for build instructions.
|
75 |
+
|
76 |
+
## Version Information
|
77 |
+
|
78 |
+
- **Package Version**: 0.0.5
|
79 |
+
- **Python**: 3.11
|
80 |
+
- **Platform**: linux_x86_64
|
81 |
+
- **Build Target**: Ada Lovelace (SM89)
|
82 |
+
|
83 |
+
## License
|
84 |
+
|
85 |
+
This package follows the same license as the original LTX-Video project.
|