Skip to content

Commit 7b1c5fa

Browse files
aimen-djariaimen-djari
authored
Feat: ai frameworks (#53)
1 parent b967903 commit 7b1c5fa

File tree

2 files changed

+183
-0
lines changed

2 files changed

+183
-0
lines changed

‎.vitepress/sidebar.ts‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -579,6 +579,10 @@ export function getSidebar() {
579579
},
580580
],
581581
},
582+
{
583+
text: '🤖 AI Framework Support',
584+
link: '/protocol/ai',
585+
},
582586
],
583587
},
584588
],

‎src/protocol/ai.md‎

Lines changed: 179 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,179 @@
1+
---
2+
title: AI Framework Support
3+
description:
4+
Comprehensive guide to AI framework compatibility and support on the iExec
5+
platform
6+
---
7+
8+
# 🤖 AI Framework Support
9+
10+
The iExec Platform provides comprehensive support for popular AI and machine
11+
learning frameworks. Deploy confidential AI with ease. iExec supports popular
12+
AI/ML frameworks, running in secure Trusted Execution Environments (TEEs) with
13+
optimized configurations.
14+
15+
## 🚀 Quick Start
16+
17+
**Want to get started immediately?**
18+
19+
- 📚
20+
**[AI Frameworks Hello World](https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world)** -
21+
Ready-to-use Docker examples for TensorFlow, PyTorch, and more
22+
- 🛠️ **[Build & Deploy](/guides/build-iapp/build-&-deploy)** - General iApp
23+
development guide (not AI-specific)
24+
- 🔬
25+
**[TDX App Guide](/guides/build-iapp/advanced/create-your-first-tdx-app)** -
26+
Build TDX applications (works well for AI workloads)
27+
28+
## 🛡️ Why iExec for AI?
29+
30+
### Confidential Computing
31+
32+
Trusted Execution Environments (TEEs) protect your AI models and data
33+
end-to-end:
34+
35+
- **Data Privacy**: TEEs isolate AI computations in secure enclaves
36+
- **Secure Training & Inference**: Unauthorized entities can never access your
37+
models and data
38+
- **Hardware-Level Security**: Intel SGX and TDX provide enterprise-grade
39+
protection
40+
41+
### AI Monetization
42+
43+
Monetize your AI assets easily and securely:
44+
45+
- **Datasets**: Encrypt and sell access to your training data
46+
- **Models**: Deploy and monetize your trained AI models
47+
- **Agents**: Create and sell AI agents and applications
48+
- **Ownership Preserved**: Your digital assets always remain yours
49+
50+
### Decentralized Infrastructure
51+
52+
Scale AI applications without centralized cloud dependencies:
53+
54+
- **On-Demand Compute**: Access powerful resources when you need them
55+
- **Fair Pricing**: Blockchain verifies execution costs transparently
56+
- **Global Network**: Deploy across a worldwide network of secure workers
57+
58+
## 🤖 AI Framework Support
59+
60+
### Overview
61+
62+
| Framework | TDX Support | SGX Support | Best For |
63+
| ---------------- | --------------- | --------------- | ----------------------------- |
64+
| **TensorFlow** | ✅ Yes (3.01GB) | ❌ No | Deep learning, production ML |
65+
| **PyTorch** | ✅ Yes (6.44GB) | ❌ No | Research, computer vision |
66+
| **Scikit-learn** | ✅ Yes (1.18GB) | ✅ Yes (1.01GB) | Traditional ML, data analysis |
67+
| **OpenVINO** | ✅ Yes (1.82GB) | ❌ No | Computer vision, inference |
68+
| **NumPy** | ✅ Yes (1.25GB) | ✅ Yes (1.08GB) | Scientific computing |
69+
| **Matplotlib** | ✅ Yes (1.25GB) | ✅ Yes (1.08GB) | Data visualization |
70+
71+
### Framework Details
72+
73+
| Framework | Version | Description | TDX Support | SGX Support | Use Cases | Resources |
74+
| ---------------- | ----------- | --------------------------------------------- | ----------- | ------------------- | -------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
75+
| **TensorFlow** | 2.19.0 | Google's ML framework for production AI | ✅ 3.01GB | ❌ Too large | Deep learning, CV, NLP | [Docs](https://www.tensorflow.org/) • [Quickstart](https://www.tensorflow.org/tutorials/quickstart/beginner) • [Docker](https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world/tree/main/tensorflow) |
76+
| **PyTorch** | 2.7.0+cu126 | Facebook's research-focused DL framework | ✅ 6.44GB | ❌ Too large | Research, DL, CV, NLP | [Docs](https://pytorch.org/docs/) • [Quickstart](https://docs.pytorch.org/tutorials/beginner/basics/quickstart_tutorial.html) • [Docker](https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world/tree/main/pytorch) |
77+
| **Scikit-learn** | 1.6.1 | Comprehensive ML library for Python | ✅ 1.18GB | ✅ 1.01GB | Classification, regression, clustering | [Docs](https://scikit-learn.org/stable/) • [Examples](https://scikit-learn.org/stable/auto_examples/index.html) • [Docker](https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world/tree/main/scikit) |
78+
| **OpenVINO** | 2024.6.0 | Intel's high-performance AI inference toolkit | ✅ 1.82GB | ❌ Execution issues | Computer vision, inference | [Docs](https://docs.openvino.ai/) • [Tutorial](https://docs.openvino.ai/2023.3/notebooks/004-hello-detection-with-output.html) • [Docker](https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world/tree/main/openvino) |
79+
| **NumPy** | 2.0.2 | Fundamental package for scientific computing | ✅ 1.25GB | ✅ 1.08GB | Scientific computing, data analysis | [Docs](https://numpy.org/doc/) • [User Guide](https://numpy.org/doc/stable/user/index.html) • [Docker](https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world/tree/main/numpy) |
80+
| **Matplotlib** | 3.9.4 | Comprehensive library for data visualization | ✅ 1.25GB | ✅ 1.08GB | Data visualization, plotting | [Docs](https://matplotlib.org/) • [Gallery](https://matplotlib.org/stable/gallery/index.html) • [Docker](https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world/tree/main/matplotlib) |
81+
82+
## 🐳 Getting Started with Docker Examples
83+
84+
### What's Included
85+
86+
Our
87+
[AI Frameworks Hello World repository](https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world)
88+
includes ready-to-use examples:
89+
90+
```
91+
ai-frameworks-hello-world/
92+
├── tensorflow/ # TensorFlow 2.19.0 example
93+
├── pytorch/ # PyTorch 2.7.0+cu126 example
94+
├── scikit/ # Scikit-learn 1.6.1 example
95+
├── openvino/ # OpenVINO 2024.6.0 example
96+
├── numpy/ # NumPy 2.0.2 example
97+
└── matplotlib/ # Matplotlib 3.9.4 example
98+
```
99+
100+
### Quick Start Commands
101+
102+
```bash
103+
# Clone the repository
104+
git clone https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world.git
105+
cd ai-frameworks-hello-world
106+
107+
# Try TensorFlow example
108+
cd tensorflow
109+
docker build -t hello-tensorflow .
110+
docker run --rm hello-tensorflow
111+
112+
# Try PyTorch example
113+
cd ../pytorch
114+
docker build -t hello-pytorch .
115+
docker run --rm hello-pytorch
116+
```
117+
118+
### Features
119+
120+
- **✅ Isolated Testing**: Each framework runs in its own container
121+
- **✅ Reproducible**: Consistent environment across systems
122+
- **✅ TDX Ready**: All containers tested for Intel TDX compatibility
123+
- **✅ Easy Deployment**: Simple build and run commands
124+
125+
## 📊 Technology Comparison
126+
127+
### TDX vs SGX for AI
128+
129+
| Feature | Intel TDX | Intel SGX |
130+
| --------------------- | -------------------------- | ---------------------------------- |
131+
| **Memory Limit** | Multi-GB+ | ~1.95GB |
132+
| **Framework Support** | All major frameworks | Limited (Scikit-learn, NumPy) |
133+
| **Code Changes** | Minimal ("lift and shift") | Significant modifications required |
134+
| **Production Ready** | ✅ Yes | ⚠️ Limited |
135+
| **AI Workloads** | ✅ Excellent | ❌ Restricted |
136+
137+
### Recommendations
138+
139+
#### For Production AI Applications
140+
141+
- **Use TDX** for TensorFlow, PyTorch, and OpenVINO
142+
- **Use SGX** for lightweight ML with Scikit-learn and NumPy
143+
144+
#### For Development and Testing
145+
146+
- **Start with SGX** for simple ML tasks
147+
- **Migrate to TDX** for complex AI workloads
148+
149+
#### Important Considerations
150+
151+
- **SGX Limitations**: Expect potential library incompatibilities and code
152+
modifications
153+
- **TDX Advantages**: Minimal code changes required ("lift and shift" approach)
154+
155+
## 📚 Next Steps
156+
157+
### Learn TEE Technologies
158+
159+
- **[Intel SGX Technology](/protocol/tee/intel-sgx)** - SGX limitations and
160+
capabilities
161+
- **[Intel TDX Technology](/protocol/tee/intel-tdx)** - TDX advantages for AI
162+
- **[SGX vs TDX Comparison](/protocol/tee/sgx-vs-tdx)** - Detailed comparison
163+
164+
### Build AI Applications
165+
166+
- **[Build & Test](/guides/build-iapp/build-&-test)** - Build and test your AI
167+
application
168+
- **[Deploy & Run](/guides/build-iapp/deploy-&-run)** - Deploy and run your AI
169+
application
170+
- **[Build Intel TDX App](/guides/build-iapp/advanced/build-your-first-tdx-iapp)** -
171+
TDX applications for AI workloads
172+
- **[Inputs and Outputs](/guides/build-iapp/inputs-and-outputs)** - Handle data
173+
flow in TEE environment
174+
175+
### Explore Examples
176+
177+
- **[AI Frameworks Hello World](https://github.com/iExecBlockchainComputing/ai-frameworks-hello-world)** -
178+
Ready-to-use Docker examples
179+
- **[iExec Discord](https://discord.com/invite/pbt9m98wnU)** - Community support

0 commit comments

Comments
 (0)