Interactive learning and development tools for Tenstorrent AI accelerators
Learn Tenstorrent hardware and software through 48 interactive lessons with guided hands-on exercises, production-ready code templates, and intelligent hardware detection. Perfect for developers new to Tenstorrent hardware and teams building production AI inference and custom training pipelines.
Overview
The Tenstorrent VSCode Toolkit is an educational extension that provides:
- ✅ 48 Interactive Lessons - From hardware detection to custom training, organized by skill level
- ✅ Custom Training Ready - Train models from scratch or fine-tune existing ones (validated on hardware!)
- ✅ Click-to-Run Commands - Execute lessons step-by-step without copy-pasting commands
- ✅ Hardware Auto-Detection - Automatically detects your Tenstorrent device and provides tailored guidance
- ✅ Production Templates - Real, tested code you can customize for your projects
- ✅ Multi-Framework Support - Learn vLLM, TT-Forge, TT-XLA, and TT-Metal
- ✅ Live Device Monitoring - Real-time temperature, power, and health status in the status bar
Target Audience:
- Developers new to Tenstorrent hardware
- AI engineers deploying models on TT accelerators
- Teams building production inference pipelines
- ML researchers training custom models
- Contributors to the Tenstorrent ecosystem
Quick Start
Try in Docker (No Installation)
Run the IDE locally in your browser:
docker run -d -p 8080:8080 -e PASSWORD=demo \
ghcr.io/tenstorrent/tt-vscode-toolkit:latest
Access at: http://localhost:8080 (password: demo)
Deploy to Cloud with Real Hardware
Get access to Tenstorrent N300 hardware in 60 seconds:
# Install Koyeb CLI
curl -fsSL https://cli.koyeb.com/install.sh | sh
# Login
koyeb login
# Deploy
koyeb services create vscode \
--app tt-vscode-toolkit \
--docker ghcr.io/tenstorrent/tt-vscode-toolkit:latest \
--ports 8080:http \
--routes /:8080 \
--env PASSWORD=yourpass \
--env MESH_DEVICE=N300 \
--regions na \
--instance-type gpu-tenstorrent-n300s \
--privileged
See Deploy to Koyeb lesson for details.
Installation
Prerequisites
Hardware:
- Tenstorrent accelerator (N150, N300, T3K, P100, P150, or Galaxy)
- 32GB+ RAM recommended (16GB minimum)
- 100GB+ free disk space for models
Software:
- Linux (Ubuntu 20.04+, RHEL 8+, or compatible)
- Python 3.10+ (3.11 for TT-XLA)
- VSCode 1.93+
- tt-metal installed and configured
Verify your environment:
tt-smi # Hardware detected?
python3 --version # Python 3.10+?
python3 -c "import ttnn; print('✓ Ready')" # tt-metal working?
Installation
Option 1: Install from VSIX Package (Recommended)
# Download the latest release
# Install extension
code --install-extension tt-vscode-toolkit-*.vsix
# Restart VSCode
Option 2: Build from Source
# Clone repository
git clone https://github.com/tenstorrent/tt-vscode-toolkit.git
cd tt-vscode-toolkit
# Install dependencies
npm install
# Build extension
npm run build
# Package extension
npm run package
# Install
code --install-extension tt-vscode-toolkit-*.vsix
First Steps
- Open VSCode - The extension activates automatically on startup
- Open Tenstorrent Sidebar - Click the Tenstorrent icon in the activity bar
- Start Learning - Begin with "Hardware Detection" lesson or open the Welcome page
Configuration: By default, all lessons are visible. To show only validated lessons, disable "Show Unvalidated Lessons" in settings.
Learning Paths
🎯 Beginner Path (4-6 hours)
Perfect for first-time users
1. Hardware Detection (5 min) → Verify your hardware setup
2. Verify Installation (5 min) → Test tt-metal installation
3. Download Model (30 min) → Get Llama-3.1-8B or Qwen3-0.6B
4. vLLM Production (20 min) → Launch production server
What you'll learn: Hardware detection, environment verification, model downloading, and production inference serving with OpenAI-compatible API.
For experienced developers
1. Hardware Detection → Verify setup
2. vLLM Production → Production serving
3. Image Generation → Stable Diffusion on TT hardware
4. TT-Forge → PyTorch model compilation
5. Coding Assistant → Build an AI coding tool
What you'll learn: Production deployment patterns, multi-modal inference (text + images), compiler workflows, and practical AI applications.
💡 Advanced Path (10-15 hours)
For contributors and power users
1. TT-XLA → JAX production compiler
2. RISC-V Programming → Low-level Tensix core programming
3. Bounty Program → Model bring-up opportunities
4. TT-Metalium Cookbook → Custom hardware projects
What you'll learn: Advanced compiler usage, low-level hardware programming, model bring-up workflows, and custom kernel development.
Lesson Catalog
39 lessons organized by category. Hardware badges show validated platforms (✅ = tested and working).
👋 Your journey begins here
0 lessons, 0 validated
🚀 Your First Inference
7 lessons, 7 validated
- Modern Setup with tt-installer 2.0 —
N150 P300C
- Hardware Detection —
N150 P300C
- Download Model and Run Inference —
N150 P300C
- Verify Your Setup —
N150 P300C
- Interactive Chat with Direct API —
N150
- HTTP API Server with Direct API —
N150
- Build tt-metal from Source —
N150 P300C
🏭 Serving Models
4 lessons, 3 validated
- Production Inference with tt-inference-server —
N150 P100
- Production Inference with vLLM —
N150 P300C
- Image Generation with Stable Diffusion XL —
N150
- Video Generation via Frame-by-Frame Diffusion — draft
2 lessons, 0 validated
- Image Classification with TT-Forge —
P300C
- JAX and PyTorch/XLA on Tenstorrent —
P300C
🎯 Applications
4 lessons, 3 validated
- Coding Assistant with Aider — draft
- Native Video Animation with AnimateDiff —
N150
- OpenClaw AI Assistant on QuietBox 2 —
P300X2
- Generating Video on QuietBox 2 —
P300X2
🎓 Advanced Topics
2 lessons, 1 validated
- Bounty Program: Model Bring-Up — draft
- Exploring TT-Metalium —
N150 P300C
🎓 Custom Training
8 lessons, 0 validated
- Understanding Custom Training —
N150
- Dataset Fundamentals —
N150
- Configuration Patterns —
N150
- Fine-tuning Basics —
N150
- Multi-Device Training —
N150
- Experiment Tracking —
N150
- Model Architecture Basics —
N150
- Training from Scratch —
N150
☁️ Deployment
2 lessons, 2 validated
- Deploy tt-vscode-toolkit to Koyeb —
N150
- Deploy Your Work to Koyeb —
N150
👨🍳 Tenstorrent Cookbook
6 lessons, 6 validated
- Tenstorrent Cookbook Overview —
N150 P300C
- Recipe 1: Conway's Game of Life —
N150 P300C
- Recipe 2: Audio Signal Processing —
N150 P300C
- Recipe 3: Mandelbrot Fractal Explorer —
N150 P300C
- Recipe 4: Custom Image Filters —
N150 P300C
- Recipe 5: Particle Life Simulator —
N150 P300C
🧠 CS Fundamentals
7 lessons, 0 validated
- Module 1: RISC-V & Computer Architecture — draft
- Module 2: The Memory Hierarchy — draft
- Module 3: Parallel Computing — draft
- Module 4: Networks and Communication — draft
- Module 5: Synchronization — draft
- Module 6: Abstraction Layers — draft
- Module 7: Computational Complexity in Practice — draft
Key Features
Intelligent Hardware Detection
- Auto-detects device type (N150, N300, T3K, P100, P150, Galaxy)
- Provides hardware-specific commands and configurations
- Real-time telemetry monitoring (temperature, power, clock speed)
- Multi-device support with aggregate health status
Interactive Learning Experience
- Click-to-run commands from lesson content
- Persistent terminal sessions maintain environment state
- Visual progress tracking
- Hierarchical organization by difficulty and category
Production-Ready Code
- Tested templates for common workflows
- Best practices from Tenstorrent engineering team
- Scripts saved to
~/tt-scratchpad/ for easy customization
- Hardware-specific optimization examples
Multi-Framework Coverage
| Framework |
Purpose |
Use Case |
| vLLM |
Production LLM serving |
OpenAI-compatible API, high throughput |
| TT-Forge |
PyTorch compilation |
MLIR-based experimental compiler |
| TT-XLA |
JAX/PyTorch XLA |
Production compiler for JAX workflows |
| TT-Metal |
Low-level kernels |
Custom ops and hardware programming |
Hands-On Cookbook Projects
The Cookbook (Lesson 16) includes 5 interactive projects that run directly on Tenstorrent hardware:
Game of Life - Classic cellular automaton with TT-NN acceleration
View full animation →
|
Particle Life - Physics simulation with 10,000+ particles
View full animation →
|
Mandelbrot Set - Fractal rendering with hardware acceleration
|
Audio Processing - Mel spectrogram computation
|
Plus: Image filters (blur, sharpen, edge detection) - all with complete source code and interactive tutorials.
Documentation
User Documentation
- FAQ.md - Comprehensive troubleshooting (covers 90% of common issues)
- Lesson Content - Interactive lessons accessible via the extension
- CHANGELOG.md - Version history and release notes
Developer Documentation
Common Issues
"No hardware detected"
tt-smi -r # Reset and rescan
sudo tt-smi # Try with elevated permissions
See FAQ.md for complete diagnostic steps.
"ImportError: undefined symbol" (TT-XLA)
unset TT_METAL_HOME
unset TT_METAL_VERSION
TT-XLA requires clean environment. See Lesson 12 for details.
"vLLM won't start"
echo $TT_METAL_HOME # Should be ~/tt-metal
echo $MESH_DEVICE # Should match your hardware (e.g., N150)
See FAQ.md for systematic vLLM debugging.
For more troubleshooting, check the FAQ or join Discord.
Contributing
We welcome contributions! Here's how to get involved:
- Report Issues - Use our issue templates for bugs, content issues, feature requests, or new lesson ideas
- Improve Content - Lessons are in
content/lessons/*.md - submit PRs for corrections or improvements
- Add Features - See CONTRIBUTING.md for development setup
- Validate Lessons - Test lessons on hardware and update metadata
- Join Discussions - Participate in Discord and GitHub Discussions
See CONTRIBUTING.md for:
- Development setup instructions
- Architecture and design principles
- Code style and standards
- Testing requirements
- Pull request workflow
- Packaging and distribution
Latest Release: v0.0.268 (2026-01-09)
Highlights:
- 🐳 Docker support with container orchestration
- 📊 Consolidated status bar with sysfs telemetry
- 🎨 Color-coded hardware validation badges
- 🤝 Open-source documentation and GitHub workflows
- 🔧 CS Fundamentals lesson fixes and ttnn API improvements
Previous Release: v0.0.243 (2026-01-09)
Highlights:
- 📚 New technical documentation structure (ARCHITECTURE, TESTING, PACKAGING)
- 🎯 README streamlined (50% reduction in size)
- 🎨 Monochrome sidebar icon for better theme compatibility
See CHANGELOG.md for complete version history.
Version Support
| Version |
Status |
Notes |
| 0.1.x |
✅ Current |
Active development, full support |
| 0.0.x |
⚠️ Legacy |
Community support only |
License
This project is licensed under the Apache License 2.0 - see the LICENSE file for details.
License Understanding
This software assists in programming Tenstorrent products. However, making, using, or selling hardware, models, or IP may require the license of rights (such as patent rights) from Tenstorrent or others. See LICENSE_understanding.txt for details.
Third-Party Licenses
This extension depends on several open-source projects:
- marked (MIT) - Markdown parsing
- DOMPurify (Apache-2.0 or MPL-2.0) - HTML sanitization
- mermaid (MIT) - Diagram rendering
- jsdom (MIT) - DOM implementation
Run npm list --prod to see all production dependencies.
Acknowledgments
Built by the Tenstorrent community with contributions from:
- Tenstorrent Developer Relations and Engineering teams
- Open-source contributors worldwide
- Community members providing feedback and hardware validation
Special thanks to:
- Beta testers who validated lessons on real hardware across all device types
- Documentation contributors who improved clarity and caught errors
- Bug reporters who helped us fix issues quickly
- Community members suggesting new lessons and features
Tenstorrent Ecosystem:
Ready to start building AI on Tenstorrent hardware? Install the extension and open the Welcome page! 🚀
Questions? Check the FAQ or join our Discord community!
