tinygrad for inference¶
Start with the complete guide, then run the CPU projects. This course pins tinygrad 0.14.0 and includes executed operator, gradient, training, checkpoint, compiler/JIT and model-contract exercises.
The mathematical references use only the standard library. The framework path requires the pinned wheel and a supported CPU compiler/runtime. Recorded results distinguish actual CPU execution from optional integrations.
| Read next | Purpose |
|---|---|
| Concept overview | Connect tensor semantics, compilation and deployment |
| Source walkthrough | Follow the actual v0.14.0 source layout |
| Operation lessons | Work elementwise, reduction and layout examples |
| Project commands | Reproduce the finite CPU progression |
The standalone package is separate from openpilot's pinned tinygrad submodule. Return to the autonomous vehicle track for the observation/action contract and validation boundaries.