tinygrad/docs/env_vars.md

List of environment variables that control tinygrad behavior.

This is a list of environment variable that control the runtime behavior of tinygrad and its examples. Most of these are self-explanatory, and are usually used to set an option at runtime.

Example: GPU=1 DEBUG=4 python3 -m pytest

However you can also decorate a function to set a value only inside that function.

# in tensor.py (probably only useful if you are a tinygrad developer)
@Context(DEBUG=4)
def numpy(self) -> ...

Or use contextmanager to temporarily set a value inside some scope:

with Context(DEBUG=0):
  a = Tensor.ones(10, 10)
  a *= 2

Global Variables

The columns of this list are are: Variable, Possible Value(s) and Description.

• A # means that the variable can take any integer value.

These control the behavior of core tinygrad even when used as a library.

Variable	Possible Value(s)	Description
DEBUG	[1-6]	enable debugging output, with 4 you get operations, timings, speed, generated code and more
GPU	[1]	enable the GPU backend
CUDA	[1]	enable CUDA backend
HSA	[1]	enable HSA backend
METAL	[1]	enable Metal backend (for Mac M1 and after)
METAL_XCODE	[1]	enable Metal using macOS Xcode SDK
CLANG	[1]	enable Clang backend
LLVM	[1]	enable LLVM backend
BEAM	[#]	number of beams in kernel beam search
GRAPH	[1]	create a graph of all operations (requires graphviz)
GRAPHUOPS	[1]	create a graph of uops (requires graphviz and saves at /tmp/uops.{svg,dot})
GRAPHPATH	[/path/to]	where to put the generated graph
DEFAULT_FLOAT	[HALF, ...]	specify the default float dtype (FLOAT32, HALF, BFLOAT16, FLOAT64, ...), default to FLOAT32
IMAGE	[1-2]	enable 2d specific optimizations
FLOAT16	[1]	use float16 for images instead of float32
PTX	[1]	enable the specialized PTX assembler for Nvidia GPUs. If not set, defaults to generic CUDA codegen backend.