triton-lang/triton

Documentation	Nightly Wheels

Triton Developer Conference registration is open: https://tritonconference2026.splashthat.com/

Folks, please join us at the Triton Developer Conference on October 19th at the San Jose Convention Center, just before PTC at the same venue! We are finalizing an agenda with keynotes, technical talks, lightning talks, and posters.

Registration and attendance are free. Breakfast, lunch, as well as a happy hour organized by our Nvidia colleagues, are included! Learn something new, share your experience, mingle with colleagues, and make new friends!

To allow us to prepare correctly, please register and reserve your spot at your earliest convenience: https://tritonconference2026.splashthat.com/ You will see more info, including a draft agenda there soon, as well.

For previous conference materials, see:

• 2025 Conference Materials
• 2024 Conference Materials
• 2023 Conference Materials

This is the development repository of Triton, a language and compiler for writing highly efficient custom Deep-Learning primitives. The aim of Triton is to provide an open-source environment to write fast code at higher productivity than CUDA, but also with higher flexibility than other existing DSLs.

The foundations of this project are described in the following MAPL2019 publication: Triton: An Intermediate Language and Compiler for Tiled Neural Network Computations. Please consider citing this work if you use Triton!

The official documentation contains installation instructions and tutorials. See also these third-party Triton puzzles, which can all be run using the Triton interpreter -- no GPU required.

You can install the latest stable release of Triton from pip:

pip install triton

Binary wheels are available for CPython 3.10-3.14.

git clone https://github.com/triton-lang/triton.git
cd triton

pip install -r python/requirements.txt # build-time dependencies
pip install -e .

Or with a virtualenv:

git clone https://github.com/triton-lang/triton.git
cd triton

python -m venv .venv --prompt triton
source .venv/bin/activate

pip install -r python/requirements.txt # build-time dependencies
pip install -e .

Triton uses LLVM to generate code for GPUs and CPUs. Normally, the Triton build downloads a prebuilt LLVM, but you can also build and use LLVM from source.

LLVM does not have a stable API, so the Triton build will not work at an arbitrary LLVM version.

For convenience, use the following command to build LLVM and install Triton with the custom LLVM:

make dev-install-llvm

• Set TRITON_BUILD_WITH_CLANG_LLD=true as an environment variable to use clang and lld. lld in particular results in faster builds.

• Set TRITON_BUILD_WITH_CCACHE=true to build with ccache.

• Set TRITON_HOME=/some/path to change the location of the .triton directory where Triton's cache is located and downloads are stored during the build. By default, this is the user's home directory. It can be changed anytime.

• If you're running out of memory when building Triton, specify the MAX_JOBS environment variable (to the pip install -e . command) to limit the number of jobs.

• Pass --no-build-isolation to pip install to make nop builds faster. Without this, every invocation of pip install uses a different symlink to cmake, and this forces ninja to rebuild most of the .a files.

• The build system creates a compile_commands.json file under the Triton repo directory. This file is used by VSCode IntelliSense and clangd to provide code completion and other features for C++ code.

  If IntelliSense does not work, you can try the following steps:

  • Do a local build. Run command pip install -e ..
  • Get the full path to the compile_commands.json file produced by the build: find ./build -name 'compile_commands.json' | xargs readlink -f. You might get a full path similar to /Users/{username}/triton/build/cmake.macosx-11.1-arm64-cpython-3.12/compile_commands.json.
  • In VSCode, install the C/C++ extension, then open the command palette (Shift + Command + P on Mac, or Shift + Ctrl + P on Windows/Linux) and open C/C++: Edit Configurations (UI).
  • Open "Advanced Settings" and paste the full path to compile_commands.json into the "Compile Commands" textbox.

There currently isn't a turnkey way to run all the Triton tests, but you can follow the following recipe:

# One-time setup.  Note this will reinstall local Triton because torch
# overwrites it with the public version.
$ make dev-install

# To run all tests (requires a GPU)
$ make test

# Or, to run tests without a gpu
$ make test-nogpu

For detailed instructions on how to debug Triton's frontend, please refer to this tutorial. The following includes additional tips for hacking on Triton's backend.

Configuration knobs

See python/triton/knobs.py for the full list of configuration knobs. You can set those knobs directly in python or use environment variables to control them. Below are some of the environment variables you can specify (see knobs.py for the full list):

• MLIR_ENABLE_DUMP=1 dumps the IR before every MLIR pass Triton runs, for all kernels. Use MLIR_ENABLE_DUMP=kernelName to dump for a specific kernel only.

  • Triton cache can interfere with the dump. In cases where MLIR_ENABLE_DUMP=1 does not work, try cleaning your triton cache: rm -r ~/.triton/cache/*.

• MLIR_DUMP_PATH specifies where MLIR_ENABLE_DUMP will dump to. If unset will dump to stderr.

• LLVM_IR_ENABLE_DUMP=1 dumps the IR before every pass run over the LLVM IR.

• TRITON_REPRODUCER_PATH=<reproducer_path> will generate an MLIR reproducer file at <reproducer_path> before each MLIR compiler stage. If any of the stages fail, <reproducer_path> will be a local MLIR reproducer captured right before the failing pass.

• TRITON_INTERPRET=1 uses the Triton interpreter instead of running on the GPU. You can insert Python breakpoints in your kernel code!

• TRITON_ENABLE_LLVM_DEBUG=1 passes -debug to LLVM, printing a lot of debugging information to stdout. If this is too noisy, run with just TRITON_LLVM_DEBUG_ONLY instead to limit the output.

  • An alternative way to reduce output noisiness is running with LLVM_IR_ENABLE_DUMP=1, extract the IR before the LLVM pass of interest, and then run LLVM's opt standalone, perhaps passing -debug-only=foo on the command line.

• TRITON_LLVM_DEBUG_ONLY=<comma-separated> is the equivalent of LLVM's -debug-only command-line option. This limits the LLVM debug output to specific pass or component names (which are specified using #define DEBUG_TYPE throughout LLVM and Triton) in order to allow the debug output to be less noisy. TRITON_LLVM_DEBUG_ONLY allows for one or more comma separated values to be specified (eg TRITON_LLVM_DEBUG_ONLY="tritongpu-remove-layout-conversions" or TRITON_LLVM_DEBUG_ONLY="tritongpu-remove-layout-conversions,regalloc").

• TRITON_ENABLE_ASAN=1 invokes the LLVM address sanitizer for memory leak and out of bounds access detection. Currently only supported on the AMD backend. This must be run using the ASAN libraries documented here.

  • When enabling the address sanitizer it is recommended to disable various memory caching strategies both within the ROCm stack and PyTorch. This will give the address sanitizer the best chance at finding the memory fault where it originates. See this test for more details.

• USE_IR_LOC={ttir,ttgir} reparses the IR such that the location information will be the line number of the IR file with that particular extension, instead of line number of the python file. This can provide a direct mapping from the IR to llir/ptx. When used with performance tools, it can provide a breakdown on IR instructions.

• TRITON_PRINT_AUTOTUNING=1 prints out the best autotuning config and total time spent for each kernel after autotuning is complete.

• DISABLE_LLVM_OPT will disable llvm optimizations for make_llir and make_ptx if its value is true when parsing as Bool. Otherwise, it will be parsed as a list of flags to disable llvm optimizations. One usage case is DISABLE_LLVM_OPT="disable-lsr" Loop strength reduction is known to cause up to 10% performance changes for certain kernels with register pressure.

• TRITON_ALWAYS_COMPILE=1 forces to compile kernels regardless of cache hit.

• MLIR_ENABLE_TIMING dumps the timing information for each MLIR pass.

• LLVM_ENABLE_TIMING dumps the timing information for each LLVM pass.

• TRITON_DEFAULT_FP_FUSION overrides the default behavior of disabling fp fusion (mul+add->fma).

• TRITON_FORCE_DISABLE_FP_FUSION=1 disables implicit fp fusion even when a kernel requests enable_fp_fusion=True. Explicit fused operations such as tl.fma are unchanged.

• MLIR_ENABLE_DIAGNOSTICS=<comma-separated> controls diagnostic emission in MLIR. Options are: warnings, remarks, stacktraces, operations. Use comma-separated values to customize output. For example, MLIR_ENABLE_DIAGNOSTICS=remarks,operations enables remarks and IR operations, while MLIR_ENABLE_DIAGNOSTICS=warnings,stacktraces enables warnings with stacktraces. By default, only errors are shown. Setting warnings includes errors and warnings; remarks includes errors, warnings, and remarks.

• MLIR_ENABLE_REMARK is deprecated. Please use MLIR_ENABLE_DIAGNOSTICS=remarks.

• TRITON_KERNEL_DUMP enables the dumping of the IR from each compilation stage and the final ptx/amdgcn.

• TRITON_DUMP_DIR specifies the directory to save the dumped IR and ptx/amdgcn when TRITON_KERNEL_DUMP is set to 1.

• TRITON_KERNEL_OVERRIDE enables the override of the compiled kernel with a user-specified IR/ptx/amdgcn at the beginning of each compilation stage.

• TRITON_OVERRIDE_DIR specifies the directory from which to load the IR/ptx/amdgcn files when TRITON_KERNEL_OVERRIDE is set to 1.

• TRITON_F32_DEFAULT sets the default input precision of tl.dot when using 32-bit floats, which can be either ieee, tf32, or tf32x3.

• TRITON_FRONT_END_DEBUGGING=1 disables exception wrapping when an error occurs in the compiler frontend, allowing the full stack trace to be seen.

• TRITON_DISABLE_LINE_INFO=1 removes all line information from the module.

• PTXAS_OPTIONS passes additional command-line options to the PTX assembler ptxas (only on NVIDIA).

• LLVM_EXTRACT_DI_LOCAL_VARIABLES emit full debug info, allowing for eval of values in gpu debuggers (ie cuda-gdb, rocm-gdb etc)

• TRITON_DEFAULT_BACKEND=<backend> optionally sets the default backend used by Triton when constructing the active driver (i.e., triton.runtime.driver.active).

Kernel Override Steps

export TRITON_ALWAYS_COMPILE=1
export TRITON_KERNEL_DUMP=1
export TRITON_DUMP_DIR=<dump_dir>
export TRITON_KERNEL_OVERRIDE=1
export TRITON_OVERRIDE_DIR=<override_dir>
# Step 1: Run the kernel once to dump kernel's IRs and ptx/amdgcn in $TRITON_DUMP_DIR
# Step 2: Copy $TRITON_DUMP_DIR/<kernel_hash> to $TRITON_OVERRIDE_DIR
# Step 3: Delete the stages that you do not want to override and modify the stage you do want to override
# Step 4: Run the kernel again to see the overridden result

Compiler Pipeline Inspection Steps To introspect the pipeline add_stages, before running your kernels, simply set the add_stages_inspection_hook like so:

def inspect_stages(_self, stages, options, language, capability):
    # inspect or modify add_stages here
triton.knobs.runtime.add_stages_inspection_hook = inspect_stages

Examples of how to use this for out of tree plugin passes is here

Version 2.0 is out! New features include:

• Many, many bug fixes
• Performance improvements
• Backend rewritten to use MLIR
• Support for kernels that contain back-to-back matmuls (e.g., flash attention)

Community contributions are more than welcome, whether it be to fix bugs or to add new features at github. For more detailed instructions, please visit our contributor's guide.

Supported Platforms:

• Linux

Supported Hardware:

• NVIDIA GPUs (Compute Capability 8.0+)
• AMD GPUs (ROCm 6.2+)
• Under development: CPUs

Dev Containers for the Triton project are available from the triton-dev-containers repository.

• Consistency: All developers can work with the same development environment, ensuring uniform behavior across different systems.
• Isolation: The container prevents potential conflicts with software installed on your local machine.
• Portability: Easily share the development environment with team members, minimizing onboarding time and setup issues.

For detailed instructions on how to use the dev containers, please see the dev container user guide.