This is an automated email from the ASF dual-hosted git repository.
ruihangl pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
The following commit(s) were added to refs/heads/main by this push:
new b8ecf2f803 [Docs] Add Relax VM architecture overview in documentation
(#19350)
b8ecf2f803 is described below
commit b8ecf2f803a7b0cdfc382ba38fc74ee53227c5fe
Author: Shushi Hong <[email protected]>
AuthorDate: Sat Apr 4 22:56:26 2026 -0400
[Docs] Add Relax VM architecture overview in documentation (#19350)
This adds Relax VM architecture overview in documentation
---
docs/arch/index.rst | 23 ++++++++++++++++++++++-
1 file changed, 22 insertions(+), 1 deletion(-)
diff --git a/docs/arch/index.rst b/docs/arch/index.rst
index 878717c355..cb02c109b7 100644
--- a/docs/arch/index.rst
+++ b/docs/arch/index.rst
@@ -225,7 +225,28 @@ for learning-based optimizations.
introduction_to_module_serialization
device_target_interactions
-.. TODO(tvm-team) add a section about relax vm here
+Relax Virtual Machine
+^^^^^^^^^^^^^^^^^^^^^
+
+Relax defines *what* to compute — it is a graph-level IR that describes the
operators and dataflow
+of a model. The Relax Virtual Machine (VM) handles *how* to run it — it is the
runtime component
+that executes the compiled result. During compilation, ``tvm.compile()``
invokes ``VMCodeGen`` to
+translate Relax functions into a compact bytecode representation. The
resulting ``VMExecutable``
+bundles the bytecode together with a constant pool and per-function metadata,
and can be serialized
+to disk for deployment.
+
+The VM uses a register-based interpreter with an intentionally minimal
instruction set — only four
+opcodes: ``Call``, ``Ret``, ``Goto``, and ``If``. The VM itself performs no
mathematical computation;
+it only orchestrates control flow (function calls, conditional branches,
loops). The actual
+compute-intensive work — matrix multiplications, convolutions, and other
operators — is carried out
+by TIR functions that have been compiled down to native GPU/CPU kernels, or by
external libraries
+such as cuBLAS and cuDNN. The VM dispatches to them through the PackedFunc
mechanism. Internally the
+VM recognizes three function kinds: *PackedFunc* for external C/C++ functions,
*VMFunc* for
+bytecode-interpreted Relax functions, and *VMTIRFunc* for compiled TIR kernels.
+
+On the Python side, users interact with the VM through
``relax.VirtualMachine(executable, device)``,
+which provides both a direct invocation interface and a stateful set-input /
invoke / get-output
+interface suitable for RPC-based remote execution.
tvm/node
--------
