Extend TVM tutorials in Colab

gallery/how_to/extend_tvm/bring_your_own_datatypes.py

@@ -23,6 +23,13 @@
 Note that the Bring Your Own Datatypes framework currently only handles **software emulated versions of datatypes**.
 The framework does not support compiling for custom accelerator datatypes out-of-the-box.
 
+To start, we must install TVM and MXNet (we will use a model from MXNet's model zoo):
+
+.. code-block:: bash
+
+    %%shell
+    pip install apache-tvm mxnet
+
 Datatype Libraries
 ------------------
 

gallery/how_to/extend_tvm/low_level_custom_pass.py

@@ -39,6 +39,13 @@
 - How a Schedule is lowered to either an IRModule class or a LLVM module. Otherwise,
   take a look at ``python/tvm/build_module.py`` to get some basics.
 
+TVM must also be installed, either from source or with a package manager:
+
+.. code-block:: bash
+
+    %%shell
+    pip install apache-tvm
+
 """
 
 # sphinx_gallery_start_ignore

gallery/how_to/extend_tvm/use_pass_infra.py

@@ -38,6 +38,14 @@
 This tutorial mainly demonstrates how developers can use the pass infra to perform
 a certain optimization and create an optimization pipeline for a Relay program.
 The same approach can be used for tir as well.
+
+To start, we'll install TVM:
+
+.. code-block:: bash
+
+    %%shell
+    pip install apache-tvm
+
 """
 
 # sphinx_gallery_start_ignore

gallery/how_to/extend_tvm/use_pass_instrument.py

@@ -31,7 +31,13 @@
 but an extension mechanism is available via the :py:func:`tvm.instrument.pass_instrument` decorator.
 
 This tutorial demonstrates how developers can use ``PassContext`` to instrument
-passes. Please also refer to the :ref:`pass-infra`.
+passes. Please also refer to the :ref:`pass-infra`. To start, we'll install TVM:
+
+.. code-block:: bash
+
+    %%shell
+    pip install apache-tvm
+
 """
 
 # sphinx_gallery_start_ignore