performance_tuning
==================

.. py:module:: performance_tuning

.. autoapi-nested-parse::

   Quick performance benchmark for GFlowNet training.

   Runs per-iteration timing across environments (HyperGrid, DiffusionSampling,
   DiscreteEBM), losses (TB, SubTB, ModifiedDB), debug on/off, and torch.compile
   on/off (compiling the standard sampling loop when possible). Produces a bar plot
   saved to ~/performance_tuning.png.



Attributes
----------

.. autoapisummary::

   performance_tuning.LR
   performance_tuning.LR_LOGF
   performance_tuning.LR_LOGZ


Classes
-------

.. autoapisummary::

   performance_tuning.BenchmarkConfig
   performance_tuning.BenchmarkResult
   performance_tuning.BenchmarkSettings


Functions
---------

.. autoapisummary::

   performance_tuning.build_components
   performance_tuning.main
   performance_tuning.make_step_fn
   performance_tuning.maybe_compile
   performance_tuning.plot_results
   performance_tuning.resolve_device
   performance_tuning.run_benchmark
   performance_tuning.run_benchmarks
   performance_tuning.synchronize
   performance_tuning.time_iterations


Module Contents
---------------

.. py:class:: BenchmarkConfig

   .. py:attribute:: debug
      :type:  bool


   .. py:attribute:: env_name
      :type:  str


   .. py:property:: label
      :type: str



   .. py:attribute:: loss_name
      :type:  str


   .. py:attribute:: use_compile
      :type:  bool


.. py:class:: BenchmarkResult

   .. py:attribute:: compiled
      :type:  bool


   .. py:attribute:: config
      :type:  BenchmarkConfig


   .. py:attribute:: elapsed
      :type:  float


   .. py:attribute:: env_name
      :type:  str


   .. py:attribute:: mean_iter_ms
      :type:  float


   .. py:attribute:: reason
      :type:  str | None
      :value: None



   .. py:attribute:: skipped
      :type:  bool


   .. py:attribute:: std_iter_ms
      :type:  float


.. py:class:: BenchmarkSettings

   .. py:attribute:: batch_size
      :type:  int
      :value: 32



   .. py:attribute:: deterministic_mode
      :type:  bool
      :value: False



   .. py:attribute:: device
      :type:  str
      :value: 'cpu'



   .. py:attribute:: n_iters
      :type:  int
      :value: 200



   .. py:attribute:: output_path
      :type:  pathlib.Path | None


   .. py:attribute:: warmup_iters
      :type:  int
      :value: 50



.. py:data:: LR
   :value: 0.001


.. py:data:: LR_LOGF
   :value: 0.001


.. py:data:: LR_LOGZ
   :value: 0.1


.. py:function:: build_components(env_name, loss_name, debug, device)

   Create env, gflownet, optimizer for a given setup.


.. py:function:: main()

.. py:function:: make_step_fn(env, gflownet, optimizer, batch_size, device)

.. py:function:: maybe_compile(fn)

.. py:function:: plot_results(results, output_path = None, return_fig = False)

.. py:function:: resolve_device(requested = 'auto')

.. py:function:: run_benchmark(config, device, settings)

.. py:function:: run_benchmarks(settings, *, return_fig = False, verbose = False)

.. py:function:: synchronize(device)

.. py:function:: time_iterations(step_fn, device, warmup, n_iters)