Internal change.

PiperOrigin-RevId: 603537825

init2winit/hyperparameters.py

@@ -102,7 +102,7 @@ def build_hparams(model_name,
                   hparam_file,
                   hparam_overrides,
                   input_pipeline_hps=None,
-                  allow_unrecognized_hparams=False):
+                  allowed_unrecognized_hparams=None):
   """Build experiment hyperparameters.
 
   Args:
@@ -115,11 +115,12 @@ def build_hparams(model_name,
       string encoding of this hyperparameter override dict. Note that this is
       applied after the hyperparameter file overrides.
     input_pipeline_hps: a dict of hyperparameters for performance tuning.
-    allow_unrecognized_hparams: if set, hparam overrides are allowed to
-      introduce new hparam keys that will most likely be ignored. Downgrading
-      unrecognized hparams from an error to a warning can be useful when trying
-      to tune using a shared search space over multiple workloads that don't all
-      support the same set of hyperparameters.
+    allowed_unrecognized_hparams: An optional list of hparam keys that hparam
+      overrides are allowed to introduce. There is no guaranteed these new
+      hparam keys will be ignored. Downgrading an explicit list of unrecognized
+      hparams from an error to a warning can be useful when trying to tune using
+      a shared search space over multiple workloads that don't all support the
+      same set of hyperparameters.
 
   Returns:
     A ConfigDict of experiment hyperparameters.
@@ -186,10 +187,13 @@ def build_hparams(model_name,
         merged['optimizer'] != hparam_overrides['optimizer']):
       merged['opt_hparams'] = {}
     hparam_overrides = expand_dot_keys(hparam_overrides)
-    if allow_unrecognized_hparams:
+    if allowed_unrecognized_hparams:
       new_keys = [k for k in hparam_overrides if k not in merged]
       if new_keys:
         logging.warning('Unrecognized top-level hparams: %s', new_keys)
+      if any(k not in allowed_unrecognized_hparams for k in new_keys):
+        raise ValueError(
+            f'Unrecognized top-level hparams not in allowlist: {new_keys}')
       with merged.unlocked():
         merged.update(hparam_overrides)
     else:

init2winit/main.py

@@ -136,9 +136,10 @@
     'hparam_file', None, 'Optional path to hparam json file for overriding '
     'hyperparameters. Hyperparameters are loaded before '
     'applying --hparam_overrides.')
-flags.DEFINE_bool(
-    'allow_unrecognized_hparams', False,
-    'Downgrades unrecognized hparam override keys from an error to a warning.')
+flags.DEFINE_list(
+    'allowed_unrecognized_hparams', [],
+    'Downgrades unrecognized hparam override keys from an error to a warning '
+    'for the supplied list of keys.')
 flags.DEFINE_string(
     'training_metrics_config', '',
     'JSON representation of the training metrics config.')
@@ -189,7 +190,7 @@ def _run(
     early_stopping_min_steps,
     eval_steps,
     hparam_file,
-    allow_unrecognized_hparams,
+    allowed_unrecognized_hparams,
     hparam_overrides,
     initializer_name,
     model_name,
@@ -221,7 +222,7 @@ def _run(
       hparam_file=hparam_file,
       hparam_overrides=hparam_overrides,
       input_pipeline_hps=input_pipeline_hps,
-      allow_unrecognized_hparams=allow_unrecognized_hparams)
+      allowed_unrecognized_hparams=allowed_unrecognized_hparams)
 
   # Note that one should never tune an RNG seed!!! The seed is only included in
   # the hparams for convenience of running hparam trials with multiple seeds per
@@ -349,7 +350,7 @@ def main(unused_argv):
         early_stopping_min_steps=FLAGS.early_stopping_min_steps,
         eval_steps=eval_steps,
         hparam_file=FLAGS.hparam_file,
-        allow_unrecognized_hparams=FLAGS.allow_unrecognized_hparams,
+        allowed_unrecognized_hparams=FLAGS.allowed_unrecognized_hparams,
         hparam_overrides=FLAGS.hparam_overrides,
         initializer_name=FLAGS.initializer,
         model_name=FLAGS.model,

init2winit/test_hyperparameters.py

@@ -64,15 +64,15 @@ def test_unrecognized_override(self):
           dataset_name='lm1b_v2',
           hparam_file=None,
           hparam_overrides=hps_overrides,
-          allow_unrecognized_hparams=False,
+          allowed_unrecognized_hparams=[],
       )
     merged_hps = hyperparameters.build_hparams(
         model_name='transformer',
         initializer_name='noop',
         dataset_name='lm1b_v2',
         hparam_file=None,
         hparam_overrides=hps_overrides,
-        allow_unrecognized_hparams=True,
+        allowed_unrecognized_hparams=['lr_hparamsTYPO'],
     )
     expected_added_field = {'base_lr': 77.0}
     self.assertEqual(merged_hps.lr_hparamsTYPO.to_dict(), expected_added_field)