Deprecate rope_scaling assignment and normalize legacy RoPE keys

src/transformers/configuration_utils.py

@@ -20,6 +20,7 @@
 import json
 import math
 import os
+import warnings
 from collections.abc import Sequence
 from dataclasses import MISSING, dataclass, fields
 from functools import wraps
@@ -713,7 +714,15 @@ def rope_scaling(self):
 
     @rope_scaling.setter
     def rope_scaling(self, value):
+        warnings.warn(
+            "`rope_scaling` is deprecated and will be removed in a future version. Set `rope_parameters` "
+            "instead, with `rope_type` and `rope_theta` included.",
+            FutureWarning,
+        )
+        if value is None:
+            return
         self.rope_parameters = value
+        self.standardize_rope_params()
 
     def save_pretrained(self, save_directory: str | os.PathLike, push_to_hub: bool = False, **kwargs):
         """

tests/utils/test_modeling_rope_utils.py

@@ -16,6 +16,7 @@
 import logging as stdlib_logging
 import math
 import unittest
+import warnings
 
 from parameterized import parameterized
 
@@ -38,6 +39,28 @@
 
 @require_torch
 class RopeTest(unittest.TestCase):
+    def test_rope_scaling_setter_is_deprecated(self):
+        config = LlamaConfig(rope_theta=500000.0)
+        with warnings.catch_warnings(record=True) as caught:
+            warnings.simplefilter("always")
+            config.rope_scaling = {
+                "type": "yarn",
+                "factor": 4.0,
+                "rope_theta": 500000.0,
+                "original_max_position_embeddings": 4096,
+            }
+        self.assertTrue(any(issubclass(w.category, FutureWarning) for w in caught))
+        self.assertEqual(config.rope_parameters["rope_type"], "yarn")
+        LlamaRotaryEmbedding(config)
+
+    def test_rope_scaling_setter_ignores_none(self):
+        config = LlamaConfig(rope_theta=500000.0)
+        with warnings.catch_warnings(record=True) as caught:
+            warnings.simplefilter("always")
+            config.rope_scaling = None
+        self.assertTrue(any(issubclass(w.category, FutureWarning) for w in caught))
+        self.assertEqual(config.rope_parameters["rope_theta"], 500000.0)
+
     def get_config_with_rope_parameters(
         self, rope_params: dict, is_nested: bool = False, same_rope_per_layer: bool = False
     ):