transformers 5.0.0rc0__py3-none-any.whl → 5.0.0rc2__py3-none-any.whl

transformers/modeling_rope_utils.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import math
+import warnings
 from functools import wraps
 from typing import TYPE_CHECKING, Optional, TypedDict
 
@@ -45,17 +46,19 @@ def dynamic_rope_update(rope_forward):
     def longrope_frequency_update(self, position_ids, device, layer_type=None):
         """Longrope uses long factor if sequence is larger than original pretraining length, short otherwise."""
         seq_len = torch.max(position_ids) + 1
-        original_max_position_embeddings = getattr(
-            self.config, "original_max_position_embeddings", self.config.max_position_embeddings
-        )
+
         if layer_type is None:
             rope_type = self.rope_type
             original_inv_freq = self.original_inv_freq
             prefix = ""
+            original_max_position_embeddings = self.config.rope_parameters["original_max_position_embeddings"]
         else:
             rope_type = self.rope_type[layer_type]
             original_inv_freq = getattr(self, f"{layer_type}_original_inv_freq")
             prefix = f"{layer_type}_"
+            original_max_position_embeddings = self.config.rope_parameters[layer_type][
+                "original_max_position_embeddings"
+            ]
 
         if seq_len > original_max_position_embeddings:
             if not hasattr(self, f"{layer_type}_long_inv_freq"):
@@ -222,7 +225,6 @@ def _compute_dynamic_ntk_parameters(
         Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the
         post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE).
     """
-    # TODO (joao): use the new `original_max_position_embeddings` from rope_parameters
     # For backward compatibility standardize the `rope_parameters_dict` if it uses old format
     config.standardize_rope_params()
     rope_parameters_dict = config.rope_parameters[layer_type] if layer_type is not None else config.rope_parameters
@@ -231,30 +233,29 @@ def _compute_dynamic_ntk_parameters(
     partial_rotary_factor = rope_parameters_dict.get("partial_rotary_factor", 1.0)
     head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
     dim = int(head_dim * partial_rotary_factor)
-    max_position_embeddings = config.max_position_embeddings
     factor = rope_parameters_dict["factor"]
     attention_factor = 1.0  # Unused in this type of RoPE
 
     # seq_len: default to max_position_embeddings, e.g. at init time
     if seq_len is None:
-        seq_len = max_position_embeddings
+        seq_len = config.max_position_embeddings
     elif isinstance(seq_len, torch.Tensor):
         seq_len = torch.maximum(
             seq_len,
-            torch.tensor(max_position_embeddings, dtype=seq_len.dtype, device=seq_len.device),
+            torch.tensor(config.max_position_embeddings, dtype=seq_len.dtype, device=seq_len.device),
         )
     else:
-        seq_len = max(seq_len, max_position_embeddings)
+        seq_len = max(seq_len, config.max_position_embeddings)
 
     # Compute the inverse frequencies
-    base = base * ((factor * seq_len / max_position_embeddings) - (factor - 1)) ** (dim / (dim - 2))
+    base = base * ((factor * seq_len / config.max_position_embeddings) - (factor - 1)) ** (dim / (dim - 2))
     inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2, dtype=torch.int64).to(device=device, dtype=torch.float) / dim))
     return inv_freq, attention_factor
 
 
 def _compute_yarn_parameters(
     config: "PreTrainedConfig",
-    device: "torch.device",
+    device: Optional["torch.device"] = None,
     seq_len: Optional[int] = None,
     layer_type: Optional[str] = None,
 ) -> tuple["torch.Tensor", float]:
@@ -291,8 +292,7 @@ def _compute_yarn_parameters(
                     `mscale_all_dim` are provided, `mscale_all_dim` acts scalar augmenting `log(factor)` when computing
                     the denominator for the inferred value of `attention_factor`. If not provided, `attention_factor`
                     will be calculated based on `factor` only.
-                *   `original_max_position_embeddings` (`int`, *optional*): The original max position embeddings used
-                    during pretraining. If not provided, the function falls back to `max_position_embeddings`.
+                *   `original_max_position_embeddings` (`int`): The original max position embeddings used during pretraining.
                 *   `truncate` (`bool`, *optional*): Whether to truncate the correction range.
 
             Additionally, this function will make use of the following properties if they are found in the config:
@@ -323,15 +323,13 @@ def _compute_yarn_parameters(
     attention_factor = rope_parameters_dict.get("attention_factor")
     mscale = rope_parameters_dict.get("mscale")
     mscale_all_dim = rope_parameters_dict.get("mscale_all_dim")
+    original_max_position_embeddings = rope_parameters_dict["original_max_position_embeddings"]
 
-    # NOTE: DeekSeek-V3 (and potentially other models) modify `max_position_embeddings` and have a
-    # `original_max_position_embeddings` field containing the pretrained value. They use the ratio between these two
-    # values to compute the default attention scaling factor, instead of using `factor`.
-    if "original_max_position_embeddings" in rope_parameters_dict:
-        original_max_position_embeddings = rope_parameters_dict["original_max_position_embeddings"]
+    # NOTE: DeekSeek-V3 (and potentially other models) have `original_max_position_embeddings` field
+    # containing the pretrained value. They use the ratio between `max_position_embeddings` and this value
+    # to compute the default attention scaling factor, instead of using `factor`.
+    if factor is None:
         factor = config.max_position_embeddings / original_max_position_embeddings
-    else:
-        original_max_position_embeddings = config.max_position_embeddings
 
     def get_mscale(scale, mscale=1):
         if scale <= 1:
@@ -392,7 +390,7 @@ def _compute_yarn_parameters(
 
 def _compute_longrope_parameters(
     config: "PreTrainedConfig",
-    device: "torch.device",
+    device: Optional["torch.device"] = None,
     seq_len: Optional[int] = None,
     layer_type: Optional[str] = None,
 ) -> tuple["torch.Tensor", float]:
@@ -439,7 +437,6 @@ def _compute_longrope_parameters(
         Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the
         post-processing scaling factor applied to the computed cos/sin.
     """
-    # TODO (joao): use the new `original_max_position_embeddings` from rope_parameters
     # For backward compatibility standardize the `rope_parameters_dict` if it uses old format
     config.standardize_rope_params()
     rope_parameters_dict = config.rope_parameters[layer_type] if layer_type is not None else config.rope_parameters
@@ -453,14 +450,13 @@ def _compute_longrope_parameters(
     short_factor = rope_parameters_dict["short_factor"]
     factor = rope_parameters_dict.get("factor")
     attention_factor = rope_parameters_dict.get("attention_factor")
+    original_max_position_embeddings = rope_parameters_dict["original_max_position_embeddings"]
 
     # NOTE: Phi3 (and potentially other models) modify `max_position_embeddings` and have a
     # `original_max_position_embeddings` field containing the pretrained value. They use the ratio between these two
     # values to compute the default attention scaling factor, instead of using `factor`.
-    if original_max_position_embeddings := getattr(config, "original_max_position_embeddings", None):
+    if factor is None:
         factor = config.max_position_embeddings / original_max_position_embeddings
-    else:
-        original_max_position_embeddings = config.max_position_embeddings
 
     # Sets the attention factor as suggested in the paper
     if attention_factor is None:
@@ -482,7 +478,7 @@ def _compute_longrope_parameters(
 
 def _compute_llama3_parameters(
     config: "PreTrainedConfig",
-    device: "torch.device",
+    device: Optional["torch.device"] = None,
     seq_len: Optional[int] = None,
     layer_type: Optional[str] = None,
 ) -> tuple["torch.Tensor", float]:
@@ -586,7 +582,7 @@ class RopeParameters(TypedDict, total=False):
             most scaling types, a `factor` of x will enable the model to handle sequences of length x *
             original maximum pre-trained length.
         original_max_position_embeddings (`int`, *optional*):
-            Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during
+            Used with 'yarn', 'longrope' and 'llama3'. The original max position embeddings used during
             pretraining.
         attention_factor (`float`, *optional*):
             Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention
@@ -640,6 +636,7 @@ class RotaryEmbeddingConfigMixin:
 
         # Standardize and validate the correctness of rotary position embeddings parameters
         self.rope_parameters.setdefault("rope_theta", kwargs.pop("rope_theta", self.default_theta))
+
         if "partial_rotary_factor" in kwargs:
             self.rope_parameters.setdefault("partial_rotary_factor", kwargs["partial_rotary_factor"])
             ignore_keys_at_rope_validation = {"partial_rotary_factor"}
@@ -653,25 +650,47 @@ class RotaryEmbeddingConfigMixin:
         Helper to standardize the config's rope params field by ensuring the params are defined for each
         later type. For old model the fn will duplicate a single rope param in each layer type (backward compatibility)
         """
-        # Move `rope_theta` and `partial_rotary_factor` to the params dict, if not there yet
+        # Move `rope_theta` and `partial_rotary_factor` to the `rope_parameters`, if not there yet
         rope_theta = getattr(self, "rope_theta", None)
         partial_rotary_factor = getattr(self, "partial_rotary_factor", None)
-        rope_parameters = self.rope_parameters
+        rope_parameters = getattr(self, "rope_parameters", None) or {}
+        layer_types = getattr(self, "layer_types", None)
 
+        # Case 0: no RoPE params defined
+        if not (rope_parameters or rope_theta):
+            # partial_rotary_factor without rope_theta is invalid, so we don't check for it here
+            logger.warning("`standardize_rope_params` was called but no RoPE parameters were found.")
+            return
         # Case 1: RoPE param keys do not intersect with possible `layer_types` -> one global dict
-        if getattr(self, "layer_types", None) is None or not set(rope_parameters.keys()).issubset(self.layer_types):
+        elif layer_types is None or rope_parameters == {} or not set(rope_parameters.keys()).issubset(layer_types):
             rope_parameters.setdefault("rope_type", rope_parameters.get("type", "default"))
             rope_parameters.setdefault("rope_theta", rope_theta)
             if partial_rotary_factor is not None:
                 rope_parameters["partial_rotary_factor"] = partial_rotary_factor
+
+            # Move pretraining-time maximum length to rope parameter dict for RoPE types with scaling
+            if rope_parameters["rope_type"] in ["llama3", "yarn", "longrope"]:
+                if hasattr(self, "original_max_position_embeddings"):
+                    # NOTE: Phi3 (and potentially other models) save `original_max_position_embeddings` field
+                    # containing the pretrained value outside rope parameters. This is an exception case where we
+                    # give priority to `self.original_max_position_embeddings
+                    self.rope_parameters["original_max_position_embeddings"] = self.original_max_position_embeddings
+                else:
+                    self.rope_parameters.setdefault("original_max_position_embeddings", self.max_position_embeddings)
+
         # Case 2: different RoPE for each layer -> several params as nested dict
         else:
-            for layer_type in self.layer_types:
+            for layer_type in set(layer_types):
                 rope_parameters[layer_type].setdefault("rope_type", rope_parameters[layer_type].get("type", "default"))
                 rope_parameters[layer_type].setdefault("rope_theta", rope_theta)
                 if partial_rotary_factor is not None:
                     rope_parameters[layer_type]["partial_rotary_factor"] = partial_rotary_factor
 
+                if rope_parameters[layer_type]["rope_type"] in ["llama3", "yarn", "longrope"]:
+                    self.rope_parameters[layer_type].setdefault(
+                        "original_max_position_embeddings", self.max_position_embeddings
+                    )
+
         self.rope_parameters = rope_parameters
 
     def validate_rope(self: "PreTrainedConfig", ignore_keys: Optional[set] = None):
@@ -691,14 +710,14 @@ class RotaryEmbeddingConfigMixin:
 
         for rope_parameters in rope_parameters_dict.values():
             rope_type = rope_parameters.get("rope_type", rope_parameters.get("type", "default"))
-            validation_fn = getattr(self, f"_validate_{rope_type}_rope_parameters")
+            validation_fn = getattr(self, f"_validate_{rope_type}_rope_parameters", None)
             rope_parameters["rope_type"] = rope_type
 
             if validation_fn is not None:
                 validation_fn(rope_parameters, ignore_keys=ignore_keys)
             else:
                 logger.warning(
-                    f"Missing validation function mapping in `ROPE_VALIDATION_FUNCTIONS` for 'rope_type'='{rope_type}'"
+                    f"Missing validation function in 'RotaryEmbeddingConfigMixin' for 'rope_type'='{rope_type}'"
                 )
 
     def _validate_default_rope_parameters(self, rope_parameters: dict, ignore_keys: Optional[set] = None):
@@ -718,26 +737,24 @@ class RotaryEmbeddingConfigMixin:
             logger.warning(f"`rope_parameters`'s factor field must be a float >= 1, got {factor}")
 
     def _validate_dynamic_rope_parameters(self, rope_parameters: dict, ignore_keys: Optional[set] = None):
-        # TODO (joao): update logic for the inclusion of `original_max_position_embeddings`
-        optional_keys = {"original_max_position_embeddings"}
         required_keys = {"rope_type", "factor"}
         received_keys = set(rope_parameters.keys())
         rope_type = rope_parameters["rope_type"]
-        self._check_received_keys(rope_type, received_keys, required_keys, optional_keys, ignore_keys=ignore_keys)
+        self._check_received_keys(rope_type, received_keys, required_keys, ignore_keys=ignore_keys)
 
         factor = rope_parameters["factor"]
         if factor is None or not isinstance(factor, float) or factor < 1.0:
             logger.warning(f"`rope_parameters`'s factor field must be a float >= 1, got {factor}")
 
     def _validate_yarn_rope_parameters(self, rope_parameters: dict, ignore_keys: Optional[set] = None):
-        required_keys = {"rope_type", "factor", "rope_theta"}
+        required_keys = {"rope_type", "factor", "rope_theta", "original_max_position_embeddings"}
         optional_keys = {
             "attention_factor",
             "beta_fast",
             "beta_slow",
-            "original_max_position_embeddings",
             "mscale",
             "mscale_all_dim",
+            "truncate",
         }
         received_keys = set(rope_parameters.keys())
         rope_type = rope_parameters["rope_type"]
@@ -765,37 +782,24 @@ class RotaryEmbeddingConfigMixin:
                 f"(defaults to 32 if None) and beta_slow={beta_slow} (defaults to 1 if None)"
             )
 
-        # Models should set `config.rope_parameters["original_max_position_embeddings"]` to their original (pre-yarn) context
-        # length, with `config.max_position_embeddings` corresponding to their post-yarn context length.
-        # However, for BC purposes, we allow the former to be unset.
-        original_max_position_embeddings = self.rope_parameters.get("original_max_position_embeddings")
-        if original_max_position_embeddings is not None:
-            # Double-check: `factor` should be the ratio between the pre-yarn and post-yarn context lengths.
-            implicit_factor = self.max_position_embeddings / original_max_position_embeddings
-            if implicit_factor != factor:
-                logger.warning_once(
-                    f"The explicitly set RoPE scaling factor (config.rope_parameters['factor'] = {factor}) does not match "
-                    "the ratio implicitly set by other parameters (implicit factor = "
-                    "post-yarn context length / pre-yarn context length = "
-                    "config.max_position_embeddings / config.rope_parameters['original_max_position_embeddings'] = "
-                    f"{implicit_factor}). Using the explicit factor ({factor}) in YaRN. This may cause unexpected "
-                    "behaviour in model usage, please correct the 'max_position_embeddings' fields in the model config."
-                )
-        # No `config.rope_parameters["original_max_position_embeddings"]`. Is `config.max_position_embeddings` the
-        # pre-yarn or the post-yarn context length?
-        # BC: we assume it is the pre-yarn context length.
-        else:
+        # Double-check: `factor` should be the ratio between the pre-yarn and post-yarn context lengths.
+        # NOTE: we might get `implicit_factor == 1` if config's `original_max_position_embeddings` was
+        # inferred from `max_position_embeddings` during standardization
+        original_max_position_embeddings = self.rope_parameters["original_max_position_embeddings"]
+        implicit_factor = self.max_position_embeddings / original_max_position_embeddings
+        if implicit_factor != factor and implicit_factor != 1:
             logger.warning_once(
-                "config.rope_parameters['original_max_position_embeddings'], the pre-yarn context length, is unset. We will "
-                "**assume** config.max_position_embeddings holds the pre-yarn context length. Some use cases may expect "
-                "config.max_position_embeddings to hold the post-yarn context length (pre-yarn context length * "
-                "factor) -- we recommend updating both fields for optimal downstream model usage."
+                f"The explicitly set RoPE scaling factor (config.rope_parameters['factor'] = {factor}) does not match "
+                "the ratio implicitly set by other parameters (implicit factor = "
+                "post-yarn context length / pre-yarn context length = "
+                "config.max_position_embeddings / config.rope_parameters['original_max_position_embeddings'] = "
+                f"{implicit_factor}). Using the explicit factor ({factor}) in YaRN. This may cause unexpected "
+                "behaviour in model usage, please correct the 'original_max_position_embeddings' fields in the model config."
             )
 
     def _validate_longrope_rope_parameters(self, rope_parameters: dict, ignore_keys: Optional[set] = None):
-        required_keys = {"rope_type", "short_factor", "long_factor", "rope_theta"}
-        # TODO (joao): update logic for the inclusion of `original_max_position_embeddings`
-        optional_keys = {"attention_factor", "factor", "original_max_position_embeddings"}
+        required_keys = {"rope_type", "short_factor", "long_factor", "rope_theta", "original_max_position_embeddings"}
+        optional_keys = {"attention_factor", "factor"}
         received_keys = set(rope_parameters.keys())
         rope_type = rope_parameters["rope_type"]
         self._check_received_keys(rope_type, received_keys, required_keys, optional_keys, ignore_keys=ignore_keys)
@@ -820,29 +824,28 @@ class RotaryEmbeddingConfigMixin:
                 f"`rope_parameters`'s long_factor field must have length {dim // 2}, got {len(long_factor)}"
             )
 
-        # Handle Phi3 divergence: prefer the use of `attention_factor` and/or `factor` over
-        # `original_max_position_embeddings` to compute internal variables. The latter lives outside `rope_parameters` and is
-        # unique to longrope (= undesirable)
-        if hasattr(self, "original_max_position_embeddings"):
+        factor = rope_parameters.get("factor")
+        original_max_position_embeddings = rope_parameters["original_max_position_embeddings"]
+
+        # Handle Phi3 divergence: we prefer the use of `attention_factor` and/or `factor` over
+        # `original_max_position_embeddings` to compute internal variables. The latter is undesirable
+        if factor is None and original_max_position_embeddings is not None:
             logger.warning_once(
-                "This model has set a `original_max_position_embeddings` field, to be used together with "
+                "This model config has set a `rope_parameters['original_max_position_embeddings']` field, to be used together with "
                 "`max_position_embeddings` to determine a scaling factor. Please set the `factor` field of `rope_parameters`"
                 "with this ratio instead -- we recommend the use of this field over `original_max_position_embeddings`, "
                 "as it is compatible with most model architectures."
             )
-        else:
-            factor = rope_parameters.get("factor")
-            if factor is None:
-                logger.warning("Missing required keys in `rope_parameters`: 'factor'")
-            elif not isinstance(factor, float) or factor < 1.0:
-                logger.warning(f"`rope_parameters`'s factor field must be a float >= 1, got {factor}")
-
-            attention_factor = rope_parameters.get("attention_factor")
-            if attention_factor is not None:
-                if not isinstance(attention_factor, float) or attention_factor < 0.0:
-                    logger.warning(
-                        f"`rope_parameters`'s attention_factor field must be a float greater than 0, got {attention_factor}"
-                    )
+        elif factor is None and original_max_position_embeddings is None:
+            logger.warning("Missing required keys in `rope_parameters`: 'factor'")
+        elif not isinstance(factor, float) or factor < 1.0:
+            logger.warning(f"`rope_parameters`'s factor field must be a float >= 1, got {factor}")
+
+        attention_factor = rope_parameters.get("attention_factor")
+        if attention_factor is not None and (not isinstance(attention_factor, float) or attention_factor < 0.0):
+            logger.warning(
+                f"`rope_parameters`'s attention_factor field must be a float greater than 0, got {attention_factor}"
+            )
 
     def _validate_llama3_rope_parameters(self, rope_parameters: dict, ignore_keys: Optional[set] = None):
         required_keys = {
@@ -899,6 +902,10 @@ class RotaryEmbeddingConfigMixin:
             received_keys -= {"type"}
             required_keys.add("rope_type")
 
+        optional_keys = optional_keys or set()
+        if "partial_rotary_factor" not in optional_keys:
+            optional_keys.add("partial_rotary_factor")
+
         # Some models need to store model-specific keys, and we don't want to throw warning at them
         if ignore_keys is not None:
             received_keys -= ignore_keys
@@ -907,9 +914,23 @@ class RotaryEmbeddingConfigMixin:
         if missing_keys:
             raise KeyError(f"Missing required keys in `rope_parameters` for 'rope_type'='{rope_type}': {missing_keys}")
 
-        if optional_keys is not None:
-            unused_keys = received_keys - required_keys - optional_keys
-        else:
-            unused_keys = received_keys - required_keys
+        unused_keys = received_keys - required_keys - optional_keys
         if unused_keys:
             logger.warning(f"Unrecognized keys in `rope_parameters` for 'rope_type'='{rope_type}': {unused_keys}")
+
+
+def rope_config_validation(config: RotaryEmbeddingConfigMixin, ignore_keys: Optional[set] = None):
+    """
+    This is a deprecated function.
+    It has been kept for backward compatibility with custom code models.
+    """
+    warnings.warn(
+        "`rope_config_validation` is deprecated and has been removed. "
+        "Its functionality has been moved to RotaryEmbeddingConfigMixin.validate_rope method. "
+        "PreTrainedConfig inherits this class, so please call self.validate_rope() instead. "
+        "Also, make sure to use the new rope_parameters syntax. "
+        "You can call self.standardize_rope_params() in the meantime.",
+        FutureWarning,
+    )
+    config.standardize_rope_params()
+    config.validate_rope(ignore_keys=ignore_keys)