configuration_e5rope.py

# coding=utf-8
# This file has been modified from the configuration_roformer.py file in the transformers library.
# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" E5Rope model configuration"""

from collections import OrderedDict
from typing import Mapping

from transformers.configuration_utils import PretrainedConfig
from transformers.onnx import OnnxConfig
from transformers.utils import logging


logger = logging.get_logger(__name__)



class E5RopeConfig(PretrainedConfig):
    r"""

    Args:
        vocab_size (`int`, *optional*, defaults to 50000):
            Vocabulary size of the E5Rope model. Defines the number of different tokens that can be represented by
            the `inputs_ids` passed when calling [`E5RopeModel`] or [`TFE5RopeModel`].
        embedding_size (`int`, *optional*, defaults to None):
            Dimensionality of the encoder layers and the pooler layer. Defaults to the `hidden_size` if not provided.
        hidden_size (`int`, *optional*, defaults to 768):
            Dimension of the encoder layers and the pooler layer.
        num_hidden_layers (`int`, *optional*, defaults to 12):
            Number of hidden layers in the Transformer encoder.
        num_attention_heads (`int`, *optional*, defaults to 12):
            Number of attention heads for each attention layer in the Transformer encoder.
        intermediate_size (`int`, *optional*, defaults to 3072):
            Dimension of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
        hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):
            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
            `"relu"`, `"selu"` and `"gelu_new"` are supported.
        hidden_dropout_prob (`float`, *optional*, defaults to 0.1):
            The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
        attention_probs_dropout_prob (`float`, *optional*, defaults to 0.1):
            The dropout ratio for the attention probabilities.
        max_position_embeddings (`int`, *optional*, defaults to 1536):
            The maximum sequence length that this model might ever be used with. Typically set this to something large
            just in case (e.g., 512 or 1024 or 1536).
        type_vocab_size (`int`, *optional*, defaults to 2):
            The vocabulary size of the `token_type_ids` passed when calling [`E5RopeModel`] or [`TFE5RopeModel`].
        initializer_range (`float`, *optional*, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        layer_norm_eps (`float`, *optional*, defaults to 1e-12):
            The epsilon used by the layer normalization layers.
        is_decoder (`bool`, *optional*, defaults to `False`):
            Whether the model is used as a decoder or not. If `False`, the model is used as an encoder.
        use_cache (`bool`, *optional*, defaults to `True`):
            Whether or not the model should return the last key/values attentions (not used by all models). Only
            relevant if `config.is_decoder=True`.
        rotary_value (`bool`, *optional*, defaults to `False`):
            Whether or not apply rotary position embeddings on value layer.
        
        rope_theta (`float`, *optional*, defaults to 10000):
            Frequency base for RoPE.
        use_pose (`bool`, *optional*, defaults to `False`):
            Whether or not to use positional skip-wise training for long context. https://arxiv.org/abs/2309.10400
        pose_target_len (`int`, *optional*, defaults to None):
            target context length if use_pose is True
    
    """

    model_type = "e5rope"

    def __init__(
        self,
        vocab_size=50000,
        embedding_size=None,
        hidden_size=768,
        num_hidden_layers=12,
        num_attention_heads=12,
        intermediate_size=3072,
        hidden_act="gelu",
        hidden_dropout_prob=0.1,
        attention_probs_dropout_prob=0.1,
        max_position_embeddings=1536,
        type_vocab_size=2,
        initializer_range=0.02,
        layer_norm_eps=1e-12,
        pad_token_id=0,
        rotary_value=False,
        use_cache=True,
        rope_theta=10000,
        use_pose=False,
        pose_target_len=None,
        **kwargs,
    ):
        super().__init__(pad_token_id=pad_token_id, **kwargs)

        self.vocab_size = vocab_size
        self.embedding_size = hidden_size if embedding_size is None else embedding_size
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.hidden_act = hidden_act
        self.intermediate_size = intermediate_size
        self.hidden_dropout_prob = hidden_dropout_prob
        self.attention_probs_dropout_prob = attention_probs_dropout_prob
        self.max_position_embeddings = max_position_embeddings
        self.type_vocab_size = type_vocab_size
        self.initializer_range = initializer_range
        self.layer_norm_eps = layer_norm_eps
        self.rotary_value = rotary_value
        self.use_cache = use_cache
        self.rope_theta = rope_theta
        self.use_pose = use_pose
        self.pose_target_len = pose_target_len


class E5RopeOnnxConfig(OnnxConfig):
    @property
    def inputs(self) -> Mapping[str, Mapping[int, str]]:
        if self.task == "multiple-choice":
            dynamic_axis = {0: "batch", 1: "choice", 2: "sequence"}
        else:
            dynamic_axis = {0: "batch", 1: "sequence"}
        dynamic_axis = {0: "batch", 1: "sequence"}
        return OrderedDict(
            [
                ("input_ids", dynamic_axis),
                ("attention_mask", dynamic_axis),
                ("token_type_ids", dynamic_axis),
            ]
        )