m4/models/vt5/configuration_vt5.py

# coding=utf-8
# Copyright 2020, The T5 Authors and HuggingFace Inc.
#
# 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.
""" VT5 model configuration"""
import os
from typing import Tuple, Union

from transformers import AutoConfig
from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging


logger = logging.get_logger(__name__)

T5_PRETRAINED_CONFIG_ARCHIVE_MAP = {
    "t5-small": "https://huggingface.co/t5-small/resolve/main/config.json",
    "t5-base": "https://huggingface.co/t5-base/resolve/main/config.json",
    "t5-large": "https://huggingface.co/t5-large/resolve/main/config.json",
    "t5-3b": "https://huggingface.co/t5-3b/resolve/main/config.json",
    "t5-11b": "https://huggingface.co/t5-11b/resolve/main/config.json",
}


class VT5Config(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`T5Model`] or a [`TFT5Model`]. It is used to
    instantiate a T5 model according to the specified arguments, defining the model architecture. Instantiating a
    configuration with the defaults will yield a similar configuration to that of the T5
    [t5-small](https://huggingface.co/t5-small) architecture.

    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
    documentation from [`PretrainedConfig`] for more information.

    TODO: this doc is completely out of sync with the actual args

    Arguments:
        vocab_size (`int`, *optional*, defaults to 32128):
            Vocabulary size of the T5 model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling [`T5Model`] or [`TFT5Model`].
        d_model (`int`, *optional*, defaults to 512):
            Size of the encoder layers and the pooler layer.
        d_kv (`int`, *optional*, defaults to 64):
            Size of the key, query, value projections per attention head. `d_kv` has to be equal to `d_model //
            num_heads`.
        d_ff (`int`, *optional*, defaults to 2048):
            Size of the intermediate feed forward layer in each `T5Block`.
        num_layers (`int`, *optional*, defaults to 6):
            Number of hidden layers in the Transformer encoder.
        num_decoder_layers (`int`, *optional*):
            Number of hidden layers in the Transformer decoder. Will use the same value as `num_layers` if not set.
        num_heads (`int`, *optional*, defaults to 8):
            Number of attention heads for each attention layer in the Transformer encoder.
        relative_attention_num_buckets (`int`, *optional*, defaults to 32):
            The number of buckets to use for each attention layer.
        relative_attention_max_distance (`int`, *optional*, defaults to 128):
            The maximum distance of the longer sequences for the bucket separation.
        dropout_rate (`float`, *optional*, defaults to 0.1):
            The ratio for all dropout layers.
        layer_norm_eps (`float`, *optional*, defaults to 1e-6):
            The epsilon used by the layer normalization layers.
        initializer_factor (`float`, *optional*, defaults to 1):
            A factor for initializing all weight matrices (should be kept to 1, used internally for initialization
            testing).
        feed_forward_proj (`string`, *optional*, defaults to `"relu"`):
            Type of feed forward layer to be used. Should be one of `"relu"` or `"gated-gelu"`. T5v1.1 uses the
            `"gated-gelu"` feed forward projection. Original T5 uses `"relu"`.
        use_cache (`bool`, *optional*, defaults to `True`):
            Whether or not the model should return the last key/values attentions (not used by all models).
        additional_vocab_size (`int`, *optional`, defaults to 0):
            Additional vocabulary size of the model, typically for the special "<img>" token. Additional vocab tokens
            are always trainable whereas regular vocab tokens can be frozen or not.
        alpha_initializer (`str`, *optional*, defaults to `"ones"`):
            Initialization type for the alphas.
        alphas_initializer_range (`float`, *optional*, defaults to 0.0):
            The standard deviation of the truncated_normal_initializer for initializing the alphas in the Gated Cross Attention.
        alpha_type (`str`, *optional*, defaults to `"vector"`):
            Whether the gating alphas should be vectors or single floats.
    """
    model_type = "vt5"
    keys_to_ignore_at_inference = ["past_key_values"]
    attribute_map = {"hidden_size": "d_model", "num_attention_heads": "num_heads", "num_hidden_layers": "num_layers"}

    def __init__(
        self,
        vocab_size=32128,
        d_model=512,
        d_kv=64,
        d_ff=2048,
        num_layers=6,
        num_decoder_layers=None,
        num_heads=8,
        relative_attention_num_buckets=32,
        relative_attention_max_distance=128,
        dropout_rate=0.1,
        layer_norm_epsilon=1e-6,
        initializer_factor=1.0,
        feed_forward_proj="relu",
        is_encoder_decoder=True,
        use_cache=True,
        pad_token_id=0,
        eos_token_id=1,
        additional_vocab_size=0,
        alpha_initializer="ones",
        alphas_initializer_range=0.0,
        alpha_type="vector",
        cross_layer_interval=1,
        tie_word_embeddings=False,
        freeze_text_layers=True,
        freeze_lm_head=False,
        freeze_vision_layers=True,
        vision_model_name="google/vit-base-patch16-224",
        vision_model_params="{}",
        vision_embed_dim=768,
        image_token_index=32128,
        **kwargs,
    ):
        self.vocab_size = vocab_size
        self.additional_vocab_size = additional_vocab_size
        self.d_model = d_model
        self.d_kv = d_kv
        self.d_ff = d_ff
        self.num_layers = num_layers
        self.num_decoder_layers = (
            num_decoder_layers if num_decoder_layers is not None else self.num_layers
        )  # default = symmetry
        self.num_heads = num_heads
        self.relative_attention_num_buckets = relative_attention_num_buckets
        self.relative_attention_max_distance = relative_attention_max_distance
        self.dropout_rate = dropout_rate
        self.layer_norm_epsilon = layer_norm_epsilon
        self.initializer_factor = initializer_factor
        self.feed_forward_proj = feed_forward_proj
        self.use_cache = use_cache

        act_info = self.feed_forward_proj.split("-")
        self.dense_act_fn = act_info[-1]
        self.is_gated_act = act_info[0] == "gated"

        if len(act_info) > 1 and act_info[0] != "gated" or len(act_info) > 2:
            raise ValueError(
                f"`feed_forward_proj`: {feed_forward_proj} is not a valid activation function of the dense layer."
                "Please make sure `feed_forward_proj` is of the format `gated-{ACT_FN}` or `{ACT_FN}`, e.g. "
                "'gated-gelu' or 'relu'"
            )

        # for backwards compatibility
        if feed_forward_proj == "gated-gelu":
            self.dense_act_fn = "gelu_new"

        self.alpha_initializer = alpha_initializer
        self.alphas_initializer_range = alphas_initializer_range
        self.alpha_type = alpha_type

        self.cross_layer_interval = cross_layer_interval
        self.freeze_vision_layers = freeze_vision_layers
        self.vision_model_name = vision_model_name
        self.vision_model_params = vision_model_params

        self.tie_word_embeddings = tie_word_embeddings
        self.freeze_text_layers = freeze_text_layers
        self.freeze_lm_head = freeze_lm_head
        self.image_token_index = image_token_index

        self.vision_embed_dim = vision_embed_dim

        # IMPORTANT: Do not do any __init__ args-based checks in the constructor, since
        # PretrainedConfig.from_dict first instantiates the class with the config dict and only then
        # updates the config object with `kwargs` from from_pretrained, so during the instantiation
        # of this object many attributes have default values and haven't yet been overridden.
        # Do any required checks inside `from_pretrained` once the superclass' `from_pretrained` was run.

        super().__init__(
            pad_token_id=pad_token_id,
            eos_token_id=eos_token_id,
            is_encoder_decoder=is_encoder_decoder,
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )

    def check_compatibilities(self):
        if self.tie_word_embeddings and (self.freeze_text_layers != self.freeze_lm_head):
            raise ValueError(
                "if `tie_word_embeddings` is True, then `freeze_lm_head` and `freeze_text_layers` must be equal."
            )

        vision_model_params = eval(self.vision_model_params)
        config = AutoConfig.from_pretrained(self.vision_model_name, **vision_model_params)
        if hasattr(config, "vision_config"):
            vison_config = config.vision_config
        else:
            vison_config = config
        vision_embed_dim = vison_config.hidden_size
        if self.vision_embed_dim != vision_embed_dim:
            raise ValueError(
                f"vision_embed_dim ({self.vision_embed_dim}) must match the hidden size of the vision model"
                f" ({vision_embed_dim})"
            )

    @classmethod
    def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> "PretrainedConfig":
        outputs = super(VT5Config, cls).from_pretrained(pretrained_model_name_or_path, **kwargs)
        if isinstance(outputs, Tuple):
            # When called with return_unused_kwargs=True, the first item will be the config
            outputs[0].check_compatibilities()
        else:
            outputs.check_compatibilities()
        return outputs