configuration_llama.py

# coding=utf-8
# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# 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.
""" LLaMA model configuration"""

from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging
from typing import Optional
import inspect
import copy
logger = logging.get_logger(__name__)

LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP = {}


class LlamaConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`LlamaModel`]. It is used to instantiate an LLaMA
    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 LLaMA-7B.

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


    Args:
        vocab_size (`int`, *optional*, defaults to 32000):
            Vocabulary size of the LLaMA model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling [`LlamaModel`]
        hidden_size (`int`, *optional*, defaults to 4096):
            Dimension of the hidden representations.
        intermediate_size (`int`, *optional*, defaults to 11008):
            Dimension of the MLP representations.
        num_hidden_layers (`int`, *optional*, defaults to 32):
            Number of hidden layers in the Transformer decoder.
        num_attention_heads (`int`, *optional*, defaults to 32):
            Number of attention heads for each attention layer in the Transformer decoder.
        num_key_value_heads (`int`, *optional*):
            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
            by meanpooling all the original heads within that group. For more details checkout [this
            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
            `num_attention_heads`.
        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
            The non-linear activation function (function or string) in the decoder.
        max_position_embeddings (`int`, *optional*, defaults to 2048):
            The maximum sequence length that this model might ever be used with. Llama 1 supports up to 2048 tokens,
            Llama 2 up to 4096, CodeLlama up to 16384.
        initializer_range (`float`, *optional*, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
            The epsilon used by the rms normalization layers.
        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`.
        pad_token_id (`int`, *optional*):
            Padding token id.
        bos_token_id (`int`, *optional*, defaults to 1):
            Beginning of stream token id.
        eos_token_id (`int`, *optional*, defaults to 2):
            End of stream token id.
        pretraining_tp (`int`, *optional*, defaults to 1):
            Experimental feature. Tensor parallelism rank used during pretraining. Please refer to [this
            document](https://huggingface.co/docs/transformers/parallelism) to understand more about it. This value is
            necessary to ensure exact reproducibility of the pretraining results. Please refer to [this
            issue](https://github.com/pytorch/pytorch/issues/76232).
        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
            Whether to tie weight embeddings
        rope_theta (`float`, *optional*, defaults to 10000.0):
            The base period of the RoPE embeddings.
        rope_scaling (`Dict`, *optional*):
            Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type
            and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value
            accordingly.
            Expected contents:
                `rope_type` (`str`):
                    The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope',
                    'llama3'], with 'default' being the original RoPE implementation.
                `factor` (`float`, *optional*):
                    Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In
                    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
                    pretraining.
                `attention_factor` (`float`, *optional*):
                    Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention
                    computation. If unspecified, it defaults to value recommended by the implementation, using the
                    `factor` field to infer the suggested value.
                `beta_fast` (`float`, *optional*):
                    Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear
                    ramp function. If unspecified, it defaults to 32.
                `beta_slow` (`float`, *optional*):
                    Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear
                    ramp function. If unspecified, it defaults to 1.
                `short_factor` (`List[float]`, *optional*):
                    Only used with 'longrope'. The scaling factor to be applied to short contexts (<
                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
                    size divided by the number of attention heads divided by 2
                `long_factor` (`List[float]`, *optional*):
                    Only used with 'longrope'. The scaling factor to be applied to long contexts (<
                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
                    size divided by the number of attention heads divided by 2
                `low_freq_factor` (`float`, *optional*):
                    Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE
                `high_freq_factor` (`float`, *optional*):
                    Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE
        attention_bias (`bool`, *optional*, defaults to `False`):
            Whether to use a bias in the query, key, value and output projection layers during self-attention.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            The dropout ratio for the attention probabilities.

    ```python
    >>> from transformers import LlamaModel, LlamaConfig

    >>> # Initializing a LLaMA llama-7b style configuration
    >>> configuration = LlamaConfig()

    >>> # Initializing a model from the llama-7b style configuration
    >>> model = LlamaModel(configuration)

    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```"""

    model_type = "llama"
    keys_to_ignore_at_inference = ["past_key_values"]

    def __init__(
        self,
        vocab_size=32000,
        hidden_size=4096,
        intermediate_size=11008,
        num_hidden_layers=32,
        num_attention_heads=32,
        num_key_value_heads=None,
        hidden_act="silu",
        max_position_embeddings=2048,
        initializer_range=0.02,
        rms_norm_eps=1e-6,
        use_cache=True,
        pad_token_id=None,
        bos_token_id=1,
        eos_token_id=2,
        pretraining_tp=1,
        tie_word_embeddings=False,
        rope_theta=10000.0,
        rope_scaling=None,
        my_rope_scaling={
    "factor": 8.0,
    "low_freq_factor": 1.0,
    "high_freq_factor": 4.0,
    "original_max_position_embeddings": 8192,
    "rope_type": "llama3"
  },
        attention_bias=False,
        attention_dropout=0.0,
        attn_implementation='flash_attention_2',
        **kwargs,
    ):  

        self.vocab_size = vocab_size
        self.max_position_embeddings = max_position_embeddings
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads

        self.attn_implementation = attn_implementation
        if self.attn_implementation is None:
            self.attn_implementation = "flash_attention_2"
            
        # for backward compatibility
        if num_key_value_heads is None:
            num_key_value_heads = num_attention_heads

        self.num_key_value_heads = num_key_value_heads
        self.hidden_act = hidden_act
        self.initializer_range = initializer_range
        self.rms_norm_eps = rms_norm_eps
        self.pretraining_tp = pretraining_tp
        self.use_cache = use_cache
        self.rope_theta = rope_theta
        self.rope_scaling = copy.deepcopy(my_rope_scaling)
        self.my_rope_scaling = my_rope_scaling
        #self._my_rope_scalingvalidation()
        self.attention_bias = attention_bias
        self.attention_dropout = attention_dropout
        if self.rope_scaling is not None and "type" not in self.rope_scaling:
            self.rope_scaling["type"] = self.rope_scaling["rope_type"]
        print('rope_scaling', self.my_rope_scaling, self.rope_scaling)

        super().__init__(
            pad_token_id=pad_token_id,
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )
        print('init done')


def _check_received_keys(
    rope_type: str,
    received_keys: set,
    required_keys: set,
    optional_keys: Optional[set] = None,
    ignore_keys: Optional[set] = None,
):
    """Compare the received keys in `config.rope_scaling` against the expected and optional keys"""
    # BC: "rope_type" was originally "type" -- let's check for "rope_type" when "type" is present
    if "type" in received_keys:
        received_keys -= {"type"}
        required_keys.add("rope_type")

    # 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

    missing_keys = required_keys - received_keys
    if missing_keys:
        raise KeyError(f"Missing required keys in `rope_scaling` 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
    if unused_keys:
        logger.warning(f"Unrecognized keys in `rope_scaling` for 'rope_type'='{rope_type}': {unused_keys}")


def _validate_llama3_parameters(config: PretrainedConfig, ignore_keys: Optional[set] = None):
    rope_scaling = config.rope_scaling
    # from IPython import embed; embed()
    print('rope_scaling2', rope_scaling)
    rope_type = rope_scaling.get("rope_type", rope_scaling.get("type", None))  # BC: "rope_type" was originally "type"
    required_keys = {"rope_type", "factor", "original_max_position_embeddings", "low_freq_factor", "high_freq_factor"}
    received_keys = set(rope_scaling.keys())
    _check_received_keys(rope_type, received_keys, required_keys, ignore_keys=ignore_keys)

    factor = rope_scaling["factor"]
    if factor is None or not isinstance(factor, float) or factor < 1.0:
        logger.warning(f"`rope_scaling`'s factor field must be a float >= 1, got {factor}")

    low_freq_factor = rope_scaling["low_freq_factor"]
    high_freq_factor = rope_scaling["high_freq_factor"]
    if low_freq_factor is None or not isinstance(low_freq_factor, float):
        logger.warning(f"`rope_scaling`'s low_freq_factor field must be a float, got {low_freq_factor}")
    if high_freq_factor is None or not isinstance(high_freq_factor, float):
        logger.warning(f"`rope_scaling`'s high_freq_factor field must be a float, got {high_freq_factor}")
    if high_freq_factor <= low_freq_factor:
        logger.warning(
            "`rope_scaling`'s high_freq_factor field must be greater than low_freq_factor, got high_freq_factor="
            f"{high_freq_factor} and low_freq_factor={low_freq_factor}"
        )

    original_max_position_embeddings = rope_scaling["original_max_position_embeddings"]
    if original_max_position_embeddings is None or not isinstance(original_max_position_embeddings, int):
        logger.warning(
            "`rope_scaling`'s original_max_position_embeddings field must be an integer, got "
            f"{original_max_position_embeddings}"
        )
    if original_max_position_embeddings >= config.max_position_embeddings:
        logger.warning(
            "`rope_scaling`'s original_max_position_embeddings field must be less than max_position_embeddings, got "
            f"{original_max_position_embeddings} and max_position_embeddings={config.max_position_embeddings}"
        )