configuration_xqwen.py

# coding=utf-8
# Copyright 2024 The Qwen team, Alibaba Group and 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.
"""xQwen model configuration"""

from transformers.configuration_utils import PretrainedConfig
from transformers.modeling_rope_utils import rope_config_validation
from transformers.utils import logging

from xlstm.xlstm_large.model import mLSTMBackendConfig


logger = logging.get_logger(__name__)


class xQwenConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`xQwenModel`]. It is used to instantiate a
    xQwen model according to the specified arguments, defining the model architecture. Instantiating a configuration
    with the defaults will yield a similar configuration to that of
    xQwen-8B [Qwen/xQwen-8B](https://huggingface.co/Qwen/xQwen-8B).

    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 151936):
            Vocabulary size of the xQwen model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling [`xQwenModel`]
        hidden_size (`int`, *optional*, defaults to 4096):
            Dimension of the hidden representations.
        intermediate_size (`int`, *optional*, defaults to 22016):
            Dimension of the MLP representations.
        num_hidden_layers (`int`, *optional*, defaults to 32):
            Number of hidden layers in the Transformer encoder.
        num_attention_heads (`int`, *optional*, defaults to 32):
            Number of attention heads for each attention layer in the Transformer encoder.
        num_key_value_heads (`int`, *optional*, defaults to 32):
            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 `32`.
        head_dim (`int`, *optional*, defaults to 128):
            The attention head dimension.
        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 32768):
            The maximum sequence length that this model might ever be used with.
        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`.
        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
            Whether the model's input and output word embeddings should be tied.
        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`, defaults to `False`, *optional*, defaults to `False`):
            Whether to use a bias in the query, key, value and output projection layers during self-attention.
        use_sliding_window (`bool`, *optional*, defaults to `False`):
            Whether to use sliding window attention.
        sliding_window (`int`, *optional*, defaults to 4096):
            Sliding window attention (SWA) window size. If not specified, will default to `4096`.
        max_window_layers (`int`, *optional*, defaults to 28):
            The number of layers that use SWA (Sliding Window Attention). The bottom layers use SWA while the top use full attention.
        layer_types (`list`, *optional*):
            Attention pattern for each layer.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            The dropout ratio for the attention probabilities.

    ```python
    >>> from transformers import xQwenModel, xQwenConfig

    >>> # Initializing a xQwen style configuration
    >>> configuration = xQwenConfig()

    >>> # Initializing a model from the xQwen-8B style configuration
    >>> model = xQwenModel(configuration)

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

    model_type = "xqwen"
    keys_to_ignore_at_inference = ["past_key_values"]

    # Default tensor parallel plan for base model `xQwen`
    base_model_tp_plan = {
        "layers.*.self_attn.q_proj": "colwise",
        "layers.*.self_attn.k_proj": "colwise",
        "layers.*.self_attn.v_proj": "colwise",
        "layers.*.self_attn.o_proj": "rowwise",
        "layers.*.mlp.gate_proj": "colwise",
        "layers.*.mlp.up_proj": "colwise",
        "layers.*.mlp.down_proj": "rowwise",
    }
    base_model_pp_plan = {
        "embed_tokens": (["input_ids"], ["inputs_embeds"]),
        "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
        "norm": (["hidden_states"], ["hidden_states"]),
    }

    def __init__(
        self,
        vocab_size=151936,
        hidden_size=4096,
        intermediate_size=22016,
        num_hidden_layers=32,
        num_attention_heads=32,
        num_key_value_heads=32,
        head_dim=128,
        hidden_act="silu",
        max_position_embeddings=32768,
        initializer_range=0.02,
        rms_norm_eps=1e-6,
        use_cache=True,
        tie_word_embeddings=False,
        rope_theta=10000.0,
        rope_scaling=None,
        attention_bias=False,
        # SWA
        use_sliding_window=False,
        sliding_window=64,
        max_window_layers=28,
        swa_with_memory=False,
        sliding_window_memory=4,
        swa_modulation="static",
        swa_modulation_bounded=False,

        # Hybdrid
        nth_layer_sa=0,
        layer_types=None,

        # Conv
        use_short_conv=False,
        conv_size=4,

        # Hedgehog
        use_hedgehog=False,

        attention_dropout=0.0,
        residual_dropout=0.0,
        mlp_dropout=0.0,
        # xLSTM specific
        qk_dim_factor=1,
        v_dim_factor=1,
        use_bias=False,
        norm_reduction_force_float32=True,
        gate_soft_cap=15.0,
        weight_mode="single",
        base_attention="xlstm_attention",
        # mLSTM specific
        max_inference_chunksize=16384,
        chunkwise_kernel="chunkwise--triton_xl_chunk",
        sequence_kernel="native_sequence__triton",
        step_kernel="triton",
        mode="train",
        chunk_size=64,
        return_last_states=True,
        autocast_kernel_dtype="bfloat16",
        eps=1e-6,
        inference_state_dtype="float32",
        **kwargs,
    ):
        self.vocab_size = vocab_size
        self.max_position_embeddings = max_position_embeddings
        self.hidden_size = hidden_size
        self.embedding_dim = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.use_sliding_window = use_sliding_window
        self.sliding_window = sliding_window if self.use_sliding_window else None
        self.max_window_layers = max_window_layers
        self.swa_with_memory = swa_with_memory
        self.sliding_window_memory = sliding_window_memory
        self.swa_modulation = swa_modulation
        self.swa_modulation_bounded = swa_modulation_bounded

        self.use_short_conv = use_short_conv
        self.conv_size = conv_size
        self.use_hedgehog = use_hedgehog



        # 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.head_dim = head_dim
        self.hidden_act = hidden_act
        self.initializer_range = initializer_range
        self.rms_norm_eps = rms_norm_eps
        self.norm_eps = rms_norm_eps
        self.use_cache = use_cache
        self.rope_theta = rope_theta
        self.rope_scaling = rope_scaling
        self.attention_bias = attention_bias
        self.attention_dropout = attention_dropout
        self.residual_dropout = residual_dropout
        self.mlp_dropout = mlp_dropout
        # Validate the correctness of rotary position embeddings parameters
        # BC: if there is a 'type' field, move it to 'rope_type'.
        if self.rope_scaling is not None and "type" in self.rope_scaling:
            self.rope_scaling["rope_type"] = self.rope_scaling["type"]
        rope_config_validation(self)
        
        self.base_attention = base_attention
        self.layer_types = layer_types
        # if self.layer_types is None:
        #     self.layer_types = [
        #         "sliding_attention"
        #         if self.sliding_window is not None and i >= self.max_window_layers
        #         else "full_attention"
        #         for i in range(self.num_hidden_layers)
        #     ]
        self.nth_layer_sa = nth_layer_sa
        if self.layer_types is None:
            self.layer_types = []
            assert nth_layer_sa >= 0, "nth_layer_sa must be non-negative"

            for i in range(self.num_hidden_layers):
                if nth_layer_sa == 0:
                    self.layer_types.append(base_attention)
                    continue

                if i % nth_layer_sa == 0:
                    self.layer_types.append("qwen_attention")
                else:
                    self.layer_types.append(base_attention)

        # layer_type_validation(self.layer_types)
        self.num_heads = self.num_attention_heads
        self.qk_dim_factor = qk_dim_factor
        self.v_dim_factor = v_dim_factor
        self.use_bias = use_bias
        self.norm_reduction_force_float32 = norm_reduction_force_float32
        self.gate_soft_cap = gate_soft_cap
        self.weight_mode = weight_mode
        self.chunkwise_kernel = chunkwise_kernel
        self.sequence_kernel = sequence_kernel
        self.step_kernel = step_kernel
        self.mode = mode
        self.chunk_size = chunk_size
        self.return_last_states = return_last_states
        self.autocast_kernel_dtype = autocast_kernel_dtype
        self.eps = eps
        self.inference_state_dtype = inference_state_dtype
        self.max_inference_chunksize = max_inference_chunksize


        super().__init__(
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )

    def mlstm_backend_config(self):
        return mLSTMBackendConfig(
            chunkwise_kernel=self.chunkwise_kernel,
            sequence_kernel=self.sequence_kernel,
            step_kernel=self.step_kernel,
            mode=self.mode,
            chunk_size=self.chunk_size,
            return_last_states=self.return_last_states,
            autocast_kernel_dtype=self.autocast_kernel_dtype,
            eps=self.eps,
            inference_state_dtype=self.inference_state_dtype,
        )


__all__ = ["xQwenConfig"]