#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/zaya/modular_zaya.py.
#               Do NOT edit this file manually as any edits will be overwritten by the generation of
#             the file from the modular. If any change should be done, please apply the change to the
#                          modular_zaya.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# Copyright 2026 Zyphra 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.

from typing import Any, Literal

from huggingface_hub.dataclasses import strict

from ...configuration_utils import PreTrainedConfig
from ...modeling_rope_utils import RopeParameters
from ...utils import auto_docstring


@auto_docstring(checkpoint="Zyphra/ZAYA1-8B")
@strict
class ZayaConfig(PreTrainedConfig):
    r"""
    lm_head_bias (`bool`, *optional*, defaults to `False`):
        Whether to add a bias to the language modeling head.
    router_hidden_size (`int`, *optional*, defaults to 256):
        Hidden size used by the ZAYA router.
    cca_time0 (`int`, *optional*, defaults to 2):
        First temporal parameter of the CCA projection.
    cca_time1 (`int`, *optional*, defaults to 2):
        Second temporal parameter of the CCA projection.

    ```python
    >>> from transformers import ZayaConfig, ZayaModel

    >>> configuration = ZayaConfig()
    >>> model = ZayaModel(configuration)

    >>> configuration = model.config
    ```
    """

    model_type = "zaya"
    keys_to_ignore_at_inference = ["past_key_values"]
    base_model_ep_plan = {
        "layers.*.mlp.gate": "ep_router",
        "layers.*.mlp.experts.gate_up_proj": "grouped_gemm",
        "layers.*.mlp.experts.down_proj": "grouped_gemm",
        "layers.*.mlp.experts": "moe_tp_experts",
    }

    vocab_size: int = 262272
    hidden_size: int = 2048
    num_hidden_layers: int = 40
    num_attention_heads: int = 8
    num_key_value_heads: int = 2
    hidden_act: str = "silu"
    max_position_embeddings: int = 131072
    initializer_range: float = 0.02
    rms_norm_eps: float = 1e-5
    use_cache: bool = True
    tie_word_embeddings: bool = True
    rope_parameters: RopeParameters | dict | None = None
    sliding_window: int | None = None
    attention_dropout: float | int = 0.0
    moe_intermediate_size: int = 2048

    num_experts_per_tok: int = 1
    num_experts: int = 16
    output_router_logits: bool = False
    layer_types: list[str] | None = None
    pad_token_id: int | None = 0
    bos_token_id: int | None = 2
    eos_token_id: int | list[int] | None = 106

    # Zaya-specific attention
    head_dim: int = 128
    attention_bias: bool = False

    lm_head_bias: bool = False
    router_hidden_size: int = 256
    cca_time0: int = 2
    cca_time1: int = 2

    def __post_init__(self, **kwargs):
        self.layer_types = ["hybrid"] * self.num_hidden_layers if self.layer_types is None else list(self.layer_types)

        default_rope_params: dict[Literal["hybrid", "hybrid_sliding"], dict[str, Any]] = {
            "hybrid": {
                "rope_type": "default",
                "rope_theta": 5_000_000.0,
                "partial_rotary_factor": 0.5,
            },
            "hybrid_sliding": {
                "rope_type": "default",
                "rope_theta": 10_000.0,
                "partial_rotary_factor": 0.5,
            },
        }
        if self.rope_parameters is None:
            self.rope_parameters = default_rope_params

        super().__post_init__(**kwargs, ignore_keys_at_rope_validation={"hybrid", "hybrid_sliding"})

    def convert_rope_params_to_dict(self, **kwargs):
        # No legacy flat RoPE format is supported here; conversion writes the nested ZAYA layer-type format directly.
        return kwargs

    def validate_architecture(self):
        """Part of ``@strict``-powered validation."""
        if self.num_experts_per_tok != 1:
            raise ValueError("ZAYA currently supports `num_experts_per_tok=1` only.")
        if self.num_attention_heads % self.num_key_value_heads != 0:
            raise ValueError("`num_attention_heads` must be a multiple of `num_key_value_heads`.")
        if "hybrid_sliding" in self.layer_types and self.sliding_window is None:
            raise ValueError("`sliding_window` must be set when `layer_types` contains `hybrid_sliding`.")


__all__ = ["ZayaConfig"]