#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/kolibri1/modular_kolibri1.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_kolibri1.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# Copyright 2026 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 huggingface_hub.dataclasses import strict

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


@auto_docstring(checkpoint="Aleph-Alpha/Kolibri-1")
@strict
class Kolibri1Config(PreTrainedConfig):
    r"""
    ```python
    >>> from transformers import Kolibri1Model, Kolibri1Config

    >>> configuration = Kolibri1Config()
    >>> model = Kolibri1Model(configuration)
    >>> configuration = model.config
    ```
    """

    model_type = "kolibri1"
    keys_to_ignore_at_inference = ["past_key_values"]

    attribute_map = {
        "num_experts": "num_local_experts",
    }
    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.q_norm": "replicated_with_grad_allreduce",
        "layers.*.self_attn.k_norm": "replicated_with_grad_allreduce",
        "layers.*.self_attn.o_proj": "rowwise",
        "layers.*.mlp.experts.gate_up_proj": "packed_colwise",
        "layers.*.mlp.experts.down_proj": "rowwise",
        "layers.*.mlp.experts": "moe_tp_experts",
        "layers.*.mlp.shared_experts.gate_proj": "colwise",
        "layers.*.mlp.shared_experts.up_proj": "colwise",
        "layers.*.mlp.shared_experts.down_proj": "rowwise",
    }
    # Token dispatch by default, so `ep_size` can exceed `tp_size` (bounded by `num_key_value_heads`)
    base_model_ep_plan = {
        "layers.*.mlp.experts.gate_up_proj": "grouped_gemm",
        "layers.*.mlp.experts.down_proj": "grouped_gemm",
        "layers.*.mlp.experts": "ep_dispatch_experts",
    }
    base_model_pp_plan = {
        "embed_tokens": (["input_ids"], ["inputs_embeds"]),
        "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
        "norm": (["hidden_states"], ["hidden_states"]),
    }

    vocab_size: int = 128000
    hidden_size: int = 2560
    num_hidden_layers: int = 50
    num_attention_heads: int = 48
    num_key_value_heads: int = 4
    hidden_act: str = "silu"
    max_position_embeddings: int = 262144
    initializer_range: float = 0.02
    rms_norm_eps: float = 1e-6
    use_cache: bool = True
    tie_word_embeddings: bool = False
    rope_parameters: RopeParameters | dict | None = None
    sliding_window: int | None = 513
    attention_dropout: float | int = 0.0
    moe_intermediate_size: int = 512
    num_experts_per_tok: int = 6
    num_experts: int = 384
    norm_topk_prob: bool = False
    output_router_logits: bool = False
    pad_token_id: int | None = None
    bos_token_id: int | None = None
    eos_token_id: int | list[int] | None = None
    head_dim: int = 128
    shared_expert_intermediate_size: int = 512
    layer_types: list[str] | None = None

    def __post_init__(self, **kwargs):
        if self.layer_types is None:
            self.layer_types = [
                "full_attention" if (i + 1) % 5 == 0 else "sliding_attention" for i in range(self.num_hidden_layers)
            ]
        super().__post_init__(**kwargs)

    def validate_architecture(self):
        super().validate_architecture()
        if "sliding_attention" in self.layer_types and (self.sliding_window is None or self.sliding_window <= 0):
            raise ValueError(f"Sliding-attention layers need a positive `sliding_window`, got {self.sliding_window}.")


__all__ = ["Kolibri1Config"]