configuration_erk.py

# Copyright 2026 eCloud Yazılım Teknolojileri. Based on Apache-2.0 licensed transformer architecture.
#
# 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.
"""Erk model configuration"""

from huggingface_hub.dataclasses import strict

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


@auto_docstring(checkpoint="ecloudtech/erk")
@strict
class ErkConfig(PreTrainedConfig):
    r"""
    ```python
    >>> from transformers import ErkModel, ErkConfig

    >>> # Initializing a Erk style configuration
    >>> configuration = ErkConfig()

    >>> # Initializing a model from the Erk-8B style configuration
    >>> model = ErkModel(configuration)

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

    model_type = "erk"
    keys_to_ignore_at_inference = ["past_key_values"]

    # Default tensor parallel plan for base model `Erk`
    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.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"]),
    }

    vocab_size: int = 151936
    hidden_size: int = 4096
    intermediate_size: int = 22016
    num_hidden_layers: int = 32
    num_attention_heads: int = 32
    num_key_value_heads: int | None = 32
    head_dim: int = 128
    hidden_act: str = "silu"
    max_position_embeddings: int = 32768
    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
    attention_bias: bool = False
    use_sliding_window: bool = False
    sliding_window: int | None = 4096
    max_window_layers: int = 28
    layer_types: list[str] | None = None
    attention_dropout: float | int = 0.0
    pad_token_id: int | None = None
    bos_token_id: int | None = None
    eos_token_id: int | list[int] | None = None

    def __post_init__(self, **kwargs):
        self.sliding_window = self.sliding_window if self.use_sliding_window else None
        if self.num_key_value_heads is None:
            self.num_key_value_heads = self.num_attention_heads

        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)
            ]
        super().__post_init__(**kwargs)


__all__ = ["ErkConfig"]