configuration_gmma_jepa.py

"""
Configuration for Gmma-JEPA model by Danger Labs.
"""
from transformers.configuration_utils import PretrainedConfig

class GmmaJEPAConfig(PretrainedConfig):
    model_type = "gmma-jepa"
    keys_to_ignore_at_inference = ["past_key_values"]

    def __init__(
        self,
        vocab_size=256000,
        hidden_size=1536,
        intermediate_size=4096,
        num_hidden_layers=18,
        num_attention_heads=8,
        num_key_value_heads=1,
        head_dim=192,
        hidden_act="gelu_pytorch_tanh",
        max_position_embeddings=8192,
        initializer_range=0.02,
        rms_norm_eps=1e-6,
        use_cache=True,
        pad_token_id=0,
        bos_token_id=2,
        eos_token_id=1,
        tie_word_embeddings=True,
        rope_theta=10000.0,
        num_jepa_layers=8,
        dag_reasoning_enabled=True,
        universal_compiler_safety=True,
        num_swarm_specialists=23,
        **kwargs,
    ):
        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,
        )
        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.head_dim = head_dim
        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.use_cache = use_cache
        self.rope_theta = rope_theta
        self.num_jepa_layers = num_jepa_layers
        self.dag_reasoning_enabled = dag_reasoning_enabled
        self.universal_compiler_safety = universal_compiler_safety
        self.num_swarm_specialists = num_swarm_specialists