config.json

{
    "activation_dropout": 0.0,
    "architectures": [
        "RodimusForCausalLM"
    ],
    "attention_dropout": 0.0,
    "attn_cfg": {
        "ffn_expand_ratio": 1.3333333333333333,
        "num_heads": 16,
        "num_heads_k": 1,
        "num_heads_v": 16,
        "rotary_emb_dim": -1,
        "rotary_emb_scale_base": null,
        "window_size": 4096
    },
    "bos_token_id": 126080,
    "d_model": 2048,
    "dropout": 0.0,
    "eos_token_id": 126081,
    "initializer_range": 0.02,
    "max_position_embeddings": 4096,
    "mixer_cfg": {
        "expand_ratio": 2.0,
        "input_gate_low_rank": "auto",
        "mem_size": 64,
        "mode": "fused_chunk",
        "normalize_epsilon": 1e-05
    },
    "model_type": "rodimus",
    "n_layer": 24,
    "norm_epsilon": 1e-05,
    "rescale_prenorm_residual": true,
    "residual_in_fp32": true,
    "tie_word_embeddings": true,
    "torch_dtype": "float32",
    "transformers_version": "4.42.0",
    "use_cache": true,
    "use_fast_path": true,
    "use_fused_cross_entropy": true,
    "use_norm_embedding": false,
    "use_scale_embedding": false,
    "vocab_size": 126464,
    "block_type": "rodimus_plus",
    "auto_map": {
        "AutoConfig": "configuration_rodimus.RodimusConfig",
        "AutoModel": "modeling_rodimus.RodimusForCausalLM",
        "AutoModelForCausalLM": "modeling_rodimus.RodimusForCausalLM"
    }
}