config.json

{
  "architectures": [
    "OmegaForCausalLM"
  ],
  "model_type": "omega_3060",
  "auto_map": {
    "AutoConfig": "configuration_omega.OmegaConfig",
    "AutoModelForCausalLM": "modeling_omega.OmegaForCausalLM"
  },
  "vocab_size": 163840,
  "hidden_size": 7168,
  "intermediate_size": 2048,
  "expert_intermediate": 2048,
  "num_hidden_layers": 61,
  "num_attention_heads": 64,
  "num_key_value_heads": 64,
  "num_experts_per_layer": 10,
  "kv_lora_rank": 512,
  "q_lora_rank": 1536,
  "qk_rope_head_dim": 64,
  "v_head_dim": 128,
  "qk_nope_head_dim": 128,
  "hidden_act": "silu",
  "max_position_embeddings": 262144,
  "rms_norm_eps": 1e-05,
  "rope_theta": 50000.0,
  "tau": 0.1,
  "expert_floor": 0.05,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.48.0",
  "use_cache": false
}