{
  "architectures": [
    "DSHybridForCausalLM"
  ],
  "attention_dropout": 0.0,
  "attention_logit_aux_loss_coef": 0.01,
  "attn_layer_offset": 2000,
  "attn_layer_period": 8,
  "attn_mode": "fused_recurrent",
  "bos_token_id": 1,
  "conv_size": 4,
  "eos_token_id": 2,
  "expand_ratio": 4,
  "expert_layer_offset": 1,
  "expert_layer_period": 2,
  "fuse_cross_entropy": true,
  "hidden_act": "swish",
  "hidden_ratio": 3,
  "hidden_size": 5120,
  "initializer_range": 0.02,
  "intermediate_size": null,
  "max_position_embeddings": 8096,
  "model_type": "ds_hybrid",
  "num_attention_heads": 32,
  "num_experts": 8,
  "num_experts_per_tok": 2,
  "num_heads": 4,
  "num_hidden_layers": 64,
  "num_key_value_heads": 8,
  "num_prediction_heads": 1,
  "output_attention_logits": false,
  "output_router_logits": false,
  "pad_token_id": 0,
  "rms_norm_eps": 1e-06,
  "router_aux_loss_coef": 0.001,
  "share_conv_kernel": true,
  "sliding_window": null,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.43.2",
  "use_cache": true,
  "use_lower_bound": false,
  "use_short_conv": false,
  "vocab_size": 128256
}