| { | |
| "bos_token_id": 0, | |
| "embedding_dropout": 0, | |
| "eos_token_id": 0, | |
| "hidden_size": 1536, | |
| "init_method": "mup", | |
| "initializer_range": 0.1, | |
| "layer_norm_epsilon": 1e-05, | |
| "m_emb": 12, | |
| "m_residual": 0.22, | |
| "m_width": 6, | |
| "max_position_embeddings": 4096, | |
| "mlp_blocks": [ | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| }, | |
| { | |
| "activation_function": "swiglu", | |
| "add_bias": false, | |
| "dropout": 0, | |
| "intermediate_size": 4096, | |
| "mlp_type": "MLP" | |
| } | |
| ], | |
| "model_type": "gpt_dolomite", | |
| "normalization_function": "rmsnorm", | |
| "num_attention_heads": 12, | |
| "num_layers": 40, | |
| "pad_token_id": 0, | |
| "position_embedding_type": "nope", | |
| "rope_scaling": null, | |
| "rope_theta": 10000, | |
| "router_aux_loss_coef": 0.01, | |
| "sequence_mixer_blocks": [ | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| }, | |
| { | |
| "add_bias": false, | |
| "attention_head_type": "gqa", | |
| "attention_multiplier": 0.0078125, | |
| "dropout": 0, | |
| "num_key_value_heads": 4, | |
| "sequence_mixer_type": "stickbreaking_attention" | |
| } | |
| ], | |
| "transformers_version": "4.49.0.dev0", | |
| "upcast_logits_for_loss": false, | |
| "use_cache": true, | |
| "vocab_size": 50304 | |
| } | |