transformer-400m / config.json
mayank-mishra's picture
Upload folder using huggingface_hub
535621d verified
{
"bos_token_id": 100257,
"embedding_dropout": 0,
"eos_token_id": 100257,
"hidden_size": 1024,
"init_method": "normal",
"initializer_range": 0.02,
"layer_norm_epsilon": 1e-05,
"m_emb": null,
"m_residual": null,
"m_width": null,
"max_position_embeddings": 4096,
"mlp_blocks": [
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 2752,
"mlp_type": "MLP"
}
],
"model_type": "gpt_base",
"normalization_function": "rmsnorm",
"num_layers": 24,
"pad_token_id": 100256,
"position_embedding_type": "rope",
"rope_dim": 64,
"rope_scaling": null,
"rope_theta": 10000,
"router_aux_loss_coef": 0.001,
"sequence_mixer_blocks": [
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"add_bias": false,
"attention_multiplier": null,
"dropout": 0,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
}
],
"transformers_version": "4.57.3",
"use_cache": true,
"vocab_size": 100352
}