rigel-base / config.json
mayank-mishra's picture
add code
4961452
Raw History Blame Contribute Delete
34 kB
Invalid JSON:Unexpected token 'I', ...", Infinity "... is not valid JSON
{
"architectures": [
"RigelForCausalLM"
],
"auto_map": {
"AutoConfig": "configuration_rigel.RigelConfig",
"AutoModel": "modeling_rigel.RigelModel",
"AutoModelForCausalLM": "modeling_rigel.RigelForCausalLM"
},
"bos_token_id": 100257,
"embedding_dropout": 0,
"embedding_init_method": "normal",
"eos_token_id": 100257,
"hidden_size": 1024,
"init_method": "mup",
"initializer_range": 0.1,
"layer_norm_epsilon": 1e-05,
"m_emb": 12,
"m_residual": 0.22,
"m_width": 4,
"max_position_embeddings": 4096,
"tie_word_embeddings": true,
"mlp_blocks": [
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
},
{
"activation_function": "swiglu",
"add_bias": false,
"dropout": 0,
"intermediate_size": 128,
"mlp_type": "MoE",
"normalized_topk": true,
"num_experts": 128,
"num_experts_per_tok": 2,
"shared_expert_gating": false,
"shared_intermediate_size": null
}
],
"model_type": "gpt_base",
"normalization_function": "rmsnorm",
"num_layers": 40,
"pad_token_id": 100256,
"position_embedding_type": "nope",
"rope_dim": null,
"rope_scaling": null,
"rope_theta": 10000,
"router_aux_loss_coef": 0.001,
"sequence_mixer_blocks": [
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"A_init_max": 16,
"A_init_min": 0,
"activation_function": "silu",
"add_bias": false,
"chunk_size": 256,
"conv_kernel_size": 4,
"dt_init_floor": 0.0001,
"dt_init_max": 0.1,
"dt_init_min": 0.001,
"intermediate_size": 2048,
"normalization_function": "rmsnorm",
"num_groups": 1,
"num_heads": 32,
"sequence_mixer_type": "mamba2",
"state_size": 128,
"time_step_limit": [
0,
Infinity
],
"use_conv_bias": true
},
{
"add_bias": false,
"attention_gate": false,
"attention_multiplier": null,
"attention_multiplier_method": "1 / head_dim",
"dropout": 0,
"exclusive_self_attention": true,
"head_dim": null,
"num_attention_heads": 16,
"num_key_value_heads": 4,
"sequence_mixer_type": "softmax_attention",
"sliding_window": null,
"softmax_dropout": 0
}
],
"transformers_version": "4.57.3",
"use_depth_scaled_init": true,
"vocab_size": 100352
}