laguna-xs.2:nvfp4

231 19 hours ago

Laguna XS.2 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local machine.

tools thinking
... /
json
f42d36014db8 · 3.7kB
{
"architectures": [
"LagunaForCausalLM"
],
"auto_map": {
"AutoConfig": "configuration_laguna.LagunaConfig",
"AutoModelForCausalLM": "modeling_laguna.LagunaForCausalLM"
},
"model_type": "laguna",
"vocab_size": 100352,
"hidden_size": 2048,
"intermediate_size": 8192,
"num_hidden_layers": 40,
"num_attention_heads": 48,
"num_key_value_heads": 8,
"head_dim": 128,
"max_position_embeddings": 131072,
"attention_bias": false,
"attention_dropout": 0.0,
"rms_norm_eps": 1e-06,
"num_experts": 256,
"num_experts_per_tok": 8,
"moe_intermediate_size": 512,
"shared_expert_intermediate_size": 512,
"router_aux_loss_coef": 0.0,
"bos_token_id": 2,
"eos_token_id": [
2,
24
],
"pad_token_id": 9,
"tie_word_embeddings": false,
"use_cache": true,
"torch_dtype": "bfloat16",
"gating": true,
"sliding_window": 512,
"rope_parameters": {
"full_attention": {
"rope_theta": 500000.0,
"rope_type": "yarn",
"factor": 32.0,
"original_max_position_embeddings": 4096,
"beta_slow": 1.0,
"beta_fast": 64.0,
"attention_factor": 1.0,
"partial_rotary_factor": 0.5
},
"sliding_attention": {
"rope_type": "default",
"rope_theta": 10000.0,
"partial_rotary_factor": 1.0
}
},
"layer_types": [
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention"
],
"moe_apply_router_weight_on_input": false,
"partial_rotary_factor": 0.5,
"mlp_layer_types": [
"dense",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse",
"sparse"
],
"use_bidirectional_attention": false,
"moe_routed_scaling_factor": 2.5,
"num_attention_heads_per_layer": [
48,
64,
64,
64,
48,
64,
64,
64,
48,
64,
64,
64,
48,
64,
64,
64,
48,
64,
64,
64,
48,
64,
64,
64,
48,
64,
64,
64,
48,
64,
64,
64,
48,
64,
64,
64,
48,
64,
64,
64
],
"compression_config": {
"mode": null,
"group_size": 32,
"eps": 1e-05,
"filter_fqns": [
"output"
],
"recompute_fake_quantize": false
},
"quantization_config": {
"mode": null,
"group_size": 32,
"eps": 1e-05,
"filter_fqns": [
"output"
],
"recompute_fake_quantize": false
}
}