Hugging Face's logo

mconcat
/
Trinity-Large-TrueBase-NVFP4

Safetensors
tensorrt_llm
afmoe
nvidia
nvfp4
fp4
quantized
tensorrt-llm
modelopt
mixture-of-experts
Mixture of Experts
blackwell
custom_code
8-bit precision
Model card Files
xet
Community
Trinity-Large-TrueBase-NVFP4 / config.json
mconcat's picture
mconcat
Upload folder using huggingface_hub
13be8a1 verified
raw
history blame contribute delete
8.8 kB
 {
"architectures": [
"AfmoeForCausalLM"
],
"attention_bias": false,
"attention_dropout": 0.0,
"auto_map": {
"AutoConfig": "configuration_afmoe.AfmoeConfig",
"AutoModel": "modeling_afmoe.AfmoeModel",
"AutoModelForCausalLM": "modeling_afmoe.AfmoeForCausalLM"
},
"bos_token_id": null,
"dtype": "bfloat16",
"eos_token_id": null,
"global_attn_every_n_layers": 4,
"head_dim": 128,
"hidden_act": "silu",
"hidden_size": 3072,
"initializer_range": 0.02,
"intermediate_size": 12288,
"layer_types": [
"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",
"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"
],
"load_balance_coeff": 5e-05,
"max_position_embeddings": 8192,
"model_type": "afmoe",
"moe_intermediate_size": 3072,
"mup_enabled": true,
"n_group": 1,
"num_attention_heads": 48,
"num_dense_layers": 6,
"num_expert_groups": 1,
"num_experts": 256,
"num_experts_per_tok": 4,
"num_hidden_layers": 60,
"num_key_value_heads": 8,
"num_limited_groups": 1,
"num_shared_experts": 1,
"pad_token_id": null,
"rms_norm_eps": 1e-05,
"rope_parameters": {
"rope_theta": 10000.0,
"rope_type": "default"
},
"rope_theta": 10000,
"route_norm": true,
"route_scale": 2.448,
"score_func": "sigmoid",
"sliding_window": 4096,
"tie_word_embeddings": false,
"topk_group": 1,
"transformers_version": "5.1.0",
"use_cache": true,
"use_grouped_mm": true,
"vocab_size": 200192,
"quantization_config": {
"config_groups": {
"group_0": {
"input_activations": {
"dynamic": false,
"num_bits": 4,
"type": "float",
"group_size": 16
},
"weights": {
"dynamic": false,
"num_bits": 4,
"type": "float",
"group_size": 16
},
"targets": [
"Linear"
]
}
},
"ignore": [
"lm_head",
"model.layers.0.self_attn*",
"model.layers.1.self_attn*",
"model.layers.10.mlp.router*",
"model.layers.10.self_attn*",
"model.layers.11.mlp.router*",
"model.layers.11.self_attn*",
"model.layers.12.mlp.router*",
"model.layers.12.self_attn*",
"model.layers.13.mlp.router*",
"model.layers.13.self_attn*",
"model.layers.14.mlp.router*",
"model.layers.14.self_attn*",
"model.layers.15.mlp.router*",
"model.layers.15.self_attn*",
"model.layers.16.mlp.router*",
"model.layers.16.self_attn*",
"model.layers.17.mlp.router*",
"model.layers.17.self_attn*",
"model.layers.18.mlp.router*",
"model.layers.18.self_attn*",
"model.layers.19.mlp.router*",
"model.layers.19.self_attn*",
"model.layers.2.self_attn*",
"model.layers.20.mlp.router*",
"model.layers.20.self_attn*",
"model.layers.21.mlp.router*",
"model.layers.21.self_attn*",
"model.layers.22.mlp.router*",
"model.layers.22.self_attn*",
"model.layers.23.mlp.router*",
"model.layers.23.self_attn*",
"model.layers.24.mlp.router*",
"model.layers.24.self_attn*",
"model.layers.25.mlp.router*",
"model.layers.25.self_attn*",
"model.layers.26.mlp.router*",
"model.layers.26.self_attn*",
"model.layers.27.mlp.router*",
"model.layers.27.self_attn*",
"model.layers.28.mlp.router*",
"model.layers.28.self_attn*",
"model.layers.29.mlp.router*",
"model.layers.29.self_attn*",
"model.layers.3.self_attn*",
"model.layers.30.mlp.router*",
"model.layers.30.self_attn*",
"model.layers.31.mlp.router*",
"model.layers.31.self_attn*",
"model.layers.32.mlp.router*",
"model.layers.32.self_attn*",
"model.layers.33.mlp.router*",
"model.layers.33.self_attn*",
"model.layers.34.mlp.router*",
"model.layers.34.self_attn*",
"model.layers.35.mlp.router*",
"model.layers.35.self_attn*",
"model.layers.36.mlp.router*",
"model.layers.36.self_attn*",
"model.layers.37.mlp.router*",
"model.layers.37.self_attn*",
"model.layers.38.mlp.router*",
"model.layers.38.self_attn*",
"model.layers.39.mlp.router*",
"model.layers.39.self_attn*",
"model.layers.4.self_attn*",
"model.layers.40.mlp.router*",
"model.layers.40.self_attn*",
"model.layers.41.mlp.router*",
"model.layers.41.self_attn*",
"model.layers.42.mlp.router*",
"model.layers.42.self_attn*",
"model.layers.43.mlp.router*",
"model.layers.43.self_attn*",
"model.layers.44.mlp.router*",
"model.layers.44.self_attn*",
"model.layers.45.mlp.router*",
"model.layers.45.self_attn*",
"model.layers.46.mlp.router*",
"model.layers.46.self_attn*",
"model.layers.47.mlp.router*",
"model.layers.47.self_attn*",
"model.layers.48.mlp.router*",
"model.layers.48.self_attn*",
"model.layers.49.mlp.router*",
"model.layers.49.self_attn*",
"model.layers.5.self_attn*",
"model.layers.50.mlp.router*",
"model.layers.50.self_attn*",
"model.layers.51.mlp.router*",
"model.layers.51.self_attn*",
"model.layers.52.mlp.router*",
"model.layers.52.self_attn*",
"model.layers.53.mlp.router*",
"model.layers.53.self_attn*",
"model.layers.54.mlp.router*",
"model.layers.54.self_attn*",
"model.layers.55.mlp.router*",
"model.layers.55.self_attn*",
"model.layers.56.mlp.router*",
"model.layers.56.self_attn*",
"model.layers.57.mlp.router*",
"model.layers.57.self_attn*",
"model.layers.58.mlp.router*",
"model.layers.58.self_attn*",
"model.layers.59.mlp.router*",
"model.layers.59.self_attn*",
"model.layers.6.mlp.router*",
"model.layers.6.self_attn*",
"model.layers.7.mlp.router*",
"model.layers.7.self_attn*",
"model.layers.8.mlp.router*",
"model.layers.8.self_attn*",
"model.layers.9.mlp.router*",
"model.layers.9.self_attn*"
],
"quant_algo": "NVFP4",
"producer": {
"name": "modelopt",
"version": "0.41.0"
},
"quant_method": "modelopt"
}
}