nemotron-3-nano:4b-mlx

918.6K an hour ago

Nemotron-3-Nano is a new Standard for Efficient, Open, and Intelligent Agentic Models, now updated with a 4B parameter count model.

tools thinking cloud 4b 30b
... /
json
47a802a9ab95 ยท 17kB
{
"architectures": [
"NemotronHForCausalLM"
],
"attention_bias": false,
"attention_dropout": 0.0,
"head_dim": 128,
"auto_map": {
"AutoConfig": "configuration_nemotron_h.NemotronHConfig",
"AutoModelForCausalLM": "modeling_nemotron_h.NemotronHForCausalLM"
},
"bos_token_id": 1,
"chunk_size": 256,
"conv_kernel": 4,
"eos_token_id": 2,
"expand": 2,
"hidden_dropout": 0.0,
"hidden_size": 3136,
"hybrid_override_pattern": "M-M-M-MM-M-M*-M-M*-M-M-M*-M-M-MM*-MMM-M-M-",
"initializer_range": 0.02,
"intermediate_size": 12544,
"layer_norm_epsilon": 1e-05,
"mamba_head_dim": 80,
"mamba_hidden_act": "silu",
"mamba_num_heads": 96,
"mamba_proj_bias": false,
"max_position_embeddings": 262144,
"mlp_bias": false,
"mlp_hidden_act": "relu2",
"model_type": "nemotron_h",
"n_groups": 8,
"num_attention_heads": 40,
"num_hidden_layers": 42,
"num_key_value_heads": 8,
"num_logits_to_keep": 1,
"pad_token_id": 0,
"rescale_prenorm_residual": true,
"residual_in_fp32": false,
"rms_norm_eps": 1e-05,
"sliding_window": null,
"ssm_state_size": 128,
"tie_word_embeddings": false,
"time_step_floor": 0.0001,
"time_step_max": 0.1,
"time_step_min": 0.001,
"time_step_rank": 256,
"torch_dtype": "bfloat16",
"transformers_version": "4.53.0",
"use_bias": false,
"use_cache": true,
"use_conv_bias": true,
"use_mamba_kernels": true,
"vocab_size": 131072,
"quantization_config": {
"config_groups": {
"group_0": {
"input_activations": {
"dynamic": false,
"num_bits": 8,
"type": "float",
"group_size": 32
},
"weights": {
"dynamic": false,
"num_bits": 8,
"type": "float",
"group_size": 32
},
"targets": [
"backbone.embeddings",
"backbone.layers.0.mixer.out_proj",
"backbone.layers.11.mixer.out_proj",
"backbone.layers.12.mixer.k_proj",
"backbone.layers.12.mixer.o_proj",
"backbone.layers.12.mixer.q_proj",
"backbone.layers.12.mixer.v_proj",
"backbone.layers.14.mixer.out_proj",
"backbone.layers.16.mixer.out_proj",
"backbone.layers.17.mixer.k_proj",
"backbone.layers.17.mixer.o_proj",
"backbone.layers.17.mixer.q_proj",
"backbone.layers.17.mixer.v_proj",
"backbone.layers.19.mixer.out_proj",
"backbone.layers.2.mixer.out_proj",
"backbone.layers.21.mixer.out_proj",
"backbone.layers.23.mixer.out_proj",
"backbone.layers.24.mixer.k_proj",
"backbone.layers.24.mixer.o_proj",
"backbone.layers.24.mixer.q_proj",
"backbone.layers.24.mixer.v_proj",
"backbone.layers.26.mixer.out_proj",
"backbone.layers.28.mixer.out_proj",
"backbone.layers.30.mixer.out_proj",
"backbone.layers.31.mixer.out_proj",
"backbone.layers.32.mixer.k_proj",
"backbone.layers.32.mixer.o_proj",
"backbone.layers.32.mixer.q_proj",
"backbone.layers.32.mixer.v_proj",
"backbone.layers.34.mixer.out_proj",
"backbone.layers.35.mixer.out_proj",
"backbone.layers.36.mixer.out_proj",
"backbone.layers.38.mixer.out_proj",
"backbone.layers.4.mixer.out_proj",
"backbone.layers.40.mixer.out_proj",
"backbone.layers.6.mixer.out_proj",
"backbone.layers.7.mixer.out_proj",
"backbone.layers.9.mixer.out_proj",
"lm_head"
]
},
"group_1": {
"weights": {
"dynamic": false,
"num_bits": 4,
"type": "float",
"group_size": 16
},
"targets": [
"backbone.layers.0.mixer.in_proj",
"backbone.layers.11.mixer.in_proj",
"backbone.layers.14.mixer.in_proj",
"backbone.layers.16.mixer.in_proj",
"backbone.layers.19.mixer.in_proj",
"backbone.layers.2.mixer.in_proj",
"backbone.layers.21.mixer.in_proj",
"backbone.layers.23.mixer.in_proj",
"backbone.layers.26.mixer.in_proj",
"backbone.layers.28.mixer.in_proj",
"backbone.layers.30.mixer.in_proj",
"backbone.layers.31.mixer.in_proj",
"backbone.layers.34.mixer.in_proj",
"backbone.layers.35.mixer.in_proj",
"backbone.layers.36.mixer.in_proj",
"backbone.layers.38.mixer.in_proj",
"backbone.layers.4.mixer.in_proj",
"backbone.layers.40.mixer.in_proj",
"backbone.layers.6.mixer.in_proj",
"backbone.layers.7.mixer.in_proj",
"backbone.layers.9.mixer.in_proj"
]
},
"group_2": {
"input_activations": {
"dynamic": false,
"num_bits": 4,
"type": "float",
"group_size": 16
},
"weights": {
"dynamic": false,
"num_bits": 4,
"type": "float",
"group_size": 16
},
"targets": [
"backbone.layers.1.mixer.down_proj",
"backbone.layers.1.mixer.up_proj",
"backbone.layers.10.mixer.down_proj",
"backbone.layers.10.mixer.up_proj",
"backbone.layers.13.mixer.down_proj",
"backbone.layers.13.mixer.up_proj",
"backbone.layers.15.mixer.down_proj",
"backbone.layers.15.mixer.up_proj",
"backbone.layers.18.mixer.down_proj",
"backbone.layers.18.mixer.up_proj",
"backbone.layers.20.mixer.down_proj",
"backbone.layers.20.mixer.up_proj",
"backbone.layers.22.mixer.down_proj",
"backbone.layers.22.mixer.up_proj",
"backbone.layers.25.mixer.down_proj",
"backbone.layers.25.mixer.up_proj",
"backbone.layers.27.mixer.down_proj",
"backbone.layers.27.mixer.up_proj",
"backbone.layers.29.mixer.down_proj",
"backbone.layers.29.mixer.up_proj",
"backbone.layers.3.mixer.down_proj",
"backbone.layers.3.mixer.up_proj",
"backbone.layers.33.mixer.down_proj",
"backbone.layers.33.mixer.up_proj",
"backbone.layers.37.mixer.down_proj",
"backbone.layers.37.mixer.up_proj",
"backbone.layers.39.mixer.down_proj",
"backbone.layers.39.mixer.up_proj",
"backbone.layers.41.mixer.down_proj",
"backbone.layers.41.mixer.up_proj",
"backbone.layers.5.mixer.down_proj",
"backbone.layers.5.mixer.up_proj",
"backbone.layers.8.mixer.down_proj",
"backbone.layers.8.mixer.up_proj"
]
}
},
"quantized_layers": {
"backbone.embeddings": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.0.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.0.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.1.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.1.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.2.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.2.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.3.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.3.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.4.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.4.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.5.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.5.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.6.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.6.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.7.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.7.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.8.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.8.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.9.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.9.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.10.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.10.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.11.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.11.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.12.mixer.q_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.12.mixer.k_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.12.mixer.v_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.12.mixer.o_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.13.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.13.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.14.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.14.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.15.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.15.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.16.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.16.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.17.mixer.q_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.17.mixer.k_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.17.mixer.v_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.17.mixer.o_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.18.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.18.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.19.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.19.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.20.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.20.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.21.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.21.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.22.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.22.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.23.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.23.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.24.mixer.q_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.24.mixer.k_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.24.mixer.v_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.24.mixer.o_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.25.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.25.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.26.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.26.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.27.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.27.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.28.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.28.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.29.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.29.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.30.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.30.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.31.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.31.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.32.mixer.q_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.32.mixer.k_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.32.mixer.v_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.32.mixer.o_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.33.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.33.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.34.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.34.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.35.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.35.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.36.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.36.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.37.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.37.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.38.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.38.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.39.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.39.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.40.mixer.in_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"backbone.layers.40.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"backbone.layers.41.mixer.up_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"backbone.layers.41.mixer.down_proj": {
"quant_algo": "NVFP4",
"group_size": 16
},
"lm_head": {
"quant_algo": "MXFP8",
"group_size": 32
}
},
"ignore": [],
"quant_algo": "MIXED_PRECISION",
"kv_cache_scheme": "NVFP4",
"producer": {
"name": "modelopt",
"version": "0.0.0"
},
"quant_method": "modelopt"
},
"quantization": {
"bits": 8,
"mode": "mxfp8"
}
}