GitHub
183.3K
Download
Models
Docs
Pricing
Sign in
Download
Models
Docs
Pricing
Sign in
nemotron-3.5-super
:latest
2
Downloads
Updated
7 minutes ago
Nemotron 3.5 Super is NVIDIA’s vision-language model for coding, reasoning, tool use, and document understanding
Nemotron 3.5 Super is NVIDIA’s vision-language model for coding, reasoning, tool use, and document understanding
Cancel
vision
tools
thinking
120b
nemotron-3.5-super:latest
...
/
json
922b311ed7ab · 47kB
{
"architectures": [
"NemotronH_Omni_Reasoning_V3"
],
"auto_map": {
"AutoConfig": "configuration_nemotron_h_omni.NemotronH_Omni_Reasoning_V3_Config",
"AutoModel": "modeling_nemotron_h_omni.NemotronH_Omni_Reasoning_V3",
"AutoModelForCausalLM": "modeling_nemotron_h_omni.NemotronH_Omni_Reasoning_V3",
"AutoModelForImageTextToText": "modeling_nemotron_h_omni.NemotronH_Omni_Reasoning_V3"
},
"downsample_ratio": 0.5,
"dtype": "bfloat16",
"eos_token_id": 11,
"force_image_size": 512,
"image_tag_type": "internvl",
"img_context_token": "<image>",
"img_context_token_id": 18,
"llm_config": {
"architectures": [
"NemotronHForCausalLM"
],
"attention_bias": false,
"attention_dropout": 0.0,
"bos_token_id": 1,
"chunk_size": 128,
"conv_kernel": 4,
"dtype": "bfloat16",
"eos_token_id": 11,
"expand": 2,
"head_dim": 128,
"hidden_dropout": 0.0,
"hidden_size": 4096,
"initializer_range": 0.02,
"intermediate_size": 2688,
"layer_norm_epsilon": 1e-05,
"layers_block_type": [
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"attention",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"attention",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"attention",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"attention",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"attention",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"attention",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"attention",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"attention",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe",
"mamba",
"moe"
],
"mamba_head_dim": 64,
"mamba_hidden_act": "silu",
"mamba_num_heads": 128,
"mamba_proj_bias": false,
"mamba_ssm_cache_dtype": "float32",
"max_position_embeddings": 1048576,
"mlp_bias": false,
"mlp_hidden_act": "relu2",
"model_type": "nemotron_h",
"moe_intermediate_size": 2688,
"moe_latent_size": 1024,
"moe_shared_expert_intermediate_size": 5376,
"moe_shared_expert_overlap": false,
"mtp_layers_block_type": [
"attention",
"moe"
],
"n_group": 1,
"n_groups": 8,
"n_routed_experts": 512,
"n_shared_experts": 1,
"norm_topk_prob": true,
"num_attention_heads": 32,
"num_experts_per_tok": 22,
"num_key_value_heads": 2,
"num_logits_to_keep": 1,
"num_nextn_predict_layers": 1,
"pad_token_id": 0,
"rescale_prenorm_residual": true,
"residual_in_fp32": false,
"routed_scaling_factor": 5.0,
"sliding_window": null,
"ssm_state_size": 128,
"tie_word_embeddings": false,
"time_step_floor": 0.0001,
"time_step_limit": [
0.0,
{
"__float__": "Infinity"
}
],
"time_step_max": 0.1,
"time_step_min": 0.001,
"topk_group": 1,
"use_bias": false,
"use_cache": true,
"use_conv_bias": true,
"use_mamba_kernels": true,
"vocab_size": 131072
},
"model_type": "nemotron_h_omni",
"norm_mean": [
0.48145466,
0.4578275,
0.40821073
],
"norm_std": [
0.26862954,
0.26130258,
0.27577711
],
"patch_size": 16,
"processor_class": "NemotronH_Omni_Reasoning_V3Processor",
"projector_hidden_size": 20480,
"ps_version": "v2",
"sound_config": null,
"sound_context_token": "<audio>",
"sound_context_token_id": null,
"template": "n5h_5p5_nanov2",
"tie_word_embeddings": false,
"transformers_version": "5.12.1",
"use_thumbnail": false,
"video_context_token": "<video>",
"video_context_token_id": null,
"video_pruning_rate": 0.7,
"video_temporal_patch_size": 2,
"vision_config": {
"args": {
"cls_token_per_teacher": true,
"cpe_max_size": 2048,
"max_num_patches": 13312,
"min_num_patches": 1024,
"model": "vit_huge_patch16_224",
"register_multiple": 10,
"teachers": [
{
"name": 0,
"use_summary": true
},
{
"name": 1,
"use_summary": true
},
{
"name": 2,
"use_summary": false
}
]
},
"attention_probs_dropout_prob": 0.0,
"drop_path_rate": 0.0,
"dtype": "bfloat16",
"hidden_act": "gelu",
"hidden_dropout_prob": 0.0,
"hidden_size": 1280,
"image_size": 224,
"initializer_range": 0.02,
"layer_norm_eps": 1e-06,
"layerscale_value": 1.0,
"max_img_size": 2048,
"max_num_patches": 13312,
"min_num_patches": 1024,
"mlp_ratio": 4.0,
"model_type": "radio",
"norm_mean": [
0.48145466,
0.4578275,
0.40821073
],
"norm_std": [
0.26862954,
0.26130258,
0.27577711
],
"num_attention_heads": 16,
"num_channels": 3,
"num_cls_tokens": 3,
"num_hidden_layers": 32,
"num_registers": 7,
"patch_size": 16,
"qkv_bias": true,
"summary_idxs": [
0,
1
],
"use_swiglu_ffn": false,
"video_temporal_patch_size": 2
},
"vit_hidden_size": 1280,
"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": [
"language_model.lm_head",
"language_model.backbone.layers.0.mixer.in_proj",
"language_model.backbone.layers.0.mixer.out_proj",
"language_model.backbone.layers.11.mixer.in_proj",
"language_model.backbone.layers.11.mixer.out_proj",
"language_model.backbone.layers.13.mixer.in_proj",
"language_model.backbone.layers.13.mixer.out_proj",
"language_model.backbone.layers.15.mixer.in_proj",
"language_model.backbone.layers.15.mixer.out_proj",
"language_model.backbone.layers.18.mixer.in_proj",
"language_model.backbone.layers.18.mixer.out_proj",
"language_model.backbone.layers.2.mixer.in_proj",
"language_model.backbone.layers.2.mixer.out_proj",
"language_model.backbone.layers.20.mixer.in_proj",
"language_model.backbone.layers.20.mixer.out_proj",
"language_model.backbone.layers.22.mixer.in_proj",
"language_model.backbone.layers.22.mixer.out_proj",
"language_model.backbone.layers.24.mixer.in_proj",
"language_model.backbone.layers.24.mixer.out_proj",
"language_model.backbone.layers.27.mixer.in_proj",
"language_model.backbone.layers.27.mixer.out_proj",
"language_model.backbone.layers.29.mixer.in_proj",
"language_model.backbone.layers.29.mixer.out_proj",
"language_model.backbone.layers.31.mixer.in_proj",
"language_model.backbone.layers.31.mixer.out_proj",
"language_model.backbone.layers.33.mixer.in_proj",
"language_model.backbone.layers.33.mixer.out_proj",
"language_model.backbone.layers.35.mixer.in_proj",
"language_model.backbone.layers.35.mixer.out_proj",
"language_model.backbone.layers.38.mixer.in_proj",
"language_model.backbone.layers.38.mixer.out_proj",
"language_model.backbone.layers.4.mixer.in_proj",
"language_model.backbone.layers.4.mixer.out_proj",
"language_model.backbone.layers.40.mixer.in_proj",
"language_model.backbone.layers.40.mixer.out_proj",
"language_model.backbone.layers.42.mixer.in_proj",
"language_model.backbone.layers.42.mixer.out_proj",
"language_model.backbone.layers.44.mixer.in_proj",
"language_model.backbone.layers.44.mixer.out_proj",
"language_model.backbone.layers.46.mixer.in_proj",
"language_model.backbone.layers.46.mixer.out_proj",
"language_model.backbone.layers.49.mixer.in_proj",
"language_model.backbone.layers.49.mixer.out_proj",
"language_model.backbone.layers.51.mixer.in_proj",
"language_model.backbone.layers.51.mixer.out_proj",
"language_model.backbone.layers.53.mixer.in_proj",
"language_model.backbone.layers.53.mixer.out_proj",
"language_model.backbone.layers.55.mixer.in_proj",
"language_model.backbone.layers.55.mixer.out_proj",
"language_model.backbone.layers.57.mixer.in_proj",
"language_model.backbone.layers.57.mixer.out_proj",
"language_model.backbone.layers.6.mixer.in_proj",
"language_model.backbone.layers.6.mixer.out_proj",
"language_model.backbone.layers.60.mixer.in_proj",
"language_model.backbone.layers.60.mixer.out_proj",
"language_model.backbone.layers.62.mixer.in_proj",
"language_model.backbone.layers.62.mixer.out_proj",
"language_model.backbone.layers.64.mixer.in_proj",
"language_model.backbone.layers.64.mixer.out_proj",
"language_model.backbone.layers.66.mixer.in_proj",
"language_model.backbone.layers.66.mixer.out_proj",
"language_model.backbone.layers.68.mixer.in_proj",
"language_model.backbone.layers.68.mixer.out_proj",
"language_model.backbone.layers.71.mixer.in_proj",
"language_model.backbone.layers.71.mixer.out_proj",
"language_model.backbone.layers.73.mixer.in_proj",
"language_model.backbone.layers.73.mixer.out_proj",
"language_model.backbone.layers.75.mixer.in_proj",
"language_model.backbone.layers.75.mixer.out_proj",
"language_model.backbone.layers.77.mixer.in_proj",
"language_model.backbone.layers.77.mixer.out_proj",
"language_model.backbone.layers.80.mixer.in_proj",
"language_model.backbone.layers.80.mixer.out_proj",
"language_model.backbone.layers.82.mixer.in_proj",
"language_model.backbone.layers.82.mixer.out_proj",
"language_model.backbone.layers.84.mixer.in_proj",
"language_model.backbone.layers.84.mixer.out_proj",
"language_model.backbone.layers.86.mixer.in_proj",
"language_model.backbone.layers.86.mixer.out_proj",
"language_model.backbone.layers.9.mixer.in_proj",
"language_model.backbone.layers.9.mixer.out_proj"
]
},
"group_1": {
"weights": {
"dynamic": false,
"num_bits": 4,
"type": "float",
"group_size": 16
},
"targets": [
"language_model.backbone.layers.1.mixer.experts",
"language_model.backbone.layers.1.mixer.shared_experts.down_proj",
"language_model.backbone.layers.1.mixer.shared_experts.up_proj",
"language_model.backbone.layers.10.mixer.experts",
"language_model.backbone.layers.10.mixer.shared_experts.down_proj",
"language_model.backbone.layers.10.mixer.shared_experts.up_proj",
"language_model.backbone.layers.12.mixer.experts",
"language_model.backbone.layers.12.mixer.shared_experts.down_proj",
"language_model.backbone.layers.12.mixer.shared_experts.up_proj",
"language_model.backbone.layers.14.mixer.experts",
"language_model.backbone.layers.14.mixer.shared_experts.down_proj",
"language_model.backbone.layers.14.mixer.shared_experts.up_proj",
"language_model.backbone.layers.17.mixer.experts",
"language_model.backbone.layers.17.mixer.shared_experts.down_proj",
"language_model.backbone.layers.17.mixer.shared_experts.up_proj",
"language_model.backbone.layers.19.mixer.experts",
"language_model.backbone.layers.19.mixer.shared_experts.down_proj",
"language_model.backbone.layers.19.mixer.shared_experts.up_proj",
"language_model.backbone.layers.21.mixer.experts",
"language_model.backbone.layers.21.mixer.shared_experts.down_proj",
"language_model.backbone.layers.21.mixer.shared_experts.up_proj",
"language_model.backbone.layers.23.mixer.experts",
"language_model.backbone.layers.23.mixer.shared_experts.down_proj",
"language_model.backbone.layers.23.mixer.shared_experts.up_proj",
"language_model.backbone.layers.26.mixer.experts",
"language_model.backbone.layers.26.mixer.shared_experts.down_proj",
"language_model.backbone.layers.26.mixer.shared_experts.up_proj",
"language_model.backbone.layers.28.mixer.experts",
"language_model.backbone.layers.28.mixer.shared_experts.down_proj",
"language_model.backbone.layers.28.mixer.shared_experts.up_proj",
"language_model.backbone.layers.3.mixer.experts",
"language_model.backbone.layers.3.mixer.shared_experts.down_proj",
"language_model.backbone.layers.3.mixer.shared_experts.up_proj",
"language_model.backbone.layers.30.mixer.experts",
"language_model.backbone.layers.30.mixer.shared_experts.down_proj",
"language_model.backbone.layers.30.mixer.shared_experts.up_proj",
"language_model.backbone.layers.32.mixer.experts",
"language_model.backbone.layers.32.mixer.shared_experts.down_proj",
"language_model.backbone.layers.32.mixer.shared_experts.up_proj",
"language_model.backbone.layers.34.mixer.experts",
"language_model.backbone.layers.34.mixer.shared_experts.down_proj",
"language_model.backbone.layers.34.mixer.shared_experts.up_proj",
"language_model.backbone.layers.37.mixer.experts",
"language_model.backbone.layers.37.mixer.shared_experts.down_proj",
"language_model.backbone.layers.37.mixer.shared_experts.up_proj",
"language_model.backbone.layers.39.mixer.experts",
"language_model.backbone.layers.39.mixer.shared_experts.down_proj",
"language_model.backbone.layers.39.mixer.shared_experts.up_proj",
"language_model.backbone.layers.41.mixer.experts",
"language_model.backbone.layers.41.mixer.shared_experts.down_proj",
"language_model.backbone.layers.41.mixer.shared_experts.up_proj",
"language_model.backbone.layers.43.mixer.experts",
"language_model.backbone.layers.43.mixer.shared_experts.down_proj",
"language_model.backbone.layers.43.mixer.shared_experts.up_proj",
"language_model.backbone.layers.45.mixer.experts",
"language_model.backbone.layers.45.mixer.shared_experts.down_proj",
"language_model.backbone.layers.45.mixer.shared_experts.up_proj",
"language_model.backbone.layers.48.mixer.experts",
"language_model.backbone.layers.48.mixer.shared_experts.down_proj",
"language_model.backbone.layers.48.mixer.shared_experts.up_proj",
"language_model.backbone.layers.5.mixer.experts",
"language_model.backbone.layers.5.mixer.shared_experts.down_proj",
"language_model.backbone.layers.5.mixer.shared_experts.up_proj",
"language_model.backbone.layers.50.mixer.experts",
"language_model.backbone.layers.50.mixer.shared_experts.down_proj",
"language_model.backbone.layers.50.mixer.shared_experts.up_proj",
"language_model.backbone.layers.52.mixer.experts",
"language_model.backbone.layers.52.mixer.shared_experts.down_proj",
"language_model.backbone.layers.52.mixer.shared_experts.up_proj",
"language_model.backbone.layers.54.mixer.experts",
"language_model.backbone.layers.54.mixer.shared_experts.down_proj",
"language_model.backbone.layers.54.mixer.shared_experts.up_proj",
"language_model.backbone.layers.56.mixer.experts",
"language_model.backbone.layers.56.mixer.shared_experts.down_proj",
"language_model.backbone.layers.56.mixer.shared_experts.up_proj",
"language_model.backbone.layers.59.mixer.experts",
"language_model.backbone.layers.59.mixer.shared_experts.down_proj",
"language_model.backbone.layers.59.mixer.shared_experts.up_proj",
"language_model.backbone.layers.61.mixer.experts",
"language_model.backbone.layers.61.mixer.shared_experts.down_proj",
"language_model.backbone.layers.61.mixer.shared_experts.up_proj",
"language_model.backbone.layers.63.mixer.experts",
"language_model.backbone.layers.63.mixer.shared_experts.down_proj",
"language_model.backbone.layers.63.mixer.shared_experts.up_proj",
"language_model.backbone.layers.65.mixer.experts",
"language_model.backbone.layers.65.mixer.shared_experts.down_proj",
"language_model.backbone.layers.65.mixer.shared_experts.up_proj",
"language_model.backbone.layers.67.mixer.experts",
"language_model.backbone.layers.67.mixer.shared_experts.down_proj",
"language_model.backbone.layers.67.mixer.shared_experts.up_proj",
"language_model.backbone.layers.70.mixer.experts",
"language_model.backbone.layers.70.mixer.shared_experts.down_proj",
"language_model.backbone.layers.70.mixer.shared_experts.up_proj",
"language_model.backbone.layers.72.mixer.experts",
"language_model.backbone.layers.72.mixer.shared_experts.down_proj",
"language_model.backbone.layers.72.mixer.shared_experts.up_proj",
"language_model.backbone.layers.74.mixer.experts",
"language_model.backbone.layers.74.mixer.shared_experts.down_proj",
"language_model.backbone.layers.74.mixer.shared_experts.up_proj",
"language_model.backbone.layers.76.mixer.experts",
"language_model.backbone.layers.76.mixer.shared_experts.down_proj",
"language_model.backbone.layers.76.mixer.shared_experts.up_proj",
"language_model.backbone.layers.79.mixer.experts",
"language_model.backbone.layers.79.mixer.shared_experts.down_proj",
"language_model.backbone.layers.79.mixer.shared_experts.up_proj",
"language_model.backbone.layers.8.mixer.experts",
"language_model.backbone.layers.8.mixer.shared_experts.down_proj",
"language_model.backbone.layers.8.mixer.shared_experts.up_proj",
"language_model.backbone.layers.81.mixer.experts",
"language_model.backbone.layers.81.mixer.shared_experts.down_proj",
"language_model.backbone.layers.81.mixer.shared_experts.up_proj",
"language_model.backbone.layers.83.mixer.experts",
"language_model.backbone.layers.83.mixer.shared_experts.down_proj",
"language_model.backbone.layers.83.mixer.shared_experts.up_proj",
"language_model.backbone.layers.85.mixer.experts",
"language_model.backbone.layers.85.mixer.shared_experts.down_proj",
"language_model.backbone.layers.85.mixer.shared_experts.up_proj",
"language_model.backbone.layers.87.mixer.experts",
"language_model.backbone.layers.87.mixer.shared_experts.down_proj",
"language_model.backbone.layers.87.mixer.shared_experts.up_proj"
]
}
},
"quantized_layers": {
"language_model.backbone.layers.0.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.0.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.1.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.1.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.1.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.2.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.2.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.3.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.3.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.3.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.4.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.4.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.5.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.5.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.5.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.6.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.6.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.8.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.8.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.8.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.9.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.9.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.10.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.10.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.10.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.11.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.11.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.12.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.12.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.12.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.13.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.13.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.14.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.14.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.14.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.15.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.15.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.17.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.17.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.17.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.18.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.18.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.19.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.19.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.19.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.20.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.20.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.21.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.21.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.21.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.22.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.22.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.23.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.23.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.23.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.24.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.24.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.26.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.26.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.26.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.27.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.27.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.28.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.28.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.28.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.29.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.29.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.30.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.30.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.30.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.31.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.31.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.32.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.32.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.32.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.33.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.33.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.34.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.34.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.34.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.35.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.35.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.37.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.37.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.37.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.38.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.38.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.39.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.39.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.39.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.40.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.40.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.41.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.41.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.41.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.42.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.42.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.43.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.43.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.43.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.44.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.44.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.45.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.45.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.45.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.46.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.46.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.48.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.48.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.48.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.49.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.49.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.50.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.50.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.50.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.51.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.51.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.52.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.52.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.52.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.53.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.53.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.54.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.54.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.54.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.55.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.55.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.56.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.56.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.56.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.57.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.57.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.59.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.59.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.59.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.60.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.60.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.61.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.61.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.61.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.62.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.62.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.63.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.63.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.63.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.64.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.64.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.65.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.65.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.65.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.66.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.66.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.67.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.67.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.67.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.68.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.68.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.70.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.70.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.70.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.71.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.71.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.72.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.72.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.72.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.73.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.73.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.74.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.74.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.74.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.75.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.75.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.76.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.76.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.76.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.77.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.77.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.79.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.79.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.79.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.80.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.80.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.81.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.81.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.81.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.82.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.82.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.83.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.83.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.83.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.84.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.84.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.85.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.85.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.85.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.86.mixer.in_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.86.mixer.out_proj": {
"quant_algo": "MXFP8",
"group_size": 32
},
"language_model.backbone.layers.87.mixer.experts": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.87.mixer.shared_experts.up_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.backbone.layers.87.mixer.shared_experts.down_proj": {
"quant_algo": "W4A16_NVFP4",
"group_size": 16
},
"language_model.lm_head": {
"quant_algo": "MXFP8",
"group_size": 32
}
},
"ignore": [
"language_model.mtp*",
"language_model.mtp.layers.0*",
"language_model.mtp.layers.1*"
],
"quant_algo": "MIXED_PRECISION",
"producer": {
"name": "modelopt",
"version": "0.47.0rc1.dev36+g613e5e8b2"
},
"quant_method": "modelopt"
},
"quantization": {
"bits": 8,
"mode": "mxfp8"
}
}