lfm3-omni:2b-mxfp8

12 9 minutes ago

LFM3-17B is a tri-hybrid for local agents and specialized applications.

vision tools audio 2b
... /
json
17e3792551ef ยท 5.9kB
{
"architectures": [
"Lfm3OmniForConditionalGeneration"
],
"audio_encoder_config": {
"conv_channels": 256,
"hidden_size": 768,
"initializer_range": 0.02,
"model_type": "lfm3_omni_encoder",
"num_mel_bins": 128,
"sliding_window": 32,
"text_config": {
"_name_or_path": "",
"architectures": null,
"block_auto_adjust_ff_dim": true,
"block_ffn_dim_multiplier": 1.0,
"block_multiple_of": 256,
"bos_token_id": 1,
"chunk_size_feed_forward": 0,
"conv_L_cache": 3,
"conv_bias": false,
"dtype": null,
"eos_token_id": null,
"full_attn_idxs": null,
"hidden_size": 768,
"id2label": {
"0": "LABEL_0",
"1": "LABEL_1"
},
"initializer_range": 0.02,
"intermediate_size": 2560,
"is_encoder_decoder": false,
"label2id": {
"LABEL_0": 0,
"LABEL_1": 1
},
"layer_types": [
"conv",
"conv",
"full_attention",
"conv",
"conv",
"full_attention",
"conv",
"conv",
"full_attention",
"conv",
"conv",
"full_attention",
"conv",
"conv",
"full_attention",
"full_attention",
"full_attention"
],
"max_position_embeddings": 128000,
"model_type": "lfm2",
"norm_eps": 1e-05,
"num_attention_heads": 8,
"num_hidden_layers": 17,
"num_key_value_heads": 4,
"output_attentions": false,
"output_hidden_states": false,
"pad_token_id": 0,
"problem_type": null,
"return_dict": true,
"rope_parameters": {
"rope_theta": 1000000.0,
"rope_type": "default"
},
"tie_word_embeddings": true,
"use_cache": true,
"vocab_size": 2
}
},
"audio_eos_code": 2016,
"audio_loss_weight": 1.0,
"audio_out_token_id": 125032,
"audio_start_token_id": 125012,
"audio_token_id": 125031,
"codebook_size": 2017,
"depth_config": {
"bos_token_id": null,
"conv_L_cache": 7,
"conv_bias": false,
"eos_token_id": null,
"head_dim": 32,
"hidden_size": 1024,
"initializer_range": 0.02,
"intermediate_size": 2816,
"layer_types": [
"full_attention",
"full_attention",
"full_attention"
],
"max_position_embeddings": 12,
"model_type": "lfm3",
"norm_eps": 1e-05,
"num_attention_heads": 32,
"num_hidden_layers": 3,
"num_key_value_heads": 8,
"pad_token_id": null,
"rope_parameters": {
"full_attention": {
"partial_rotary_factor": 1.0,
"rope_theta": 1000000.0,
"rope_type": "default"
}
},
"sliding_window": 256,
"tie_word_embeddings": true,
"use_cache": false,
"use_sandwich_norm": false,
"vocab_size": 1
},
"downsample_factor": 2,
"image_token_id": 125030,
"initializer_range": 0.02,
"interleaved_n_audio": 12,
"interleaved_n_text": 6,
"maskgit_schedule": [
4,
4,
4
],
"mixed_start_token_id": 125015,
"model_type": "lfm3_omni",
"num_codebooks": 12,
"projector_bias": true,
"projector_hidden_act": "gelu",
"projector_hidden_size": 4096,
"projector_layer_norm_eps": 1e-06,
"text_config": {
"bos_token_id": 124894,
"conv_L_cache": 7,
"conv_bias": false,
"eos_token_id": 124900,
"head_dim": 128,
"hidden_size": 2048,
"initializer_range": 0.02,
"intermediate_size": 7680,
"layer_types": [
"conv",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv"
],
"max_position_embeddings": 128000,
"model_type": "lfm3",
"norm_eps": 1e-05,
"num_attention_heads": 16,
"num_hidden_layers": 24,
"num_key_value_heads": 4,
"pad_token_id": 124893,
"rope_parameters": {
"full_attention": {
"partial_rotary_factor": 0.5,
"rope_theta": 1000000.0,
"rope_type": "default"
},
"sliding_attention": {
"rope_theta": 10000,
"rope_type": "default"
}
},
"sliding_window": 256,
"tie_word_embeddings": true,
"use_cache": true,
"use_sandwich_norm": true,
"vocab_size": 128000
},
"text_end_token_id": 125014,
"text_loss_weight": 1.0,
"tie_word_embeddings": true,
"transformers_version": "5.19.0.dev0",
"video_token_id": 125029,
"vision_config": {
"conv_kernel_size": 3,
"dtype": "bfloat16",
"hidden_size": 1024,
"in_channels": 3,
"initializer_range": 0.02,
"intermediate_size": 2560,
"layer_types": [
"conv",
"conv",
"full_attention",
"conv",
"full_attention",
"conv",
"full_attention",
"conv",
"full_attention",
"conv",
"full_attention",
"conv",
"full_attention",
"conv"
],
"model_type": "lfm3_vl_vision",
"norm_eps": 1e-05,
"num_attention_heads": 16,
"num_hidden_layers": 14,
"num_key_value_heads": 8,
"patch_size": 16,
"rope_parameters": {
"rope_theta": 1000000.0,
"rope_type": "axial"
},
"spatial_merge_size": 2,
"temporal_patch_size": 2
},
"auto_map": {
"AutoConfig": "configuration_lfm3_omni.Lfm3OmniConfig",
"AutoModel": "modeling_lfm3_omni.Lfm3OmniModel",
"AutoModelForImageTextToText": "modeling_lfm3_omni.Lfm3OmniForConditionalGeneration",
"AutoModelForMultimodalLM": "modeling_lfm3_omni.Lfm3OmniForConditionalGeneration",
"AutoModelForSeq2SeqLM": "modeling_lfm3_omni.Lfm3OmniForConditionalGeneration"
}
}