Upload config.json with huggingface_hub
Browse files- config.json +15 -95
config.json
CHANGED
|
@@ -1,8 +1,4 @@
|
|
| 1 |
{
|
| 2 |
-
"add_classification_head": false,
|
| 3 |
-
"anyres_pooling_size": 2,
|
| 4 |
-
"anyres_vit_max_image_size": 2048,
|
| 5 |
-
"anyres_vit_two_views": false,
|
| 6 |
"architectures": [
|
| 7 |
"HunYuanVLForConditionalGeneration"
|
| 8 |
],
|
|
@@ -10,59 +6,29 @@
|
|
| 10 |
"attention_dropout": 0.0,
|
| 11 |
"attention_head_dim": 128,
|
| 12 |
"bos_token_id": 120000,
|
| 13 |
-
"cla_share_factor": 2,
|
| 14 |
-
"class_num": 0,
|
| 15 |
-
"dense_list": [
|
| 16 |
-
1024,
|
| 17 |
-
0
|
| 18 |
-
],
|
| 19 |
"eod_token_id": 120020,
|
| 20 |
"eos_token_id": 120020,
|
| 21 |
-
"group_limited_greedy": false,
|
| 22 |
"head_dim": 128,
|
| 23 |
"hidden_act": "silu",
|
| 24 |
"hidden_size": 1024,
|
| 25 |
-
"
|
| 26 |
-
"
|
| 27 |
-
"im_start_id": 120118,
|
| 28 |
"image_token_id": 120120,
|
|
|
|
| 29 |
"initializer_range": 0.02,
|
| 30 |
"intermediate_size": 3584,
|
| 31 |
-
"kv_lora_rank": null,
|
| 32 |
-
"mask_init_id": 13,
|
| 33 |
"max_position_embeddings": 32768,
|
| 34 |
-
"tie_word_embeddings": true,
|
| 35 |
"mlp_bias": false,
|
| 36 |
"model_type": "hunyuan_vl",
|
| 37 |
-
"moe_drop_tokens": false,
|
| 38 |
-
"moe_intermediate_size": null,
|
| 39 |
-
"moe_layer_num_skipped": 0,
|
| 40 |
-
"moe_random_routing_dropped_token": false,
|
| 41 |
-
"moe_topk": null,
|
| 42 |
-
"mtp_loss_factor": 0.3,
|
| 43 |
-
"mtp_moe_layer_num_skipped": 0,
|
| 44 |
-
"mtp_no_bias": false,
|
| 45 |
-
"mtp_type": "moe",
|
| 46 |
-
"n_group": null,
|
| 47 |
-
"norm_topk_prob": true,
|
| 48 |
"norm_type": "rms",
|
| 49 |
"num_attention_heads": 16,
|
| 50 |
"num_experts": 1,
|
| 51 |
"num_hidden_layers": 24,
|
| 52 |
"num_key_value_heads": 8,
|
| 53 |
-
"num_media_embeds": 257,
|
| 54 |
-
"num_nextn_predict_layers": 1,
|
| 55 |
-
"num_predictor_layers": 1,
|
| 56 |
-
"num_shared_expert": null,
|
| 57 |
"org_vocab_size": 120818,
|
| 58 |
"pad_id": 120002,
|
| 59 |
-
"pad_token_id":
|
| 60 |
-
"pool_type": "last",
|
| 61 |
-
"position_embedding_xdrope": true,
|
| 62 |
"pretraining_tp": 1,
|
| 63 |
-
"q_lora_rank": null,
|
| 64 |
-
"qk_nope_head_dim": null,
|
| 65 |
-
"qk_rope_head_dim": null,
|
| 66 |
"rms_norm_eps": 1e-05,
|
| 67 |
"rope_scaling": {
|
| 68 |
"alpha": 1000.0,
|
|
@@ -71,90 +37,44 @@
|
|
| 71 |
"factor": 1.0,
|
| 72 |
"mscale": 1.0,
|
| 73 |
"mscale_all_dim": 1.0,
|
| 74 |
-
"type": "xdrope"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 75 |
},
|
| 76 |
"rope_theta": 10000.0,
|
| 77 |
"routed_scaling_factor": 1.0,
|
| 78 |
"sep_token_id": 0,
|
| 79 |
-
"skip_cls_token": true,
|
| 80 |
"text_end_id": 8,
|
| 81 |
"text_start_id": 7,
|
| 82 |
"tie_word_embeddings": true,
|
| 83 |
-
"
|
| 84 |
-
"torch_dtype": "bfloat16",
|
| 85 |
"transformers_version": "4.49.0",
|
| 86 |
"use_cache": true,
|
| 87 |
-
"use_cla": false,
|
| 88 |
-
"use_mixed_mlp_moe": false,
|
| 89 |
-
"use_mla": false,
|
| 90 |
"use_qk_norm": true,
|
| 91 |
-
"
|
| 92 |
-
"v_head_dim": null,
|
| 93 |
-
"video_end_id": 120123,
|
| 94 |
-
"video_start_id": 120122,
|
| 95 |
"vision_config": {
|
| 96 |
-
"adaptor_patch_size": 2,
|
| 97 |
"add_patchemb_bias": true,
|
| 98 |
"attention_dropout": 0.0,
|
| 99 |
"cat_extra_token": 1,
|
| 100 |
-
"fuse_vit_feature": 0,
|
| 101 |
"hidden_act": "gelu",
|
| 102 |
"hidden_dropout": 0.0,
|
| 103 |
"hidden_size": 1152,
|
| 104 |
"img_max_token_num": 4096,
|
| 105 |
"intermediate_size": 4304,
|
| 106 |
"interpolate_mode": "bilinear",
|
| 107 |
-
"layer_norm_eps": 1e-05,
|
| 108 |
"max_image_size": 2048,
|
| 109 |
"max_vit_seq_len": 16384,
|
| 110 |
-
"num_key_value_heads": 16,
|
| 111 |
"num_attention_heads": 16,
|
| 112 |
"num_channels": 3,
|
| 113 |
"num_hidden_layers": 27,
|
| 114 |
"out_hidden_size": 1024,
|
| 115 |
"patch_size": 16,
|
| 116 |
-
"use_qk_norm": false,
|
| 117 |
-
"attention_bias": true,
|
| 118 |
"rms_norm_eps": 1e-05,
|
| 119 |
-
"
|
| 120 |
-
0.25,
|
| 121 |
-
0.25,
|
| 122 |
-
0.25,
|
| 123 |
-
0.25
|
| 124 |
-
],
|
| 125 |
-
"use_rotary_pos_emb": false,
|
| 126 |
-
"position_embedding_xdrope": false,
|
| 127 |
-
"max_position_embeddings": 262144,
|
| 128 |
-
"rope_theta": 10000.0,
|
| 129 |
-
"mlp_bias": true,
|
| 130 |
-
"norm_type": "torch_nn",
|
| 131 |
-
"anyres_pooling_size": 2,
|
| 132 |
-
"anyres_vit_max_image_size": 2048,
|
| 133 |
-
"anyres_vit_two_views": false,
|
| 134 |
-
"perceive_pre_norm": true,
|
| 135 |
-
"perceive_post_norm": true,
|
| 136 |
-
"poolmlp": 0,
|
| 137 |
-
"remove_prenorm": true,
|
| 138 |
-
"two_view": true,
|
| 139 |
-
"use_normalize_pool": false,
|
| 140 |
-
"use_fusion_block": false,
|
| 141 |
-
"use_layernorm": true
|
| 142 |
},
|
| 143 |
-
"
|
| 144 |
-
"vit_input_resolution": 224,
|
| 145 |
-
"vit_mapping_type": "simple_conv_mlp",
|
| 146 |
-
"vit_norm_type": "fused",
|
| 147 |
-
"vit_patch": 1,
|
| 148 |
-
"vit_path": null,
|
| 149 |
-
"vit_remove_prenorm": true,
|
| 150 |
-
"vit_token": 58,
|
| 151 |
-
"vit_type": "AnyResVit-tp",
|
| 152 |
-
"vit_used_rms_norm": true,
|
| 153 |
-
"vocab_size": 120818,
|
| 154 |
-
"xdrope_section": [
|
| 155 |
-
0.25,
|
| 156 |
-
0.25,
|
| 157 |
-
0.25,
|
| 158 |
-
0.25
|
| 159 |
-
]
|
| 160 |
}
|
|
|
|
| 1 |
{
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
"architectures": [
|
| 3 |
"HunYuanVLForConditionalGeneration"
|
| 4 |
],
|
|
|
|
| 6 |
"attention_dropout": 0.0,
|
| 7 |
"attention_head_dim": 128,
|
| 8 |
"bos_token_id": 120000,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
"eod_token_id": 120020,
|
| 10 |
"eos_token_id": 120020,
|
|
|
|
| 11 |
"head_dim": 128,
|
| 12 |
"hidden_act": "silu",
|
| 13 |
"hidden_size": 1024,
|
| 14 |
+
"image_start_token_id": 120118,
|
| 15 |
+
"image_end_token_id": 120119,
|
|
|
|
| 16 |
"image_token_id": 120120,
|
| 17 |
+
"image_newline_token_id": 120121,
|
| 18 |
"initializer_range": 0.02,
|
| 19 |
"intermediate_size": 3584,
|
|
|
|
|
|
|
| 20 |
"max_position_embeddings": 32768,
|
|
|
|
| 21 |
"mlp_bias": false,
|
| 22 |
"model_type": "hunyuan_vl",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
"norm_type": "rms",
|
| 24 |
"num_attention_heads": 16,
|
| 25 |
"num_experts": 1,
|
| 26 |
"num_hidden_layers": 24,
|
| 27 |
"num_key_value_heads": 8,
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
"org_vocab_size": 120818,
|
| 29 |
"pad_id": 120002,
|
| 30 |
+
"pad_token_id": -1,
|
|
|
|
|
|
|
| 31 |
"pretraining_tp": 1,
|
|
|
|
|
|
|
|
|
|
| 32 |
"rms_norm_eps": 1e-05,
|
| 33 |
"rope_scaling": {
|
| 34 |
"alpha": 1000.0,
|
|
|
|
| 37 |
"factor": 1.0,
|
| 38 |
"mscale": 1.0,
|
| 39 |
"mscale_all_dim": 1.0,
|
| 40 |
+
"type": "xdrope",
|
| 41 |
+
"xdrope_section": [
|
| 42 |
+
16,
|
| 43 |
+
16,
|
| 44 |
+
16,
|
| 45 |
+
16
|
| 46 |
+
]
|
| 47 |
},
|
| 48 |
"rope_theta": 10000.0,
|
| 49 |
"routed_scaling_factor": 1.0,
|
| 50 |
"sep_token_id": 0,
|
|
|
|
| 51 |
"text_end_id": 8,
|
| 52 |
"text_start_id": 7,
|
| 53 |
"tie_word_embeddings": true,
|
| 54 |
+
"dtype": "bfloat16",
|
|
|
|
| 55 |
"transformers_version": "4.49.0",
|
| 56 |
"use_cache": true,
|
|
|
|
|
|
|
|
|
|
| 57 |
"use_qk_norm": true,
|
| 58 |
+
"use_cla": false,
|
|
|
|
|
|
|
|
|
|
| 59 |
"vision_config": {
|
|
|
|
| 60 |
"add_patchemb_bias": true,
|
| 61 |
"attention_dropout": 0.0,
|
| 62 |
"cat_extra_token": 1,
|
|
|
|
| 63 |
"hidden_act": "gelu",
|
| 64 |
"hidden_dropout": 0.0,
|
| 65 |
"hidden_size": 1152,
|
| 66 |
"img_max_token_num": 4096,
|
| 67 |
"intermediate_size": 4304,
|
| 68 |
"interpolate_mode": "bilinear",
|
|
|
|
| 69 |
"max_image_size": 2048,
|
| 70 |
"max_vit_seq_len": 16384,
|
|
|
|
| 71 |
"num_attention_heads": 16,
|
| 72 |
"num_channels": 3,
|
| 73 |
"num_hidden_layers": 27,
|
| 74 |
"out_hidden_size": 1024,
|
| 75 |
"patch_size": 16,
|
|
|
|
|
|
|
| 76 |
"rms_norm_eps": 1e-05,
|
| 77 |
+
"spatial_merge_size": 2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
},
|
| 79 |
+
"vocab_size": 120818
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
}
|