Update config.json
Browse files- config.json +9 -36
config.json
CHANGED
|
@@ -54,18 +54,9 @@
|
|
| 54 |
"transformers_version": "4.46.0.dev0",
|
| 55 |
"vision_config": {
|
| 56 |
"add_cross_attention": false,
|
| 57 |
-
"
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
9,
|
| 61 |
-
1
|
| 62 |
-
],
|
| 63 |
-
"dim_embed": [
|
| 64 |
-
128,
|
| 65 |
-
256,
|
| 66 |
-
512,
|
| 67 |
-
1024
|
| 68 |
-
],
|
| 69 |
"drop_path_rate": 0.1,
|
| 70 |
"image_feature_source": [
|
| 71 |
"spatial_avg_pool",
|
|
@@ -76,30 +67,12 @@
|
|
| 76 |
"type": "learned_abs_2d"
|
| 77 |
},
|
| 78 |
"model_type": "davit",
|
| 79 |
-
"num_groups": [
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
],
|
| 85 |
-
"num_heads": [
|
| 86 |
-
4,
|
| 87 |
-
8,
|
| 88 |
-
16,
|
| 89 |
-
32
|
| 90 |
-
],
|
| 91 |
-
"patch_size": [
|
| 92 |
-
7,
|
| 93 |
-
3,
|
| 94 |
-
3,
|
| 95 |
-
3
|
| 96 |
-
],
|
| 97 |
-
"patch_stride": [
|
| 98 |
-
4,
|
| 99 |
-
2,
|
| 100 |
-
2,
|
| 101 |
-
2
|
| 102 |
-
],
|
| 103 |
"visual_temporal_embedding": {
|
| 104 |
"max_temporal_embeddings": 100,
|
| 105 |
"type": "COSINE"
|
|
|
|
| 54 |
"transformers_version": "4.46.0.dev0",
|
| 55 |
"vision_config": {
|
| 56 |
"add_cross_attention": false,
|
| 57 |
+
"architectures": null,
|
| 58 |
+
"depths": [1, 1, 9, 1],
|
| 59 |
+
"dim_embed": [128, 256, 512, 1024],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
"drop_path_rate": 0.1,
|
| 61 |
"image_feature_source": [
|
| 62 |
"spatial_avg_pool",
|
|
|
|
| 67 |
"type": "learned_abs_2d"
|
| 68 |
},
|
| 69 |
"model_type": "davit",
|
| 70 |
+
"num_groups": [4, 8, 16, 32],
|
| 71 |
+
"num_heads": [4, 8, 16, 32],
|
| 72 |
+
"patch_padding": [3, 1, 1, 1],
|
| 73 |
+
"patch_prenorm": [false, true, true, true],
|
| 74 |
+
"patch_size": [7, 3, 3, 3],
|
| 75 |
+
"patch_stride": [4, 2, 2, 2],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 76 |
"visual_temporal_embedding": {
|
| 77 |
"max_temporal_embeddings": 100,
|
| 78 |
"type": "COSINE"
|