arvisioncode
commited on
Commit
•
d142326
1
Parent(s):
72763f4
Update config.json
Browse files- config.json +13 -13
config.json
CHANGED
@@ -12,7 +12,7 @@
|
|
12 |
"ignore_index": -100,
|
13 |
"model_type": "florence2",
|
14 |
"pad_token_id": 1,
|
15 |
-
"projection_dim":
|
16 |
"text_config": {
|
17 |
"vocab_size": 51289,
|
18 |
"activation_dropout": 0.1,
|
@@ -23,18 +23,18 @@
|
|
23 |
"bos_token_id": 0,
|
24 |
"classif_dropout": 0.1,
|
25 |
"classifier_dropout": 0.0,
|
26 |
-
"d_model":
|
27 |
-
"decoder_attention_heads":
|
28 |
-
"decoder_ffn_dim":
|
29 |
"decoder_layerdrop": 0.0,
|
30 |
-
"decoder_layers":
|
31 |
"decoder_start_token_id": 2,
|
32 |
"dropout": 0.1,
|
33 |
"early_stopping": true,
|
34 |
-
"encoder_attention_heads":
|
35 |
-
"encoder_ffn_dim":
|
36 |
"encoder_layerdrop": 0.0,
|
37 |
-
"encoder_layers":
|
38 |
"eos_token_id": 2,
|
39 |
"forced_eos_token_id": 2,
|
40 |
"forced_bos_token_id": 0,
|
@@ -49,7 +49,7 @@
|
|
49 |
"max_position_embeddings": 1024,
|
50 |
"no_repeat_ngram_size": 3,
|
51 |
"normalize_before": false,
|
52 |
-
"num_hidden_layers":
|
53 |
"pad_token_id": 1,
|
54 |
"scale_embedding": false,
|
55 |
"num_beams": 3
|
@@ -62,12 +62,12 @@
|
|
62 |
"patch_padding": [3, 1, 1, 1],
|
63 |
"patch_prenorm": [false, true, true, true],
|
64 |
"enable_checkpoint": false,
|
65 |
-
"dim_embed": [
|
66 |
-
"num_heads": [
|
67 |
-
"num_groups": [
|
68 |
"depths": [1, 1, 9, 1],
|
69 |
"window_size": 12,
|
70 |
-
"projection_dim":
|
71 |
"visual_temporal_embedding": {
|
72 |
"type": "COSINE",
|
73 |
"max_temporal_embeddings": 100
|
|
|
12 |
"ignore_index": -100,
|
13 |
"model_type": "florence2",
|
14 |
"pad_token_id": 1,
|
15 |
+
"projection_dim": 1024,
|
16 |
"text_config": {
|
17 |
"vocab_size": 51289,
|
18 |
"activation_dropout": 0.1,
|
|
|
23 |
"bos_token_id": 0,
|
24 |
"classif_dropout": 0.1,
|
25 |
"classifier_dropout": 0.0,
|
26 |
+
"d_model": 1024,
|
27 |
+
"decoder_attention_heads": 16,
|
28 |
+
"decoder_ffn_dim": 4096,
|
29 |
"decoder_layerdrop": 0.0,
|
30 |
+
"decoder_layers": 12,
|
31 |
"decoder_start_token_id": 2,
|
32 |
"dropout": 0.1,
|
33 |
"early_stopping": true,
|
34 |
+
"encoder_attention_heads": 16,
|
35 |
+
"encoder_ffn_dim": 4096,
|
36 |
"encoder_layerdrop": 0.0,
|
37 |
+
"encoder_layers": 12,
|
38 |
"eos_token_id": 2,
|
39 |
"forced_eos_token_id": 2,
|
40 |
"forced_bos_token_id": 0,
|
|
|
49 |
"max_position_embeddings": 1024,
|
50 |
"no_repeat_ngram_size": 3,
|
51 |
"normalize_before": false,
|
52 |
+
"num_hidden_layers": 12,
|
53 |
"pad_token_id": 1,
|
54 |
"scale_embedding": false,
|
55 |
"num_beams": 3
|
|
|
62 |
"patch_padding": [3, 1, 1, 1],
|
63 |
"patch_prenorm": [false, true, true, true],
|
64 |
"enable_checkpoint": false,
|
65 |
+
"dim_embed": [256, 512, 1024, 2048],
|
66 |
+
"num_heads": [8, 16, 32, 64],
|
67 |
+
"num_groups": [8, 16, 32, 64],
|
68 |
"depths": [1, 1, 9, 1],
|
69 |
"window_size": 12,
|
70 |
+
"projection_dim": 1024,
|
71 |
"visual_temporal_embedding": {
|
72 |
"type": "COSINE",
|
73 |
"max_temporal_embeddings": 100
|