arjunanand13
commited on
Commit
•
42c212c
1
Parent(s):
316aed3
Create config.json
Browse files- config.json +68 -0
config.json
ADDED
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "microsoft/Florence-2-base-ft",
|
3 |
+
"architectures": ["Florence2ForConditionalGeneration"],
|
4 |
+
"auto_map": {
|
5 |
+
"AutoConfig": "configuration_florence2.Florence2Config",
|
6 |
+
"AutoModelForCausalLM": "modeling_florence2.Florence2ForConditionalGeneration"
|
7 |
+
},
|
8 |
+
"bos_token_id": 2,
|
9 |
+
"eos_token_id": 1,
|
10 |
+
"ignore_index": -100,
|
11 |
+
"is_encoder_decoder": true,
|
12 |
+
"model_type": "florence2",
|
13 |
+
"pad_token_id": 0,
|
14 |
+
"projection_dim": 768,
|
15 |
+
"vocab_size": 51289,
|
16 |
+
"torch_dtype": "float32", // Keep fine-tuned dtype if trained in float32
|
17 |
+
"text_config": {
|
18 |
+
"activation_dropout": 0.1,
|
19 |
+
"activation_function": "gelu",
|
20 |
+
"attention_dropout": 0.1,
|
21 |
+
"bos_token_id": 2,
|
22 |
+
"decoder_attention_heads": 12,
|
23 |
+
"decoder_ffn_dim": 3072,
|
24 |
+
"decoder_layers": 6,
|
25 |
+
"d_model": 768,
|
26 |
+
"encoder_attention_heads": 12,
|
27 |
+
"encoder_ffn_dim": 3072,
|
28 |
+
"encoder_layers": 6,
|
29 |
+
"eos_token_id": 1,
|
30 |
+
"forced_bos_token_id": 2,
|
31 |
+
"forced_eos_token_id": 1,
|
32 |
+
"num_beams": 3,
|
33 |
+
"no_repeat_ngram_size": 3,
|
34 |
+
"dropout": 0.1,
|
35 |
+
"label2id": {
|
36 |
+
"LABEL_0": 0,
|
37 |
+
"LABEL_1": 1,
|
38 |
+
"LABEL_2": 2
|
39 |
+
},
|
40 |
+
"max_position_embeddings": 1024,
|
41 |
+
"is_encoder_decoder": true,
|
42 |
+
"pad_token_id": 0
|
43 |
+
},
|
44 |
+
"vision_config": {
|
45 |
+
"model_type": "davit",
|
46 |
+
"drop_path_rate": 0.1,
|
47 |
+
"patch_size": [7, 3, 3, 3],
|
48 |
+
"patch_stride": [4, 2, 2, 2],
|
49 |
+
"patch_padding": [3, 1, 1, 1],
|
50 |
+
"patch_prenorm": [false, true, true, true],
|
51 |
+
"dim_embed": [128, 256, 512, 1024],
|
52 |
+
"num_heads": [4, 8, 16, 32],
|
53 |
+
"num_groups": [4, 8, 16, 32],
|
54 |
+
"depths": [1, 1, 9, 1],
|
55 |
+
"window_size": 12,
|
56 |
+
"projection_dim": 768,
|
57 |
+
"visual_temporal_embedding": {
|
58 |
+
"type": "COSINE",
|
59 |
+
"max_temporal_embeddings": 100
|
60 |
+
},
|
61 |
+
"image_pos_embed": {
|
62 |
+
"type": "learned_abs_2d",
|
63 |
+
"max_pos_embeddings": 50
|
64 |
+
},
|
65 |
+
"image_feature_source": ["spatial_avg_pool", "temporal_avg_pool"]
|
66 |
+
},
|
67 |
+
"transformers_version": "4.41.0.dev0"
|
68 |
+
}
|