Upload UnifiedIOPreprocessing
Browse files- preprocessor_config.json +268 -0
preprocessor_config.json
ADDED
@@ -0,0 +1,268 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"config": {
|
3 |
+
"audio_history_cfg": {
|
4 |
+
"attn_qk_norm": true,
|
5 |
+
"attn_scaled_cosine": false,
|
6 |
+
"clip_attn_logit": null,
|
7 |
+
"dropout_broadcast_dims": [
|
8 |
+
-2
|
9 |
+
],
|
10 |
+
"dropout_rate": 0.0,
|
11 |
+
"droppath_rate": 0.0,
|
12 |
+
"dtype": "float32",
|
13 |
+
"emb_dim": 768,
|
14 |
+
"float32_attention_logits": true,
|
15 |
+
"head_dim": 64,
|
16 |
+
"latents_size": 16,
|
17 |
+
"layer_drop": 0.0,
|
18 |
+
"max_frames": 8,
|
19 |
+
"mlp_activations": [
|
20 |
+
"gelu"
|
21 |
+
],
|
22 |
+
"mlp_dim": 2048,
|
23 |
+
"num_heads": 12,
|
24 |
+
"num_layers": 2,
|
25 |
+
"resampler_type": "perceiver",
|
26 |
+
"xattention_index": [
|
27 |
+
0,
|
28 |
+
1
|
29 |
+
],
|
30 |
+
"xattn_qk_norm": true,
|
31 |
+
"xattn_scaled_cosine": false
|
32 |
+
},
|
33 |
+
"audio_vit_cfg": {
|
34 |
+
"default_input_size": [
|
35 |
+
256,
|
36 |
+
128
|
37 |
+
],
|
38 |
+
"dropout_broadcast_dims": [],
|
39 |
+
"dropout_rate": 0.0,
|
40 |
+
"dtype": "float32",
|
41 |
+
"emb_dim": 768,
|
42 |
+
"float32_attention_logits": true,
|
43 |
+
"head_dim": 64,
|
44 |
+
"mlp_activations": [
|
45 |
+
"gelu"
|
46 |
+
],
|
47 |
+
"mlp_dim": 3072,
|
48 |
+
"num_heads": 12,
|
49 |
+
"num_layers": 11,
|
50 |
+
"patch_size": 16,
|
51 |
+
"pos_patch_size": 16,
|
52 |
+
"transpose_input": true,
|
53 |
+
"vit_embed": true
|
54 |
+
},
|
55 |
+
"audio_vqgan": {
|
56 |
+
"act_fn": "relu",
|
57 |
+
"attention_dropout_rate": 0.0,
|
58 |
+
"checkpoint_path": "",
|
59 |
+
"decoder_head_dim": 64,
|
60 |
+
"decoder_hidden_size": 512,
|
61 |
+
"decoder_mlp_dim": 2048,
|
62 |
+
"decoder_num_heads": 8,
|
63 |
+
"decoder_num_layers": 8,
|
64 |
+
"default_input_size": [
|
65 |
+
128,
|
66 |
+
256
|
67 |
+
],
|
68 |
+
"dropout_rate": 0.0,
|
69 |
+
"droppath_rate": 0.0,
|
70 |
+
"dtype": "float32",
|
71 |
+
"encoder_head_dim": 64,
|
72 |
+
"encoder_hidden_size": 512,
|
73 |
+
"encoder_mlp_dim": 2048,
|
74 |
+
"encoder_num_heads": 8,
|
75 |
+
"encoder_num_layers": 8,
|
76 |
+
"output_channel": 1,
|
77 |
+
"patch_size": [
|
78 |
+
8,
|
79 |
+
8
|
80 |
+
],
|
81 |
+
"proj_dim": 32,
|
82 |
+
"use_bias": false,
|
83 |
+
"use_decoder": true,
|
84 |
+
"vocab_size": 8192
|
85 |
+
},
|
86 |
+
"freeze_vit": true,
|
87 |
+
"image_history_cfg": {
|
88 |
+
"attn_qk_norm": true,
|
89 |
+
"attn_scaled_cosine": false,
|
90 |
+
"clip_attn_logit": null,
|
91 |
+
"dropout_broadcast_dims": [
|
92 |
+
-2
|
93 |
+
],
|
94 |
+
"dropout_rate": 0.0,
|
95 |
+
"droppath_rate": 0.0,
|
96 |
+
"dtype": "float32",
|
97 |
+
"emb_dim": 768,
|
98 |
+
"float32_attention_logits": true,
|
99 |
+
"head_dim": 64,
|
100 |
+
"latents_size": 32,
|
101 |
+
"layer_drop": 0.0,
|
102 |
+
"max_frames": 8,
|
103 |
+
"mlp_activations": [
|
104 |
+
"gelu"
|
105 |
+
],
|
106 |
+
"mlp_dim": 2048,
|
107 |
+
"num_heads": 12,
|
108 |
+
"num_layers": 2,
|
109 |
+
"resampler_type": "perceiver",
|
110 |
+
"xattention_index": [
|
111 |
+
0,
|
112 |
+
1
|
113 |
+
],
|
114 |
+
"xattn_qk_norm": true,
|
115 |
+
"xattn_scaled_cosine": false
|
116 |
+
},
|
117 |
+
"image_vit_cfg": {
|
118 |
+
"default_input_size": [
|
119 |
+
256,
|
120 |
+
256
|
121 |
+
],
|
122 |
+
"dropout_broadcast_dims": [],
|
123 |
+
"dropout_rate": 0.0,
|
124 |
+
"dtype": "float32",
|
125 |
+
"emb_dim": 768,
|
126 |
+
"float32_attention_logits": true,
|
127 |
+
"head_dim": 64,
|
128 |
+
"mlp_activations": [
|
129 |
+
"gelu"
|
130 |
+
],
|
131 |
+
"mlp_dim": 3072,
|
132 |
+
"num_heads": 12,
|
133 |
+
"num_layers": 11,
|
134 |
+
"num_pos": 197,
|
135 |
+
"patch_size": 16,
|
136 |
+
"pos_patch_size": 16
|
137 |
+
},
|
138 |
+
"image_vqgan": {
|
139 |
+
"attn_resolutions": [
|
140 |
+
32
|
141 |
+
],
|
142 |
+
"ch": 128,
|
143 |
+
"ch_mult": [
|
144 |
+
1,
|
145 |
+
2,
|
146 |
+
2,
|
147 |
+
4
|
148 |
+
],
|
149 |
+
"checkpoint_path": "",
|
150 |
+
"default_input_size": [
|
151 |
+
256,
|
152 |
+
256
|
153 |
+
],
|
154 |
+
"double_z": false,
|
155 |
+
"dropout": 0,
|
156 |
+
"dtype": "float32",
|
157 |
+
"embed_dim": 4,
|
158 |
+
"in_channels": 3,
|
159 |
+
"n_embed": 16384,
|
160 |
+
"num_res_blocks": 2,
|
161 |
+
"out_ch": 3,
|
162 |
+
"patch_size": [
|
163 |
+
8,
|
164 |
+
8
|
165 |
+
],
|
166 |
+
"resolution": 256,
|
167 |
+
"z_channels": 4
|
168 |
+
},
|
169 |
+
"input_modalities": [
|
170 |
+
"text",
|
171 |
+
"image",
|
172 |
+
"image_history",
|
173 |
+
"audio",
|
174 |
+
"audio_history"
|
175 |
+
],
|
176 |
+
"sequence_length": {
|
177 |
+
"audio_history_input_samples": 128,
|
178 |
+
"audio_input_samples": 128,
|
179 |
+
"image_history_input_samples": 256,
|
180 |
+
"image_input_samples": 576,
|
181 |
+
"is_training": true,
|
182 |
+
"num_frames": 4
|
183 |
+
},
|
184 |
+
"t5_config": {
|
185 |
+
"audio_history_pos_emb": "llama_rope",
|
186 |
+
"audio_patch_size": 16,
|
187 |
+
"audio_pos_emb": "llama_rope",
|
188 |
+
"audio_vit_patch_size": 16,
|
189 |
+
"audio_vocab_size": 8320,
|
190 |
+
"dalle_attn_mask": true,
|
191 |
+
"decoder_max_audio_length": 512,
|
192 |
+
"decoder_max_image_length": 1024,
|
193 |
+
"decoder_max_text_length": 512,
|
194 |
+
"decoder_xattention_internval": 1,
|
195 |
+
"default_audio_history_vit_size": [
|
196 |
+
256,
|
197 |
+
128
|
198 |
+
],
|
199 |
+
"default_audio_size": [
|
200 |
+
256,
|
201 |
+
128
|
202 |
+
],
|
203 |
+
"default_audio_vit_size": [
|
204 |
+
256,
|
205 |
+
128
|
206 |
+
],
|
207 |
+
"default_image_history_vit_size": [
|
208 |
+
256,
|
209 |
+
256
|
210 |
+
],
|
211 |
+
"default_image_size": [
|
212 |
+
256,
|
213 |
+
256
|
214 |
+
],
|
215 |
+
"default_image_vit_size": [
|
216 |
+
384,
|
217 |
+
384
|
218 |
+
],
|
219 |
+
"dropout_broadcast_dims": [
|
220 |
+
-2
|
221 |
+
],
|
222 |
+
"dropout_rate": 0.0,
|
223 |
+
"dtype": "float32",
|
224 |
+
"dynamic_unk_mask": true,
|
225 |
+
"emb_dim": 1024,
|
226 |
+
"encoder_max_audio_length": 128,
|
227 |
+
"encoder_max_image_length": 576,
|
228 |
+
"encoder_max_text_length": 512,
|
229 |
+
"float32_attention_logits": true,
|
230 |
+
"head_dim": 64,
|
231 |
+
"image_history_pos_emb": "llama_rope",
|
232 |
+
"image_patch_size": 16,
|
233 |
+
"image_pos_emb": "llama_rope",
|
234 |
+
"image_tokenizer_type": "vqgan",
|
235 |
+
"image_vit_patch_size": 16,
|
236 |
+
"image_vocab_size": 16512,
|
237 |
+
"logits_via_embedding": true,
|
238 |
+
"mlp_activations": [
|
239 |
+
"silu",
|
240 |
+
"linear"
|
241 |
+
],
|
242 |
+
"mlp_dim": 2816,
|
243 |
+
"num_decoder_layers": 24,
|
244 |
+
"num_encoder_layers": 24,
|
245 |
+
"num_heads": 16,
|
246 |
+
"qk_norm": true,
|
247 |
+
"text_pos_emb": "llama_rope",
|
248 |
+
"vocab_size": 33280
|
249 |
+
},
|
250 |
+
"target_modalities": [
|
251 |
+
"text",
|
252 |
+
"image",
|
253 |
+
"audio"
|
254 |
+
],
|
255 |
+
"use_audio_history_vit": true,
|
256 |
+
"use_audio_vit": true,
|
257 |
+
"use_image_history_vit": true,
|
258 |
+
"use_image_vit": true
|
259 |
+
},
|
260 |
+
"sequence_length": {
|
261 |
+
"audio_history_input_samples": 128,
|
262 |
+
"audio_input_samples": 128,
|
263 |
+
"image_history_input_samples": 256,
|
264 |
+
"image_input_samples": 576,
|
265 |
+
"is_training": true,
|
266 |
+
"num_frames": 4
|
267 |
+
}
|
268 |
+
}
|