Upload config.json with huggingface_hub
Browse files- config.json +92 -0
config.json
ADDED
|
@@ -0,0 +1,92 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"in_channels": 768,
|
| 3 |
+
"audio_codec": {
|
| 4 |
+
"encoder_dim": 64,
|
| 5 |
+
"encoder_rates": [
|
| 6 |
+
2,
|
| 7 |
+
8,
|
| 8 |
+
10,
|
| 9 |
+
12
|
| 10 |
+
],
|
| 11 |
+
"latent_dim": 1024,
|
| 12 |
+
"decoder_dim": 1536,
|
| 13 |
+
"decoder_rates": [
|
| 14 |
+
12,
|
| 15 |
+
10,
|
| 16 |
+
8,
|
| 17 |
+
2
|
| 18 |
+
],
|
| 19 |
+
"n_codebooks": 16,
|
| 20 |
+
"codebook_size": 1024,
|
| 21 |
+
"codebook_dim": 128,
|
| 22 |
+
"quantizer_dropout": false,
|
| 23 |
+
"sample_rate": 48000,
|
| 24 |
+
"mean": 0.0,
|
| 25 |
+
"std": 1.0
|
| 26 |
+
},
|
| 27 |
+
"text_encoder": {
|
| 28 |
+
"dim": 768,
|
| 29 |
+
"name": "t5-base",
|
| 30 |
+
"max_length": 512,
|
| 31 |
+
"pad_mode": "longest"
|
| 32 |
+
},
|
| 33 |
+
"vision_encoder": {
|
| 34 |
+
"dim": 1024,
|
| 35 |
+
"batch_size": 300,
|
| 36 |
+
"name": "PE-Core-L14-336",
|
| 37 |
+
"normalize_feature": true,
|
| 38 |
+
"interpolation_mode": "BICUBIC",
|
| 39 |
+
"image_size": 336
|
| 40 |
+
},
|
| 41 |
+
"transformer": {
|
| 42 |
+
"dim": 2816,
|
| 43 |
+
"n_heads": 22,
|
| 44 |
+
"n_layers": 22,
|
| 45 |
+
"dropout": 0.1,
|
| 46 |
+
"norm_eps": 1e-05,
|
| 47 |
+
"qk_norm": true,
|
| 48 |
+
"fc_bias": false,
|
| 49 |
+
"ffn_exp": 4,
|
| 50 |
+
"ffn_dim_multiplier": 1,
|
| 51 |
+
"multiple_of": 64,
|
| 52 |
+
"non_linearity": "swiglu",
|
| 53 |
+
"use_rope": true,
|
| 54 |
+
"max_positions": 10000,
|
| 55 |
+
"frequency_embedding_dim": 256,
|
| 56 |
+
"timestep_non_linearity": "swiglu",
|
| 57 |
+
"t_block_non_linearity": "silu",
|
| 58 |
+
"t_block_bias": true,
|
| 59 |
+
"context_dim": 2816,
|
| 60 |
+
"context_non_linearity": "swiglu",
|
| 61 |
+
"context_embedder_dropout": 0.0,
|
| 62 |
+
"context_norm": false,
|
| 63 |
+
"out_channels": 256,
|
| 64 |
+
"in_channels": null
|
| 65 |
+
},
|
| 66 |
+
"num_anchors": 3,
|
| 67 |
+
"anchor_embedding_dim": 128,
|
| 68 |
+
"visual_ranker": {
|
| 69 |
+
"checkpoint": null,
|
| 70 |
+
"kind": "imagebind"
|
| 71 |
+
},
|
| 72 |
+
"text_ranker": {
|
| 73 |
+
"rankers": {
|
| 74 |
+
"clap": [
|
| 75 |
+
{
|
| 76 |
+
"checkpoint": null,
|
| 77 |
+
"kind": "clap"
|
| 78 |
+
},
|
| 79 |
+
5.0
|
| 80 |
+
],
|
| 81 |
+
"judge": [
|
| 82 |
+
{
|
| 83 |
+
"checkpoint_or_model_id": "facebook/sam-audio-judge",
|
| 84 |
+
"kind": "judge"
|
| 85 |
+
},
|
| 86 |
+
1.0
|
| 87 |
+
]
|
| 88 |
+
},
|
| 89 |
+
"kind": "ensemble"
|
| 90 |
+
},
|
| 91 |
+
"span_predictor": "pe-a-frame-large"
|
| 92 |
+
}
|