Adding safetensor for HF transformer support
Browse filesSigned-off-by: taejinp <[email protected]>
- config.json +79 -0
- model.safetensors +3 -0
- processor_config.json +16 -0
config.json
ADDED
|
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"SortformerOffline"
|
| 4 |
+
],
|
| 5 |
+
"ats_weight": 0.5,
|
| 6 |
+
"dtype": "float32",
|
| 7 |
+
"fc_encoder_config": {
|
| 8 |
+
"activation_dropout": 0.1,
|
| 9 |
+
"attention_bias": true,
|
| 10 |
+
"attention_dropout": 0.1,
|
| 11 |
+
"conv_kernel_size": 9,
|
| 12 |
+
"dropout": 0.1,
|
| 13 |
+
"dropout_positions": 0.0,
|
| 14 |
+
"hidden_act": "silu",
|
| 15 |
+
"hidden_size": 512,
|
| 16 |
+
"initializer_range": 0.02,
|
| 17 |
+
"intermediate_size": 2048,
|
| 18 |
+
"layerdrop": 0.1,
|
| 19 |
+
"max_position_embeddings": 5000,
|
| 20 |
+
"model_type": "sortformer_fc_encoder",
|
| 21 |
+
"num_attention_heads": 8,
|
| 22 |
+
"num_hidden_layers": 18,
|
| 23 |
+
"num_key_value_heads": 8,
|
| 24 |
+
"num_mel_bins": 80,
|
| 25 |
+
"scale_input": true,
|
| 26 |
+
"subsampling_conv_channels": 256,
|
| 27 |
+
"subsampling_conv_kernel_size": 3,
|
| 28 |
+
"subsampling_conv_stride": 2,
|
| 29 |
+
"subsampling_factor": 8
|
| 30 |
+
},
|
| 31 |
+
"initializer_range": 0.02,
|
| 32 |
+
"model_type": "sortformer",
|
| 33 |
+
"modules_config": {
|
| 34 |
+
"causal_attn_rate": 0.5,
|
| 35 |
+
"causal_attn_rc": 30,
|
| 36 |
+
"chunk_left_context": 1,
|
| 37 |
+
"chunk_len": 188,
|
| 38 |
+
"chunk_right_context": 1,
|
| 39 |
+
"dropout_rate": 0.5,
|
| 40 |
+
"fc_d_model": 512,
|
| 41 |
+
"fifo_len": 0,
|
| 42 |
+
"max_index": 10000,
|
| 43 |
+
"min_pos_scores_rate": 0.5,
|
| 44 |
+
"model_type": "sortformer_modules",
|
| 45 |
+
"num_speakers": 4,
|
| 46 |
+
"pred_score_threshold": 1e-06,
|
| 47 |
+
"scores_add_rnd": 2.0,
|
| 48 |
+
"scores_boost_latest": 0.5,
|
| 49 |
+
"sil_threshold": 0.1,
|
| 50 |
+
"spkcache_len": 188,
|
| 51 |
+
"spkcache_sil_frames_per_spk": 5,
|
| 52 |
+
"spkcache_update_period": 188,
|
| 53 |
+
"strong_boost_rate": 0.3,
|
| 54 |
+
"subsampling_factor": 8,
|
| 55 |
+
"tf_d_model": 192,
|
| 56 |
+
"weak_boost_rate": 0.7
|
| 57 |
+
},
|
| 58 |
+
"num_speakers": 4,
|
| 59 |
+
"pil_weight": 0.5,
|
| 60 |
+
"tf_encoder_config": {
|
| 61 |
+
"activation_dropout": 0.5,
|
| 62 |
+
"activation_function": "relu",
|
| 63 |
+
"attention_dropout": 0.5,
|
| 64 |
+
"d_model": 192,
|
| 65 |
+
"dropout": 0.5,
|
| 66 |
+
"encoder_attention_heads": 8,
|
| 67 |
+
"encoder_ffn_dim": 768,
|
| 68 |
+
"encoder_layerdrop": 0.5,
|
| 69 |
+
"encoder_layers": 18,
|
| 70 |
+
"init_std": 0.02,
|
| 71 |
+
"initializer_range": 0.02,
|
| 72 |
+
"layer_norm_eps": 1e-05,
|
| 73 |
+
"max_source_positions": 1500,
|
| 74 |
+
"model_type": "sortformer_tf_encoder",
|
| 75 |
+
"num_mel_bins": 80,
|
| 76 |
+
"scale_embedding": false
|
| 77 |
+
},
|
| 78 |
+
"transformers_version": "5.0.0.dev0"
|
| 79 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e8abcc5f3a82ff23134c98a37f70fef3f159611f394bb191a0ad0a6f4b052974
|
| 3 |
+
size 494206256
|
processor_config.json
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"feature_extractor": {
|
| 3 |
+
"feature_extractor_type": "SortformerFeatureExtractor",
|
| 4 |
+
"feature_size": 80,
|
| 5 |
+
"hop_length": 160,
|
| 6 |
+
"n_fft": 512,
|
| 7 |
+
"padding_side": "right",
|
| 8 |
+
"padding_value": 0.0,
|
| 9 |
+
"preemphasis": 0.97,
|
| 10 |
+
"processor_class": "SortformerProcessor",
|
| 11 |
+
"return_attention_mask": true,
|
| 12 |
+
"sampling_rate": 16000,
|
| 13 |
+
"win_length": 400
|
| 14 |
+
},
|
| 15 |
+
"processor_class": "SortformerProcessor"
|
| 16 |
+
}
|