Update tiny models for NemotronAsrStreamingEncoder
Browse files- config.json +32 -0
- model.safetensors +3 -0
- preprocessor_config.json +12 -0
- tokenizer.json +0 -0
- tokenizer_config.json +11 -0
config.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation_dropout": 0.0,
|
| 3 |
+
"architectures": [
|
| 4 |
+
"NemotronAsrStreamingEncoder"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": true,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"conv_kernel_size": 9,
|
| 9 |
+
"convolution_bias": true,
|
| 10 |
+
"default_num_lookahead_tokens": 13,
|
| 11 |
+
"dropout": 0.0,
|
| 12 |
+
"dropout_positions": 0.0,
|
| 13 |
+
"dtype": "float32",
|
| 14 |
+
"hidden_act": "silu",
|
| 15 |
+
"hidden_size": 64,
|
| 16 |
+
"initializer_range": 0.02,
|
| 17 |
+
"intermediate_size": 256,
|
| 18 |
+
"layerdrop": 0.0,
|
| 19 |
+
"max_position_embeddings": 5000,
|
| 20 |
+
"model_type": "nemotron_asr_streaming_encoder",
|
| 21 |
+
"num_attention_heads": 4,
|
| 22 |
+
"num_hidden_layers": 2,
|
| 23 |
+
"num_key_value_heads": 4,
|
| 24 |
+
"num_mel_bins": 80,
|
| 25 |
+
"scale_input": true,
|
| 26 |
+
"sliding_window": 71,
|
| 27 |
+
"subsampling_conv_channels": 32,
|
| 28 |
+
"subsampling_conv_kernel_size": 3,
|
| 29 |
+
"subsampling_conv_stride": 2,
|
| 30 |
+
"subsampling_factor": 8,
|
| 31 |
+
"transformers_version": "5.16.0.dev0"
|
| 32 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:67f10d50ef549b00981b54eab45d89f1897d9220f2bf40f8b99cb58a87b899b9
|
| 3 |
+
size 918408
|
preprocessor_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"feature_extractor_type": "NemotronAsrStreamingFeatureExtractor",
|
| 3 |
+
"feature_size": 128,
|
| 4 |
+
"hop_length": 160,
|
| 5 |
+
"n_fft": 512,
|
| 6 |
+
"padding_side": "right",
|
| 7 |
+
"padding_value": 0.0,
|
| 8 |
+
"preemphasis": 0.97,
|
| 9 |
+
"return_attention_mask": true,
|
| 10 |
+
"sampling_rate": 16000,
|
| 11 |
+
"win_length": 400
|
| 12 |
+
}
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"backend": "tokenizers",
|
| 3 |
+
"clean_up_tokenization_spaces": false,
|
| 4 |
+
"is_local": true,
|
| 5 |
+
"local_files_only": false,
|
| 6 |
+
"model_max_length": 5000,
|
| 7 |
+
"pad_token": "<pad>",
|
| 8 |
+
"processor_class": "NemotronAsrStreamingProcessor",
|
| 9 |
+
"tokenizer_class": "ParakeetTokenizer",
|
| 10 |
+
"unk_token": "<unk>"
|
| 11 |
+
}
|