Instructions to use Reza2kn/visualears-fastconformer-fa-full-ab-qat-w2-materialized with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- NeMo
How to use Reza2kn/visualears-fastconformer-fa-full-ab-qat-w2-materialized with NeMo:
import nemo.collections.asr as nemo_asr asr_model = nemo_asr.models.ASRModel.from_pretrained("Reza2kn/visualears-fastconformer-fa-full-ab-qat-w2-materialized") transcriptions = asr_model.transcribe(["file.wav"]) - Notebooks
- Google Colab
- Kaggle
Upload QAT selection summary
Browse files- qat_selection_summary.json +143 -0
qat_selection_summary.json
ADDED
|
@@ -0,0 +1,143 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"base_model": "/workspace/.hf_home/hub/models--Reza2kn--visualears-fastconformer-fa-full-ab/snapshots/5c5cca3f3df79ab2bde501a162ec4e4f81d5b4c9/fa_fastconformer_ab_final.nemo",
|
| 3 |
+
"resolved_model": "/workspace/.hf_home/hub/models--Reza2kn--visualears-fastconformer-fa-full-ab/snapshots/5c5cca3f3df79ab2bde501a162ec4e4f81d5b4c9/fa_fastconformer_ab_final.nemo",
|
| 4 |
+
"weight_bits": 2,
|
| 5 |
+
"activation_bits": 0,
|
| 6 |
+
"scale_mode": "learnable-tensor",
|
| 7 |
+
"stochastic_precision": true,
|
| 8 |
+
"stochastic_schedule": "log-linear",
|
| 9 |
+
"stochastic_binary_prob_min": 0.2,
|
| 10 |
+
"stochastic_binary_prob_max": 0.9,
|
| 11 |
+
"one_bit_cotraining": true,
|
| 12 |
+
"cotraining_teacher_bits": 2,
|
| 13 |
+
"cotraining_student_bits": 1,
|
| 14 |
+
"cotraining_teacher_loss_weight": 1.0,
|
| 15 |
+
"cotraining_student_loss_weight": 1.0,
|
| 16 |
+
"cotraining_kl_weight": 0.7,
|
| 17 |
+
"cotraining_stochastic_loss_weight": 0.5,
|
| 18 |
+
"cotraining_stochastic_kl_weight": 0.7,
|
| 19 |
+
"repr_distill_weight": 0.2,
|
| 20 |
+
"repr_distill_stochastic_weight": 0.1,
|
| 21 |
+
"repr_distill_layer": 2,
|
| 22 |
+
"repr_distill_max_frames": 96,
|
| 23 |
+
"repr_distill_temperature": 1.0,
|
| 24 |
+
"encoder_layer_indices": [
|
| 25 |
+
0,
|
| 26 |
+
1,
|
| 27 |
+
2
|
| 28 |
+
],
|
| 29 |
+
"include_attention": false,
|
| 30 |
+
"include_subsampling": false,
|
| 31 |
+
"exclude_name_regex": "(?:conv|convolution|subsampling|pre_encode|decoder|ctc_decoder)",
|
| 32 |
+
"num_qat_linear": 12,
|
| 33 |
+
"qat_linear_modules": [
|
| 34 |
+
{
|
| 35 |
+
"name": "encoder.layers.0.feed_forward1.linear1",
|
| 36 |
+
"layer_index": 0,
|
| 37 |
+
"in_features": 512,
|
| 38 |
+
"out_features": 2048,
|
| 39 |
+
"bias": true,
|
| 40 |
+
"reason": "selected",
|
| 41 |
+
"stochastic_binary_prob": 0.2
|
| 42 |
+
},
|
| 43 |
+
{
|
| 44 |
+
"name": "encoder.layers.0.feed_forward1.linear2",
|
| 45 |
+
"layer_index": 0,
|
| 46 |
+
"in_features": 2048,
|
| 47 |
+
"out_features": 512,
|
| 48 |
+
"bias": true,
|
| 49 |
+
"reason": "selected",
|
| 50 |
+
"stochastic_binary_prob": 0.2
|
| 51 |
+
},
|
| 52 |
+
{
|
| 53 |
+
"name": "encoder.layers.0.feed_forward2.linear1",
|
| 54 |
+
"layer_index": 0,
|
| 55 |
+
"in_features": 512,
|
| 56 |
+
"out_features": 2048,
|
| 57 |
+
"bias": true,
|
| 58 |
+
"reason": "selected",
|
| 59 |
+
"stochastic_binary_prob": 0.2
|
| 60 |
+
},
|
| 61 |
+
{
|
| 62 |
+
"name": "encoder.layers.0.feed_forward2.linear2",
|
| 63 |
+
"layer_index": 0,
|
| 64 |
+
"in_features": 2048,
|
| 65 |
+
"out_features": 512,
|
| 66 |
+
"bias": true,
|
| 67 |
+
"reason": "selected",
|
| 68 |
+
"stochastic_binary_prob": 0.2
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"name": "encoder.layers.1.feed_forward1.linear1",
|
| 72 |
+
"layer_index": 1,
|
| 73 |
+
"in_features": 512,
|
| 74 |
+
"out_features": 2048,
|
| 75 |
+
"bias": true,
|
| 76 |
+
"reason": "selected",
|
| 77 |
+
"stochastic_binary_prob": 0.4242640687119285
|
| 78 |
+
},
|
| 79 |
+
{
|
| 80 |
+
"name": "encoder.layers.1.feed_forward1.linear2",
|
| 81 |
+
"layer_index": 1,
|
| 82 |
+
"in_features": 2048,
|
| 83 |
+
"out_features": 512,
|
| 84 |
+
"bias": true,
|
| 85 |
+
"reason": "selected",
|
| 86 |
+
"stochastic_binary_prob": 0.4242640687119285
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"name": "encoder.layers.1.feed_forward2.linear1",
|
| 90 |
+
"layer_index": 1,
|
| 91 |
+
"in_features": 512,
|
| 92 |
+
"out_features": 2048,
|
| 93 |
+
"bias": true,
|
| 94 |
+
"reason": "selected",
|
| 95 |
+
"stochastic_binary_prob": 0.4242640687119285
|
| 96 |
+
},
|
| 97 |
+
{
|
| 98 |
+
"name": "encoder.layers.1.feed_forward2.linear2",
|
| 99 |
+
"layer_index": 1,
|
| 100 |
+
"in_features": 2048,
|
| 101 |
+
"out_features": 512,
|
| 102 |
+
"bias": true,
|
| 103 |
+
"reason": "selected",
|
| 104 |
+
"stochastic_binary_prob": 0.4242640687119285
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"name": "encoder.layers.2.feed_forward1.linear1",
|
| 108 |
+
"layer_index": 2,
|
| 109 |
+
"in_features": 512,
|
| 110 |
+
"out_features": 2048,
|
| 111 |
+
"bias": true,
|
| 112 |
+
"reason": "selected",
|
| 113 |
+
"stochastic_binary_prob": 0.9
|
| 114 |
+
},
|
| 115 |
+
{
|
| 116 |
+
"name": "encoder.layers.2.feed_forward1.linear2",
|
| 117 |
+
"layer_index": 2,
|
| 118 |
+
"in_features": 2048,
|
| 119 |
+
"out_features": 512,
|
| 120 |
+
"bias": true,
|
| 121 |
+
"reason": "selected",
|
| 122 |
+
"stochastic_binary_prob": 0.9
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"name": "encoder.layers.2.feed_forward2.linear1",
|
| 126 |
+
"layer_index": 2,
|
| 127 |
+
"in_features": 512,
|
| 128 |
+
"out_features": 2048,
|
| 129 |
+
"bias": true,
|
| 130 |
+
"reason": "selected",
|
| 131 |
+
"stochastic_binary_prob": 0.9
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"name": "encoder.layers.2.feed_forward2.linear2",
|
| 135 |
+
"layer_index": 2,
|
| 136 |
+
"in_features": 2048,
|
| 137 |
+
"out_features": 512,
|
| 138 |
+
"bias": true,
|
| 139 |
+
"reason": "selected",
|
| 140 |
+
"stochastic_binary_prob": 0.9
|
| 141 |
+
}
|
| 142 |
+
]
|
| 143 |
+
}
|