Andrewsab commited on
Commit
2cf18a9
·
verified ·
1 Parent(s): c66aafc

Voice Scribe mirror parakeet from FluidInference/parakeet-tdt-0.6b-v3-ov@dfd55eb6c85a

Browse files
README.md ADDED
@@ -0,0 +1,120 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ language:
4
+ - en
5
+ - es
6
+ - it
7
+ - fr
8
+ - de
9
+ - nl
10
+ - ru
11
+ - pl
12
+ - uk
13
+ - sk
14
+ - bg
15
+ - fi
16
+ - ro
17
+ - hr
18
+ - cs
19
+ - sv
20
+ - et
21
+ - hu
22
+ - lt
23
+ - da
24
+ - mt
25
+ - sl
26
+ - lv
27
+ - el
28
+ pipeline_tag: automatic-speech-recognition
29
+ thumbnail: null
30
+ tags:
31
+ - automatic-speech-recognition
32
+ - speech
33
+ - audio
34
+ - Transducer
35
+ - TDT
36
+ - FastConformer
37
+ - Conformer
38
+ - multilingual
39
+ - NeMo
40
+ - OpenVINO
41
+ base_model:
42
+ - nvidia/parakeet-tdt-1.1b
43
+ ---
44
+
45
+ # Parakeet TDT 1.1B V3 - OpenVINO
46
+
47
+ [![Discord](https://img.shields.io/badge/Discord-Join%20Chat-7289da.svg)](https://discord.gg/WNsvaCtmDe)
48
+ [![GitHub Repo stars](https://img.shields.io/github/stars/FluidInference/eddy?style=flat&logo=github)](https://github.com/FluidInference/eddy)
49
+
50
+ OpenVINO-optimized version of NVIDIA's Parakeet TDT 1.1B V3 model for high-performance multilingual automatic speech recognition on Intel NPUs and CPUs.
51
+
52
+ ## Benchmark Results
53
+
54
+ **Hardware**: Intel Core Ultra 7 155H (Meteor Lake) with Intel AI Boost NPU
55
+ **Software**: OpenVINO 2025.x
56
+
57
+ ### LibriSpeech test-clean (English)
58
+
59
+ | Metric | Value |
60
+ |--------|-------|
61
+ | **Average WER** | 3.7% |
62
+ | **Median WER** | 0.0% |
63
+ | **Average CER** | 1.9% |
64
+ | **RTFx (NPU)** | 25.7× |
65
+ | **RTFx (CPU)** | 5-8× |
66
+ | **Files processed** | 2,620 (5.4 hours) |
67
+
68
+ ### FLEURS Multilingual (24 Languages)
69
+
70
+ | Metric | Value |
71
+ |--------|-------|
72
+ | **Average WER** | 17.0% |
73
+ | **Average CER** | 5.4% |
74
+ | **Average RTFx** | 41.1× |
75
+ | **Total samples** | ~15,000+ |
76
+
77
+ **Best performing languages** (WER): Italian 4.3%, Spanish 5.4%, English 6.1%, German 7.4%, French 7.7%
78
+
79
+ See [BENCHMARK_RESULTS.md](https://github.com/FluidInference/eddy/blob/main/BENCHMARK_RESULTS.md) for complete per-language results.
80
+
81
+ ## Performance Comparison
82
+
83
+ | Implementation | Device | RTFx (Avg) | WER (LibriSpeech) |
84
+ |----------------|--------|------------|-------------------|
85
+ | **eddy (OpenVINO)** | Intel Core Ultra 7 155H NPU | **25.7×** | 3.7% |
86
+ | Parakeet (PyTorch) | Intel Arc 140V GPU | ~20×* | ~2.5%* |
87
+ | **eddy (OpenVINO)** | Intel Core Ultra 7 155H CPU | **5-8×** | 3.7% |
88
+
89
+ > **Note**: Benchmarked on HP EliteBook Ultra G1i. eddy NPU is ~1.3× faster than PyTorch on Intel Arc GPU, with lower power consumption. *V3 estimated from V2 benchmark.
90
+
91
+ ## Supported Languages
92
+
93
+ **24 European languages**: English, Spanish, Italian, French, German, Dutch, Russian, Polish, Ukrainian, Slovak, Bulgarian, Finnish, Romanian, Croatian, Czech, Swedish, Estonian, Hungarian, Lithuanian, Danish, Maltese, Slovenian, Latvian, Greek
94
+
95
+ ## Usage
96
+
97
+ Python usage via ctypes available - see [eddy repository](https://github.com/FluidInference/eddy) for details.
98
+
99
+ ## Model Details
100
+
101
+ - **Parameters**: 1.1B
102
+ - **Architecture**: FastConformer-RNNT (4-model pipeline)
103
+ - **Languages**: 24 European languages
104
+ - **Blank token ID**: 8192
105
+ - **Context window**: 10s chunks with 3s overlap
106
+ - **Features**: LSTM state continuity, token deduplication, per-token timestamps
107
+
108
+ ## License
109
+
110
+ CC-BY-4.0 - See [LICENSE](LICENSE) for details.
111
+
112
+ ## Links
113
+
114
+ - **GitHub**: [FluidInference/eddy](https://github.com/FluidInference/eddy)
115
+ - **Base Model**: [nvidia/parakeet-tdt-1.1b](https://huggingface.co/nvidia/parakeet-tdt-1.1b)
116
+ - **Documentation**: [Benchmark Results](https://github.com/FluidInference/eddy/blob/main/BENCHMARK_RESULTS.md)
117
+
118
+ ## Acknowledgments
119
+
120
+ Based on NVIDIA's Parakeet TDT model. OpenVINO conversion and optimization by the FluidInference team.
UPSTREAM_SOURCE.md ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Voice Scribe Model Mirror
2
+
3
+ This repository is a Voice Scribe distribution mirror. The model artifacts are
4
+ copied from the upstream repository and the source revision below is pinned.
5
+
6
+ | Field | Value |
7
+ | --- | --- |
8
+ | Layout key | `parakeet` |
9
+ | Target directory in installer | `parakeet-v3-ov` |
10
+ | Upstream repo | `FluidInference/parakeet-tdt-0.6b-v3-ov` |
11
+ | Upstream revision | `dfd55eb6c85a9a8546a162bed84784245d5743c2` |
12
+ | Upstream resolved SHA | `dfd55eb6c85a9a8546a162bed84784245d5743c2` |
13
+ | Mirror created | `2026-04-23T22:20:54Z` |
14
+ | Description | Parakeet-TDT 0.6B v3 Intel/OpenVINO layout. |
15
+ | License metadata | `{"license": "cc-by-4.0", "license_files": [], "license_tags": ["license:cc-by-4.0"]}` |
16
+
17
+ ## Installer Contract
18
+
19
+ This mirror corresponds to `parakeet/installer/wrapper/model_catalog.py`.
20
+ Required files for installer validation:
21
+
22
+ ```json
23
+ [
24
+ "config.json",
25
+ "parakeet_encoder.xml",
26
+ "parakeet_encoder.bin",
27
+ "parakeet_decoder.xml",
28
+ "parakeet_decoder.bin",
29
+ "parakeet_joint.xml",
30
+ "parakeet_joint.bin",
31
+ "parakeet_melspectogram.xml",
32
+ "parakeet_melspectogram.bin",
33
+ "parakeet_vocab.json"
34
+ ]
35
+ ```
36
+
37
+ Allowed installer subset patterns:
38
+
39
+ ```json
40
+ []
41
+ ```
42
+
43
+ ## Redistribution Note
44
+
45
+ Do not make this repository public unless the upstream license and model card
46
+ allow redistribution for the intended use. Private mirrors are for operational
47
+ distribution convenience and reproducible installs.
config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {}
parakeet_decoder.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d32d703eed045f38564362b3f07a2405ff274b192001b9088d363b7b832edf8d
3
+ size 23604560
parakeet_decoder.xml ADDED
@@ -0,0 +1,1224 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="main_graph" version="11">
3
+ <layers>
4
+ <layer id="2" name="targets" type="Parameter" version="opset1">
5
+ <data shape="1,1" element_type="i64" />
6
+ <rt_info>
7
+ <attribute name="old_api_map_element_type" version="0" value="i32" />
8
+ </rt_info>
9
+ <output>
10
+ <port id="0" precision="I64" names="targets">
11
+ <dim>1</dim>
12
+ <dim>1</dim>
13
+ </port>
14
+ </output>
15
+ </layer>
16
+ <layer id="1" name="h_in" type="Parameter" version="opset1">
17
+ <data shape="2,1,640" element_type="f32" />
18
+ <output>
19
+ <port id="0" precision="FP32" names="h_in">
20
+ <dim>2</dim>
21
+ <dim>1</dim>
22
+ <dim>640</dim>
23
+ </port>
24
+ </output>
25
+ </layer>
26
+ <layer id="0" name="c_in" type="Parameter" version="opset1">
27
+ <data shape="2,1,640" element_type="f32" />
28
+ <output>
29
+ <port id="0" precision="FP32" names="c_in">
30
+ <dim>2</dim>
31
+ <dim>1</dim>
32
+ <dim>640</dim>
33
+ </port>
34
+ </output>
35
+ </layer>
36
+ <layer id="3" name="embed.weight_compressed" type="Const" version="opset1">
37
+ <data element_type="f16" shape="8193, 640" offset="0" size="10487040" />
38
+ <output>
39
+ <port id="0" precision="FP16" names="embed.weight">
40
+ <dim>8193</dim>
41
+ <dim>640</dim>
42
+ </port>
43
+ </output>
44
+ </layer>
45
+ <layer id="4" name="embed.weight" type="Convert" version="opset1">
46
+ <data destination_type="f32" />
47
+ <rt_info>
48
+ <attribute name="decompression" version="0" />
49
+ </rt_info>
50
+ <input>
51
+ <port id="0" precision="FP16">
52
+ <dim>8193</dim>
53
+ <dim>640</dim>
54
+ </port>
55
+ </input>
56
+ <output>
57
+ <port id="1" precision="FP32">
58
+ <dim>8193</dim>
59
+ <dim>640</dim>
60
+ </port>
61
+ </output>
62
+ </layer>
63
+ <layer id="5" name="Constant_29103" type="Const" version="opset1">
64
+ <data element_type="i64" shape="" offset="10487040" size="8" />
65
+ <output>
66
+ <port id="0" precision="I64" />
67
+ </output>
68
+ </layer>
69
+ <layer id="6" name="/embed/Gather" type="Gather" version="opset8">
70
+ <data batch_dims="0" />
71
+ <input>
72
+ <port id="0" precision="FP32">
73
+ <dim>8193</dim>
74
+ <dim>640</dim>
75
+ </port>
76
+ <port id="1" precision="I64">
77
+ <dim>1</dim>
78
+ <dim>1</dim>
79
+ </port>
80
+ <port id="2" precision="I64" />
81
+ </input>
82
+ <output>
83
+ <port id="3" precision="FP32" names="/embed/Gather_output_0">
84
+ <dim>1</dim>
85
+ <dim>1</dim>
86
+ <dim>640</dim>
87
+ </port>
88
+ </output>
89
+ </layer>
90
+ <layer id="7" name="Constant_29126" type="Const" version="opset1">
91
+ <data element_type="i64" shape="3" offset="10487048" size="24" />
92
+ <output>
93
+ <port id="0" precision="I64">
94
+ <dim>3</dim>
95
+ </port>
96
+ </output>
97
+ </layer>
98
+ <layer id="8" name="Transpose_29127" type="Transpose" version="opset1">
99
+ <input>
100
+ <port id="0" precision="FP32">
101
+ <dim>1</dim>
102
+ <dim>1</dim>
103
+ <dim>640</dim>
104
+ </port>
105
+ <port id="1" precision="I64">
106
+ <dim>3</dim>
107
+ </port>
108
+ </input>
109
+ <output>
110
+ <port id="2" precision="FP32">
111
+ <dim>1</dim>
112
+ <dim>1</dim>
113
+ <dim>640</dim>
114
+ </port>
115
+ </output>
116
+ </layer>
117
+ <layer id="9" name="/rnn/Constant_1" type="Const" version="opset1">
118
+ <data element_type="i64" shape="1" offset="10487040" size="8" />
119
+ <output>
120
+ <port id="0" precision="I64" names="/rnn/Constant_1_output_0">
121
+ <dim>1</dim>
122
+ </port>
123
+ </output>
124
+ </layer>
125
+ <layer id="10" name="/rnn/Constant_2" type="Const" version="opset1">
126
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
127
+ <output>
128
+ <port id="0" precision="I64" names="/rnn/Constant_2_output_0">
129
+ <dim>1</dim>
130
+ </port>
131
+ </output>
132
+ </layer>
133
+ <layer id="11" name="Broadcast_29110" type="Const" version="opset1">
134
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
135
+ <output>
136
+ <port id="0" precision="I64">
137
+ <dim>1</dim>
138
+ </port>
139
+ </output>
140
+ </layer>
141
+ <layer id="12" name="/rnn/Constant" type="Const" version="opset1">
142
+ <data element_type="i64" shape="1" offset="10487040" size="8" />
143
+ <output>
144
+ <port id="0" precision="I64" names="/rnn/Constant_output_0">
145
+ <dim>1</dim>
146
+ </port>
147
+ </output>
148
+ </layer>
149
+ <layer id="13" name="/rnn/Slice" type="Slice" version="opset8">
150
+ <input>
151
+ <port id="0" precision="FP32">
152
+ <dim>2</dim>
153
+ <dim>1</dim>
154
+ <dim>640</dim>
155
+ </port>
156
+ <port id="1" precision="I64">
157
+ <dim>1</dim>
158
+ </port>
159
+ <port id="2" precision="I64">
160
+ <dim>1</dim>
161
+ </port>
162
+ <port id="3" precision="I64">
163
+ <dim>1</dim>
164
+ </port>
165
+ <port id="4" precision="I64">
166
+ <dim>1</dim>
167
+ </port>
168
+ </input>
169
+ <output>
170
+ <port id="5" precision="FP32" names="/rnn/Slice_output_0">
171
+ <dim>1</dim>
172
+ <dim>1</dim>
173
+ <dim>640</dim>
174
+ </port>
175
+ </output>
176
+ </layer>
177
+ <layer id="14" name="Constant_29154" type="Const" version="opset1">
178
+ <data element_type="i64" shape="3" offset="10487048" size="24" />
179
+ <output>
180
+ <port id="0" precision="I64">
181
+ <dim>3</dim>
182
+ </port>
183
+ </output>
184
+ </layer>
185
+ <layer id="15" name="Transpose_29155" type="Transpose" version="opset1">
186
+ <input>
187
+ <port id="0" precision="FP32">
188
+ <dim>1</dim>
189
+ <dim>1</dim>
190
+ <dim>640</dim>
191
+ </port>
192
+ <port id="1" precision="I64">
193
+ <dim>3</dim>
194
+ </port>
195
+ </input>
196
+ <output>
197
+ <port id="2" precision="FP32">
198
+ <dim>1</dim>
199
+ <dim>1</dim>
200
+ <dim>640</dim>
201
+ </port>
202
+ </output>
203
+ </layer>
204
+ <layer id="16" name="/rnn/Constant_4" type="Const" version="opset1">
205
+ <data element_type="i64" shape="1" offset="10487040" size="8" />
206
+ <output>
207
+ <port id="0" precision="I64" names="/rnn/Constant_4_output_0">
208
+ <dim>1</dim>
209
+ </port>
210
+ </output>
211
+ </layer>
212
+ <layer id="17" name="/rnn/Constant_5" type="Const" version="opset1">
213
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
214
+ <output>
215
+ <port id="0" precision="I64" names="/rnn/Constant_5_output_0">
216
+ <dim>1</dim>
217
+ </port>
218
+ </output>
219
+ </layer>
220
+ <layer id="18" name="Broadcast_29120" type="Const" version="opset1">
221
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
222
+ <output>
223
+ <port id="0" precision="I64">
224
+ <dim>1</dim>
225
+ </port>
226
+ </output>
227
+ </layer>
228
+ <layer id="19" name="/rnn/Constant_3" type="Const" version="opset1">
229
+ <data element_type="i64" shape="1" offset="10487040" size="8" />
230
+ <output>
231
+ <port id="0" precision="I64" names="/rnn/Constant_3_output_0">
232
+ <dim>1</dim>
233
+ </port>
234
+ </output>
235
+ </layer>
236
+ <layer id="20" name="/rnn/Slice_1" type="Slice" version="opset8">
237
+ <input>
238
+ <port id="0" precision="FP32">
239
+ <dim>2</dim>
240
+ <dim>1</dim>
241
+ <dim>640</dim>
242
+ </port>
243
+ <port id="1" precision="I64">
244
+ <dim>1</dim>
245
+ </port>
246
+ <port id="2" precision="I64">
247
+ <dim>1</dim>
248
+ </port>
249
+ <port id="3" precision="I64">
250
+ <dim>1</dim>
251
+ </port>
252
+ <port id="4" precision="I64">
253
+ <dim>1</dim>
254
+ </port>
255
+ </input>
256
+ <output>
257
+ <port id="5" precision="FP32" names="/rnn/Slice_1_output_0">
258
+ <dim>1</dim>
259
+ <dim>1</dim>
260
+ <dim>640</dim>
261
+ </port>
262
+ </output>
263
+ </layer>
264
+ <layer id="21" name="Constant_29156" type="Const" version="opset1">
265
+ <data element_type="i64" shape="3" offset="10487048" size="24" />
266
+ <output>
267
+ <port id="0" precision="I64">
268
+ <dim>3</dim>
269
+ </port>
270
+ </output>
271
+ </layer>
272
+ <layer id="22" name="Transpose_29157" type="Transpose" version="opset1">
273
+ <input>
274
+ <port id="0" precision="FP32">
275
+ <dim>1</dim>
276
+ <dim>1</dim>
277
+ <dim>640</dim>
278
+ </port>
279
+ <port id="1" precision="I64">
280
+ <dim>3</dim>
281
+ </port>
282
+ </input>
283
+ <output>
284
+ <port id="2" precision="FP32">
285
+ <dim>1</dim>
286
+ <dim>1</dim>
287
+ <dim>640</dim>
288
+ </port>
289
+ </output>
290
+ </layer>
291
+ <layer id="23" name="ShapeOf_29134" type="ShapeOf" version="opset3">
292
+ <data output_type="i64" />
293
+ <input>
294
+ <port id="0" precision="FP32">
295
+ <dim>1</dim>
296
+ <dim>1</dim>
297
+ <dim>640</dim>
298
+ </port>
299
+ </input>
300
+ <output>
301
+ <port id="1" precision="I64">
302
+ <dim>3</dim>
303
+ </port>
304
+ </output>
305
+ </layer>
306
+ <layer id="24" name="Constant_29138" type="Const" version="opset1">
307
+ <data element_type="i32" shape="1" offset="10487080" size="4" />
308
+ <output>
309
+ <port id="0" precision="I32">
310
+ <dim>1</dim>
311
+ </port>
312
+ </output>
313
+ </layer>
314
+ <layer id="25" name="Constant_29135" type="Const" version="opset1">
315
+ <data element_type="i32" shape="1" offset="10487084" size="4" />
316
+ <output>
317
+ <port id="0" precision="I32">
318
+ <dim>1</dim>
319
+ </port>
320
+ </output>
321
+ </layer>
322
+ <layer id="26" name="Gather_29139" type="Gather" version="opset8">
323
+ <data batch_dims="0" />
324
+ <input>
325
+ <port id="0" precision="I64">
326
+ <dim>3</dim>
327
+ </port>
328
+ <port id="1" precision="I32">
329
+ <dim>1</dim>
330
+ </port>
331
+ <port id="2" precision="I32">
332
+ <dim>1</dim>
333
+ </port>
334
+ </input>
335
+ <output>
336
+ <port id="3" precision="I64">
337
+ <dim>1</dim>
338
+ </port>
339
+ </output>
340
+ </layer>
341
+ <layer id="27" name="Constant_31328" type="Const" version="opset1">
342
+ <data element_type="i64" shape="1" offset="10487040" size="8" />
343
+ <rt_info>
344
+ <attribute name="precise" version="0" />
345
+ </rt_info>
346
+ <output>
347
+ <port id="0" precision="I64">
348
+ <dim>1</dim>
349
+ </port>
350
+ </output>
351
+ </layer>
352
+ <layer id="28" name="Constant_31329" type="Const" version="opset1">
353
+ <data element_type="i64" shape="" offset="10487040" size="8" />
354
+ <rt_info>
355
+ <attribute name="precise" version="0" />
356
+ </rt_info>
357
+ <output>
358
+ <port id="0" precision="I64" />
359
+ </output>
360
+ </layer>
361
+ <layer id="29" name="Gather_31330" type="Gather" version="opset8">
362
+ <data batch_dims="0" />
363
+ <input>
364
+ <port id="0" precision="I64">
365
+ <dim>3</dim>
366
+ </port>
367
+ <port id="1" precision="I64">
368
+ <dim>1</dim>
369
+ </port>
370
+ <port id="2" precision="I64" />
371
+ </input>
372
+ <output>
373
+ <port id="3" precision="I64">
374
+ <dim>1</dim>
375
+ </port>
376
+ </output>
377
+ </layer>
378
+ <layer id="30" name="Broadcast_29151" type="Broadcast" version="opset3">
379
+ <data mode="numpy" />
380
+ <input>
381
+ <port id="0" precision="I64">
382
+ <dim>1</dim>
383
+ </port>
384
+ <port id="1" precision="I64">
385
+ <dim>1</dim>
386
+ </port>
387
+ </input>
388
+ <output>
389
+ <port id="2" precision="I64">
390
+ <dim>1</dim>
391
+ </port>
392
+ </output>
393
+ </layer>
394
+ <layer id="31" name="Concat_29130_compressed" type="Const" version="opset1">
395
+ <data element_type="f16" shape="1, 2560, 640" offset="10487088" size="3276800" />
396
+ <output>
397
+ <port id="0" precision="FP16">
398
+ <dim>1</dim>
399
+ <dim>2560</dim>
400
+ <dim>640</dim>
401
+ </port>
402
+ </output>
403
+ </layer>
404
+ <layer id="32" name="Concat_29130" type="Convert" version="opset1">
405
+ <data destination_type="f32" />
406
+ <rt_info>
407
+ <attribute name="decompression" version="0" />
408
+ </rt_info>
409
+ <input>
410
+ <port id="0" precision="FP16">
411
+ <dim>1</dim>
412
+ <dim>2560</dim>
413
+ <dim>640</dim>
414
+ </port>
415
+ </input>
416
+ <output>
417
+ <port id="1" precision="FP32">
418
+ <dim>1</dim>
419
+ <dim>2560</dim>
420
+ <dim>640</dim>
421
+ </port>
422
+ </output>
423
+ </layer>
424
+ <layer id="33" name="Concat_29133_compressed" type="Const" version="opset1">
425
+ <data element_type="f16" shape="1, 2560, 640" offset="13763888" size="3276800" />
426
+ <output>
427
+ <port id="0" precision="FP16">
428
+ <dim>1</dim>
429
+ <dim>2560</dim>
430
+ <dim>640</dim>
431
+ </port>
432
+ </output>
433
+ </layer>
434
+ <layer id="34" name="Concat_29133" type="Convert" version="opset1">
435
+ <data destination_type="f32" />
436
+ <rt_info>
437
+ <attribute name="decompression" version="0" />
438
+ </rt_info>
439
+ <input>
440
+ <port id="0" precision="FP16">
441
+ <dim>1</dim>
442
+ <dim>2560</dim>
443
+ <dim>640</dim>
444
+ </port>
445
+ </input>
446
+ <output>
447
+ <port id="1" precision="FP32">
448
+ <dim>1</dim>
449
+ <dim>2560</dim>
450
+ <dim>640</dim>
451
+ </port>
452
+ </output>
453
+ </layer>
454
+ <layer id="35" name="Concat_29150_compressed" type="Const" version="opset1">
455
+ <data element_type="f16" shape="1, 2560" offset="17040688" size="5120" />
456
+ <output>
457
+ <port id="0" precision="FP16">
458
+ <dim>1</dim>
459
+ <dim>2560</dim>
460
+ </port>
461
+ </output>
462
+ </layer>
463
+ <layer id="36" name="Concat_29150" type="Convert" version="opset1">
464
+ <data destination_type="f32" />
465
+ <rt_info>
466
+ <attribute name="decompression" version="0" />
467
+ </rt_info>
468
+ <input>
469
+ <port id="0" precision="FP16">
470
+ <dim>1</dim>
471
+ <dim>2560</dim>
472
+ </port>
473
+ </input>
474
+ <output>
475
+ <port id="1" precision="FP32">
476
+ <dim>1</dim>
477
+ <dim>2560</dim>
478
+ </port>
479
+ </output>
480
+ </layer>
481
+ <layer id="37" name="LSTMSequence_29165" type="LSTMSequence" version="opset5">
482
+ <data direction="forward" hidden_size="640" activations="sigmoid, tanh, tanh" activations_alpha="" activations_beta="" clip="0" />
483
+ <input>
484
+ <port id="0" precision="FP32">
485
+ <dim>1</dim>
486
+ <dim>1</dim>
487
+ <dim>640</dim>
488
+ </port>
489
+ <port id="1" precision="FP32">
490
+ <dim>1</dim>
491
+ <dim>1</dim>
492
+ <dim>640</dim>
493
+ </port>
494
+ <port id="2" precision="FP32">
495
+ <dim>1</dim>
496
+ <dim>1</dim>
497
+ <dim>640</dim>
498
+ </port>
499
+ <port id="3" precision="I64">
500
+ <dim>1</dim>
501
+ </port>
502
+ <port id="4" precision="FP32">
503
+ <dim>1</dim>
504
+ <dim>2560</dim>
505
+ <dim>640</dim>
506
+ </port>
507
+ <port id="5" precision="FP32">
508
+ <dim>1</dim>
509
+ <dim>2560</dim>
510
+ <dim>640</dim>
511
+ </port>
512
+ <port id="6" precision="FP32">
513
+ <dim>1</dim>
514
+ <dim>2560</dim>
515
+ </port>
516
+ </input>
517
+ <output>
518
+ <port id="7" precision="FP32">
519
+ <dim>1</dim>
520
+ <dim>1</dim>
521
+ <dim>1</dim>
522
+ <dim>640</dim>
523
+ </port>
524
+ <port id="8" precision="FP32">
525
+ <dim>1</dim>
526
+ <dim>1</dim>
527
+ <dim>640</dim>
528
+ </port>
529
+ <port id="9" precision="FP32">
530
+ <dim>1</dim>
531
+ <dim>1</dim>
532
+ <dim>640</dim>
533
+ </port>
534
+ </output>
535
+ </layer>
536
+ <layer id="38" name="Constant_31168" type="Const" version="opset1">
537
+ <data element_type="i64" shape="3" offset="17045808" size="24" />
538
+ <rt_info>
539
+ <attribute name="precise" version="0" />
540
+ </rt_info>
541
+ <output>
542
+ <port id="0" precision="I64">
543
+ <dim>3</dim>
544
+ </port>
545
+ </output>
546
+ </layer>
547
+ <layer id="39" name="/rnn/LSTM_/rnn/LSTM_output_2" type="Reshape" version="opset1">
548
+ <data special_zero="true" />
549
+ <input>
550
+ <port id="0" precision="FP32">
551
+ <dim>1</dim>
552
+ <dim>1</dim>
553
+ <dim>640</dim>
554
+ </port>
555
+ <port id="1" precision="I64">
556
+ <dim>3</dim>
557
+ </port>
558
+ </input>
559
+ <output>
560
+ <port id="2" precision="FP32" names="/rnn/LSTM_output_2">
561
+ <dim>1</dim>
562
+ <dim>1</dim>
563
+ <dim>640</dim>
564
+ </port>
565
+ </output>
566
+ </layer>
567
+ <layer id="40" name="Constant_31143" type="Const" version="opset1">
568
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
569
+ <output>
570
+ <port id="0" precision="I64">
571
+ <dim>1</dim>
572
+ </port>
573
+ </output>
574
+ </layer>
575
+ <layer id="41" name="Squeeze_31144" type="Squeeze" version="opset1">
576
+ <input>
577
+ <port id="0" precision="FP32">
578
+ <dim>1</dim>
579
+ <dim>1</dim>
580
+ <dim>1</dim>
581
+ <dim>640</dim>
582
+ </port>
583
+ <port id="1" precision="I64">
584
+ <dim>1</dim>
585
+ </port>
586
+ </input>
587
+ <output>
588
+ <port id="2" precision="FP32">
589
+ <dim>1</dim>
590
+ <dim>1</dim>
591
+ <dim>640</dim>
592
+ </port>
593
+ </output>
594
+ </layer>
595
+ <layer id="42" name="/rnn/Constant_8" type="Const" version="opset1">
596
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
597
+ <output>
598
+ <port id="0" precision="I64" names="/rnn/Constant_8_output_0">
599
+ <dim>1</dim>
600
+ </port>
601
+ </output>
602
+ </layer>
603
+ <layer id="43" name="/rnn/Constant_9" type="Const" version="opset1">
604
+ <data element_type="i64" shape="1" offset="17045832" size="8" />
605
+ <output>
606
+ <port id="0" precision="I64" names="/rnn/Constant_9_output_0">
607
+ <dim>1</dim>
608
+ </port>
609
+ </output>
610
+ </layer>
611
+ <layer id="44" name="Broadcast_29179" type="Const" version="opset1">
612
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
613
+ <output>
614
+ <port id="0" precision="I64">
615
+ <dim>1</dim>
616
+ </port>
617
+ </output>
618
+ </layer>
619
+ <layer id="45" name="/rnn/Constant_7" type="Const" version="opset1">
620
+ <data element_type="i64" shape="1" offset="10487040" size="8" />
621
+ <output>
622
+ <port id="0" precision="I64" names="/rnn/Constant_7_output_0">
623
+ <dim>1</dim>
624
+ </port>
625
+ </output>
626
+ </layer>
627
+ <layer id="46" name="/rnn/Slice_2" type="Slice" version="opset8">
628
+ <input>
629
+ <port id="0" precision="FP32">
630
+ <dim>2</dim>
631
+ <dim>1</dim>
632
+ <dim>640</dim>
633
+ </port>
634
+ <port id="1" precision="I64">
635
+ <dim>1</dim>
636
+ </port>
637
+ <port id="2" precision="I64">
638
+ <dim>1</dim>
639
+ </port>
640
+ <port id="3" precision="I64">
641
+ <dim>1</dim>
642
+ </port>
643
+ <port id="4" precision="I64">
644
+ <dim>1</dim>
645
+ </port>
646
+ </input>
647
+ <output>
648
+ <port id="5" precision="FP32" names="/rnn/Slice_2_output_0">
649
+ <dim>1</dim>
650
+ <dim>1</dim>
651
+ <dim>640</dim>
652
+ </port>
653
+ </output>
654
+ </layer>
655
+ <layer id="47" name="Constant_29223" type="Const" version="opset1">
656
+ <data element_type="i64" shape="3" offset="10487048" size="24" />
657
+ <output>
658
+ <port id="0" precision="I64">
659
+ <dim>3</dim>
660
+ </port>
661
+ </output>
662
+ </layer>
663
+ <layer id="48" name="Transpose_29224" type="Transpose" version="opset1">
664
+ <input>
665
+ <port id="0" precision="FP32">
666
+ <dim>1</dim>
667
+ <dim>1</dim>
668
+ <dim>640</dim>
669
+ </port>
670
+ <port id="1" precision="I64">
671
+ <dim>3</dim>
672
+ </port>
673
+ </input>
674
+ <output>
675
+ <port id="2" precision="FP32">
676
+ <dim>1</dim>
677
+ <dim>1</dim>
678
+ <dim>640</dim>
679
+ </port>
680
+ </output>
681
+ </layer>
682
+ <layer id="49" name="/rnn/Constant_11" type="Const" version="opset1">
683
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
684
+ <output>
685
+ <port id="0" precision="I64" names="/rnn/Constant_11_output_0">
686
+ <dim>1</dim>
687
+ </port>
688
+ </output>
689
+ </layer>
690
+ <layer id="50" name="/rnn/Constant_12" type="Const" version="opset1">
691
+ <data element_type="i64" shape="1" offset="17045832" size="8" />
692
+ <output>
693
+ <port id="0" precision="I64" names="/rnn/Constant_12_output_0">
694
+ <dim>1</dim>
695
+ </port>
696
+ </output>
697
+ </layer>
698
+ <layer id="51" name="Broadcast_29189" type="Const" version="opset1">
699
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
700
+ <output>
701
+ <port id="0" precision="I64">
702
+ <dim>1</dim>
703
+ </port>
704
+ </output>
705
+ </layer>
706
+ <layer id="52" name="/rnn/Constant_10" type="Const" version="opset1">
707
+ <data element_type="i64" shape="1" offset="10487040" size="8" />
708
+ <output>
709
+ <port id="0" precision="I64" names="/rnn/Constant_10_output_0">
710
+ <dim>1</dim>
711
+ </port>
712
+ </output>
713
+ </layer>
714
+ <layer id="53" name="/rnn/Slice_3" type="Slice" version="opset8">
715
+ <input>
716
+ <port id="0" precision="FP32">
717
+ <dim>2</dim>
718
+ <dim>1</dim>
719
+ <dim>640</dim>
720
+ </port>
721
+ <port id="1" precision="I64">
722
+ <dim>1</dim>
723
+ </port>
724
+ <port id="2" precision="I64">
725
+ <dim>1</dim>
726
+ </port>
727
+ <port id="3" precision="I64">
728
+ <dim>1</dim>
729
+ </port>
730
+ <port id="4" precision="I64">
731
+ <dim>1</dim>
732
+ </port>
733
+ </input>
734
+ <output>
735
+ <port id="5" precision="FP32" names="/rnn/Slice_3_output_0">
736
+ <dim>1</dim>
737
+ <dim>1</dim>
738
+ <dim>640</dim>
739
+ </port>
740
+ </output>
741
+ </layer>
742
+ <layer id="54" name="Constant_29225" type="Const" version="opset1">
743
+ <data element_type="i64" shape="3" offset="10487048" size="24" />
744
+ <output>
745
+ <port id="0" precision="I64">
746
+ <dim>3</dim>
747
+ </port>
748
+ </output>
749
+ </layer>
750
+ <layer id="55" name="Transpose_29226" type="Transpose" version="opset1">
751
+ <input>
752
+ <port id="0" precision="FP32">
753
+ <dim>1</dim>
754
+ <dim>1</dim>
755
+ <dim>640</dim>
756
+ </port>
757
+ <port id="1" precision="I64">
758
+ <dim>3</dim>
759
+ </port>
760
+ </input>
761
+ <output>
762
+ <port id="2" precision="FP32">
763
+ <dim>1</dim>
764
+ <dim>1</dim>
765
+ <dim>640</dim>
766
+ </port>
767
+ </output>
768
+ </layer>
769
+ <layer id="56" name="Concat_29199_compressed" type="Const" version="opset1">
770
+ <data element_type="f16" shape="1, 2560, 640" offset="17045840" size="3276800" />
771
+ <output>
772
+ <port id="0" precision="FP16">
773
+ <dim>1</dim>
774
+ <dim>2560</dim>
775
+ <dim>640</dim>
776
+ </port>
777
+ </output>
778
+ </layer>
779
+ <layer id="57" name="Concat_29199" type="Convert" version="opset1">
780
+ <data destination_type="f32" />
781
+ <rt_info>
782
+ <attribute name="decompression" version="0" />
783
+ </rt_info>
784
+ <input>
785
+ <port id="0" precision="FP16">
786
+ <dim>1</dim>
787
+ <dim>2560</dim>
788
+ <dim>640</dim>
789
+ </port>
790
+ </input>
791
+ <output>
792
+ <port id="1" precision="FP32">
793
+ <dim>1</dim>
794
+ <dim>2560</dim>
795
+ <dim>640</dim>
796
+ </port>
797
+ </output>
798
+ </layer>
799
+ <layer id="58" name="Concat_29202_compressed" type="Const" version="opset1">
800
+ <data element_type="f16" shape="1, 2560, 640" offset="20322640" size="3276800" />
801
+ <output>
802
+ <port id="0" precision="FP16">
803
+ <dim>1</dim>
804
+ <dim>2560</dim>
805
+ <dim>640</dim>
806
+ </port>
807
+ </output>
808
+ </layer>
809
+ <layer id="59" name="Concat_29202" type="Convert" version="opset1">
810
+ <data destination_type="f32" />
811
+ <rt_info>
812
+ <attribute name="decompression" version="0" />
813
+ </rt_info>
814
+ <input>
815
+ <port id="0" precision="FP16">
816
+ <dim>1</dim>
817
+ <dim>2560</dim>
818
+ <dim>640</dim>
819
+ </port>
820
+ </input>
821
+ <output>
822
+ <port id="1" precision="FP32">
823
+ <dim>1</dim>
824
+ <dim>2560</dim>
825
+ <dim>640</dim>
826
+ </port>
827
+ </output>
828
+ </layer>
829
+ <layer id="60" name="Concat_29219_compressed" type="Const" version="opset1">
830
+ <data element_type="f16" shape="1, 2560" offset="23599440" size="5120" />
831
+ <output>
832
+ <port id="0" precision="FP16">
833
+ <dim>1</dim>
834
+ <dim>2560</dim>
835
+ </port>
836
+ </output>
837
+ </layer>
838
+ <layer id="61" name="Concat_29219" type="Convert" version="opset1">
839
+ <data destination_type="f32" />
840
+ <rt_info>
841
+ <attribute name="decompression" version="0" />
842
+ </rt_info>
843
+ <input>
844
+ <port id="0" precision="FP16">
845
+ <dim>1</dim>
846
+ <dim>2560</dim>
847
+ </port>
848
+ </input>
849
+ <output>
850
+ <port id="1" precision="FP32">
851
+ <dim>1</dim>
852
+ <dim>2560</dim>
853
+ </port>
854
+ </output>
855
+ </layer>
856
+ <layer id="62" name="LSTMSequence_29234" type="LSTMSequence" version="opset5">
857
+ <data direction="forward" hidden_size="640" activations="sigmoid, tanh, tanh" activations_alpha="" activations_beta="" clip="0" />
858
+ <input>
859
+ <port id="0" precision="FP32">
860
+ <dim>1</dim>
861
+ <dim>1</dim>
862
+ <dim>640</dim>
863
+ </port>
864
+ <port id="1" precision="FP32">
865
+ <dim>1</dim>
866
+ <dim>1</dim>
867
+ <dim>640</dim>
868
+ </port>
869
+ <port id="2" precision="FP32">
870
+ <dim>1</dim>
871
+ <dim>1</dim>
872
+ <dim>640</dim>
873
+ </port>
874
+ <port id="3" precision="I64">
875
+ <dim>1</dim>
876
+ </port>
877
+ <port id="4" precision="FP32">
878
+ <dim>1</dim>
879
+ <dim>2560</dim>
880
+ <dim>640</dim>
881
+ </port>
882
+ <port id="5" precision="FP32">
883
+ <dim>1</dim>
884
+ <dim>2560</dim>
885
+ <dim>640</dim>
886
+ </port>
887
+ <port id="6" precision="FP32">
888
+ <dim>1</dim>
889
+ <dim>2560</dim>
890
+ </port>
891
+ </input>
892
+ <output>
893
+ <port id="7" precision="FP32">
894
+ <dim>1</dim>
895
+ <dim>1</dim>
896
+ <dim>1</dim>
897
+ <dim>640</dim>
898
+ </port>
899
+ <port id="8" precision="FP32">
900
+ <dim>1</dim>
901
+ <dim>1</dim>
902
+ <dim>640</dim>
903
+ </port>
904
+ <port id="9" precision="FP32">
905
+ <dim>1</dim>
906
+ <dim>1</dim>
907
+ <dim>640</dim>
908
+ </port>
909
+ </output>
910
+ </layer>
911
+ <layer id="63" name="Constant_31170" type="Const" version="opset1">
912
+ <data element_type="i64" shape="3" offset="17045808" size="24" />
913
+ <rt_info>
914
+ <attribute name="precise" version="0" />
915
+ </rt_info>
916
+ <output>
917
+ <port id="0" precision="I64">
918
+ <dim>3</dim>
919
+ </port>
920
+ </output>
921
+ </layer>
922
+ <layer id="64" name="/rnn/LSTM_1_/rnn/LSTM_1_output_2" type="Reshape" version="opset1">
923
+ <data special_zero="true" />
924
+ <input>
925
+ <port id="0" precision="FP32">
926
+ <dim>1</dim>
927
+ <dim>1</dim>
928
+ <dim>640</dim>
929
+ </port>
930
+ <port id="1" precision="I64">
931
+ <dim>3</dim>
932
+ </port>
933
+ </input>
934
+ <output>
935
+ <port id="2" precision="FP32" names="/rnn/LSTM_1_output_2">
936
+ <dim>1</dim>
937
+ <dim>1</dim>
938
+ <dim>640</dim>
939
+ </port>
940
+ </output>
941
+ </layer>
942
+ <layer id="65" name="c_out" type="Concat" version="opset1">
943
+ <data axis="0" />
944
+ <input>
945
+ <port id="0" precision="FP32">
946
+ <dim>1</dim>
947
+ <dim>1</dim>
948
+ <dim>640</dim>
949
+ </port>
950
+ <port id="1" precision="FP32">
951
+ <dim>1</dim>
952
+ <dim>1</dim>
953
+ <dim>640</dim>
954
+ </port>
955
+ </input>
956
+ <output>
957
+ <port id="2" precision="FP32" names="c_out">
958
+ <dim>2</dim>
959
+ <dim>1</dim>
960
+ <dim>640</dim>
961
+ </port>
962
+ </output>
963
+ </layer>
964
+ <layer id="67" name="Constant_31172" type="Const" version="opset1">
965
+ <data element_type="i64" shape="3" offset="17045808" size="24" />
966
+ <rt_info>
967
+ <attribute name="precise" version="0" />
968
+ </rt_info>
969
+ <output>
970
+ <port id="0" precision="I64">
971
+ <dim>3</dim>
972
+ </port>
973
+ </output>
974
+ </layer>
975
+ <layer id="68" name="/rnn/LSTM_/rnn/LSTM_output_1" type="Reshape" version="opset1">
976
+ <data special_zero="true" />
977
+ <input>
978
+ <port id="0" precision="FP32">
979
+ <dim>1</dim>
980
+ <dim>1</dim>
981
+ <dim>640</dim>
982
+ </port>
983
+ <port id="1" precision="I64">
984
+ <dim>3</dim>
985
+ </port>
986
+ </input>
987
+ <output>
988
+ <port id="2" precision="FP32" names="/rnn/LSTM_output_1">
989
+ <dim>1</dim>
990
+ <dim>1</dim>
991
+ <dim>640</dim>
992
+ </port>
993
+ </output>
994
+ </layer>
995
+ <layer id="69" name="Constant_31174" type="Const" version="opset1">
996
+ <data element_type="i64" shape="3" offset="17045808" size="24" />
997
+ <rt_info>
998
+ <attribute name="precise" version="0" />
999
+ </rt_info>
1000
+ <output>
1001
+ <port id="0" precision="I64">
1002
+ <dim>3</dim>
1003
+ </port>
1004
+ </output>
1005
+ </layer>
1006
+ <layer id="70" name="/rnn/LSTM_1_/rnn/LSTM_1_output_1" type="Reshape" version="opset1">
1007
+ <data special_zero="true" />
1008
+ <input>
1009
+ <port id="0" precision="FP32">
1010
+ <dim>1</dim>
1011
+ <dim>1</dim>
1012
+ <dim>640</dim>
1013
+ </port>
1014
+ <port id="1" precision="I64">
1015
+ <dim>3</dim>
1016
+ </port>
1017
+ </input>
1018
+ <output>
1019
+ <port id="2" precision="FP32" names="/rnn/LSTM_1_output_1">
1020
+ <dim>1</dim>
1021
+ <dim>1</dim>
1022
+ <dim>640</dim>
1023
+ </port>
1024
+ </output>
1025
+ </layer>
1026
+ <layer id="71" name="h_out" type="Concat" version="opset1">
1027
+ <data axis="0" />
1028
+ <input>
1029
+ <port id="0" precision="FP32">
1030
+ <dim>1</dim>
1031
+ <dim>1</dim>
1032
+ <dim>640</dim>
1033
+ </port>
1034
+ <port id="1" precision="FP32">
1035
+ <dim>1</dim>
1036
+ <dim>1</dim>
1037
+ <dim>640</dim>
1038
+ </port>
1039
+ </input>
1040
+ <output>
1041
+ <port id="2" precision="FP32" names="h_out">
1042
+ <dim>2</dim>
1043
+ <dim>1</dim>
1044
+ <dim>640</dim>
1045
+ </port>
1046
+ </output>
1047
+ </layer>
1048
+ <layer id="73" name="Constant_31150" type="Const" version="opset1">
1049
+ <data element_type="i64" shape="1" offset="10487072" size="8" />
1050
+ <output>
1051
+ <port id="0" precision="I64">
1052
+ <dim>1</dim>
1053
+ </port>
1054
+ </output>
1055
+ </layer>
1056
+ <layer id="74" name="Squeeze_31151" type="Squeeze" version="opset1">
1057
+ <input>
1058
+ <port id="0" precision="FP32">
1059
+ <dim>1</dim>
1060
+ <dim>1</dim>
1061
+ <dim>1</dim>
1062
+ <dim>640</dim>
1063
+ </port>
1064
+ <port id="1" precision="I64">
1065
+ <dim>1</dim>
1066
+ </port>
1067
+ </input>
1068
+ <output>
1069
+ <port id="2" precision="FP32">
1070
+ <dim>1</dim>
1071
+ <dim>1</dim>
1072
+ <dim>640</dim>
1073
+ </port>
1074
+ </output>
1075
+ </layer>
1076
+ <layer id="75" name="Constant_31152" type="Const" version="opset1">
1077
+ <data element_type="i64" shape="3" offset="10487048" size="24" />
1078
+ <output>
1079
+ <port id="0" precision="I64">
1080
+ <dim>3</dim>
1081
+ </port>
1082
+ </output>
1083
+ </layer>
1084
+ <layer id="76" name="decoder_output" type="Transpose" version="opset1">
1085
+ <input>
1086
+ <port id="0" precision="FP32">
1087
+ <dim>1</dim>
1088
+ <dim>1</dim>
1089
+ <dim>640</dim>
1090
+ </port>
1091
+ <port id="1" precision="I64">
1092
+ <dim>3</dim>
1093
+ </port>
1094
+ </input>
1095
+ <output>
1096
+ <port id="2" precision="FP32" names="decoder_output">
1097
+ <dim>1</dim>
1098
+ <dim>1</dim>
1099
+ <dim>640</dim>
1100
+ </port>
1101
+ </output>
1102
+ </layer>
1103
+ <layer id="77" name="decoder_output/sink_port_0" type="Result" version="opset1" output_names="decoder_output">
1104
+ <input>
1105
+ <port id="0" precision="FP32">
1106
+ <dim>1</dim>
1107
+ <dim>1</dim>
1108
+ <dim>640</dim>
1109
+ </port>
1110
+ </input>
1111
+ </layer>
1112
+ <layer id="72" name="h_out/sink_port_0" type="Result" version="opset1" output_names="h_out">
1113
+ <input>
1114
+ <port id="0" precision="FP32">
1115
+ <dim>2</dim>
1116
+ <dim>1</dim>
1117
+ <dim>640</dim>
1118
+ </port>
1119
+ </input>
1120
+ </layer>
1121
+ <layer id="66" name="c_out/sink_port_0" type="Result" version="opset1" output_names="c_out">
1122
+ <input>
1123
+ <port id="0" precision="FP32">
1124
+ <dim>2</dim>
1125
+ <dim>1</dim>
1126
+ <dim>640</dim>
1127
+ </port>
1128
+ </input>
1129
+ </layer>
1130
+ </layers>
1131
+ <edges>
1132
+ <edge from-layer="0" from-port="0" to-layer="20" to-port="0" />
1133
+ <edge from-layer="0" from-port="0" to-layer="53" to-port="0" />
1134
+ <edge from-layer="1" from-port="0" to-layer="13" to-port="0" />
1135
+ <edge from-layer="1" from-port="0" to-layer="46" to-port="0" />
1136
+ <edge from-layer="2" from-port="0" to-layer="6" to-port="1" />
1137
+ <edge from-layer="3" from-port="0" to-layer="4" to-port="0" />
1138
+ <edge from-layer="4" from-port="1" to-layer="6" to-port="0" />
1139
+ <edge from-layer="5" from-port="0" to-layer="6" to-port="2" />
1140
+ <edge from-layer="6" from-port="3" to-layer="8" to-port="0" />
1141
+ <edge from-layer="7" from-port="0" to-layer="8" to-port="1" />
1142
+ <edge from-layer="8" from-port="2" to-layer="23" to-port="0" />
1143
+ <edge from-layer="8" from-port="2" to-layer="37" to-port="0" />
1144
+ <edge from-layer="9" from-port="0" to-layer="13" to-port="1" />
1145
+ <edge from-layer="10" from-port="0" to-layer="13" to-port="2" />
1146
+ <edge from-layer="11" from-port="0" to-layer="13" to-port="3" />
1147
+ <edge from-layer="12" from-port="0" to-layer="13" to-port="4" />
1148
+ <edge from-layer="13" from-port="5" to-layer="15" to-port="0" />
1149
+ <edge from-layer="14" from-port="0" to-layer="15" to-port="1" />
1150
+ <edge from-layer="15" from-port="2" to-layer="37" to-port="1" />
1151
+ <edge from-layer="16" from-port="0" to-layer="20" to-port="1" />
1152
+ <edge from-layer="17" from-port="0" to-layer="20" to-port="2" />
1153
+ <edge from-layer="18" from-port="0" to-layer="20" to-port="3" />
1154
+ <edge from-layer="19" from-port="0" to-layer="20" to-port="4" />
1155
+ <edge from-layer="20" from-port="5" to-layer="22" to-port="0" />
1156
+ <edge from-layer="21" from-port="0" to-layer="22" to-port="1" />
1157
+ <edge from-layer="22" from-port="2" to-layer="37" to-port="2" />
1158
+ <edge from-layer="23" from-port="1" to-layer="26" to-port="0" />
1159
+ <edge from-layer="23" from-port="1" to-layer="29" to-port="0" />
1160
+ <edge from-layer="24" from-port="0" to-layer="26" to-port="1" />
1161
+ <edge from-layer="25" from-port="0" to-layer="26" to-port="2" />
1162
+ <edge from-layer="26" from-port="3" to-layer="30" to-port="0" />
1163
+ <edge from-layer="27" from-port="0" to-layer="29" to-port="1" />
1164
+ <edge from-layer="28" from-port="0" to-layer="29" to-port="2" />
1165
+ <edge from-layer="29" from-port="3" to-layer="30" to-port="1" />
1166
+ <edge from-layer="30" from-port="2" to-layer="37" to-port="3" />
1167
+ <edge from-layer="30" from-port="2" to-layer="62" to-port="3" />
1168
+ <edge from-layer="31" from-port="0" to-layer="32" to-port="0" />
1169
+ <edge from-layer="32" from-port="1" to-layer="37" to-port="4" />
1170
+ <edge from-layer="33" from-port="0" to-layer="34" to-port="0" />
1171
+ <edge from-layer="34" from-port="1" to-layer="37" to-port="5" />
1172
+ <edge from-layer="35" from-port="0" to-layer="36" to-port="0" />
1173
+ <edge from-layer="36" from-port="1" to-layer="37" to-port="6" />
1174
+ <edge from-layer="37" from-port="9" to-layer="39" to-port="0" />
1175
+ <edge from-layer="37" from-port="7" to-layer="41" to-port="0" />
1176
+ <edge from-layer="37" from-port="8" to-layer="68" to-port="0" />
1177
+ <edge from-layer="38" from-port="0" to-layer="39" to-port="1" />
1178
+ <edge from-layer="39" from-port="2" to-layer="65" to-port="0" />
1179
+ <edge from-layer="40" from-port="0" to-layer="41" to-port="1" />
1180
+ <edge from-layer="41" from-port="2" to-layer="62" to-port="0" />
1181
+ <edge from-layer="42" from-port="0" to-layer="46" to-port="1" />
1182
+ <edge from-layer="43" from-port="0" to-layer="46" to-port="2" />
1183
+ <edge from-layer="44" from-port="0" to-layer="46" to-port="3" />
1184
+ <edge from-layer="45" from-port="0" to-layer="46" to-port="4" />
1185
+ <edge from-layer="46" from-port="5" to-layer="48" to-port="0" />
1186
+ <edge from-layer="47" from-port="0" to-layer="48" to-port="1" />
1187
+ <edge from-layer="48" from-port="2" to-layer="62" to-port="1" />
1188
+ <edge from-layer="49" from-port="0" to-layer="53" to-port="1" />
1189
+ <edge from-layer="50" from-port="0" to-layer="53" to-port="2" />
1190
+ <edge from-layer="51" from-port="0" to-layer="53" to-port="3" />
1191
+ <edge from-layer="52" from-port="0" to-layer="53" to-port="4" />
1192
+ <edge from-layer="53" from-port="5" to-layer="55" to-port="0" />
1193
+ <edge from-layer="54" from-port="0" to-layer="55" to-port="1" />
1194
+ <edge from-layer="55" from-port="2" to-layer="62" to-port="2" />
1195
+ <edge from-layer="56" from-port="0" to-layer="57" to-port="0" />
1196
+ <edge from-layer="57" from-port="1" to-layer="62" to-port="4" />
1197
+ <edge from-layer="58" from-port="0" to-layer="59" to-port="0" />
1198
+ <edge from-layer="59" from-port="1" to-layer="62" to-port="5" />
1199
+ <edge from-layer="60" from-port="0" to-layer="61" to-port="0" />
1200
+ <edge from-layer="61" from-port="1" to-layer="62" to-port="6" />
1201
+ <edge from-layer="62" from-port="9" to-layer="64" to-port="0" />
1202
+ <edge from-layer="62" from-port="8" to-layer="70" to-port="0" />
1203
+ <edge from-layer="62" from-port="7" to-layer="74" to-port="0" />
1204
+ <edge from-layer="63" from-port="0" to-layer="64" to-port="1" />
1205
+ <edge from-layer="64" from-port="2" to-layer="65" to-port="1" />
1206
+ <edge from-layer="65" from-port="2" to-layer="66" to-port="0" />
1207
+ <edge from-layer="67" from-port="0" to-layer="68" to-port="1" />
1208
+ <edge from-layer="68" from-port="2" to-layer="71" to-port="0" />
1209
+ <edge from-layer="69" from-port="0" to-layer="70" to-port="1" />
1210
+ <edge from-layer="70" from-port="2" to-layer="71" to-port="1" />
1211
+ <edge from-layer="71" from-port="2" to-layer="72" to-port="0" />
1212
+ <edge from-layer="73" from-port="0" to-layer="74" to-port="1" />
1213
+ <edge from-layer="74" from-port="2" to-layer="76" to-port="0" />
1214
+ <edge from-layer="75" from-port="0" to-layer="76" to-port="1" />
1215
+ <edge from-layer="76" from-port="2" to-layer="77" to-port="0" />
1216
+ </edges>
1217
+ <rt_info>
1218
+ <Runtime_version value="2025.3.0-19807-44526285f24-releases/2025/3" />
1219
+ <conversion_parameters>
1220
+ <input_model value="DIR\parakeet_decoder.onnx" />
1221
+ <is_python_object value="False" />
1222
+ </conversion_parameters>
1223
+ </rt_info>
1224
+ </net>
parakeet_encoder.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:52999e4ed1595470a80b9e0666ddcf9948c0b36f437967cb319c5c332c13e88e
3
+ size 1185869036
parakeet_encoder.xml ADDED
The diff for this file is too large to render. See raw diff
 
parakeet_joint.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e88f298ad8ca8d90ff50b0147281a3ca8cf677550619513c0013a3160b57f001
3
+ size 12642332
parakeet_joint.xml ADDED
@@ -0,0 +1,497 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="main_graph" version="11">
3
+ <layers>
4
+ <layer id="1" name="encoder_outputs" type="Parameter" version="opset1">
5
+ <data shape="1,1,1024" element_type="f32" />
6
+ <output>
7
+ <port id="0" precision="FP32" names="encoder_outputs">
8
+ <dim>1</dim>
9
+ <dim>1</dim>
10
+ <dim>1024</dim>
11
+ </port>
12
+ </output>
13
+ </layer>
14
+ <layer id="0" name="decoder_outputs" type="Parameter" version="opset1">
15
+ <data shape="1,1,640" element_type="f32" />
16
+ <output>
17
+ <port id="0" precision="FP32" names="decoder_outputs">
18
+ <dim>1</dim>
19
+ <dim>1</dim>
20
+ <dim>640</dim>
21
+ </port>
22
+ </output>
23
+ </layer>
24
+ <layer id="2" name="Constant_33450_compressed" type="Const" version="opset1">
25
+ <data element_type="f16" shape="1, 1, 1, 8198" offset="0" size="16396" />
26
+ <output>
27
+ <port id="0" precision="FP16">
28
+ <dim>1</dim>
29
+ <dim>1</dim>
30
+ <dim>1</dim>
31
+ <dim>8198</dim>
32
+ </port>
33
+ </output>
34
+ </layer>
35
+ <layer id="3" name="Constant_33450" type="Convert" version="opset1">
36
+ <data destination_type="f32" />
37
+ <rt_info>
38
+ <attribute name="decompression" version="0" />
39
+ </rt_info>
40
+ <input>
41
+ <port id="0" precision="FP16">
42
+ <dim>1</dim>
43
+ <dim>1</dim>
44
+ <dim>1</dim>
45
+ <dim>8198</dim>
46
+ </port>
47
+ </input>
48
+ <output>
49
+ <port id="1" precision="FP32">
50
+ <dim>1</dim>
51
+ <dim>1</dim>
52
+ <dim>1</dim>
53
+ <dim>8198</dim>
54
+ </port>
55
+ </output>
56
+ </layer>
57
+ <layer id="4" name="Constant_33448_compressed" type="Const" version="opset1">
58
+ <data element_type="f16" shape="1, 1, 640" offset="16396" size="1280" />
59
+ <output>
60
+ <port id="0" precision="FP16">
61
+ <dim>1</dim>
62
+ <dim>1</dim>
63
+ <dim>640</dim>
64
+ </port>
65
+ </output>
66
+ </layer>
67
+ <layer id="5" name="Constant_33448" type="Convert" version="opset1">
68
+ <data destination_type="f32" />
69
+ <rt_info>
70
+ <attribute name="decompression" version="0" />
71
+ </rt_info>
72
+ <input>
73
+ <port id="0" precision="FP16">
74
+ <dim>1</dim>
75
+ <dim>1</dim>
76
+ <dim>640</dim>
77
+ </port>
78
+ </input>
79
+ <output>
80
+ <port id="1" precision="FP32">
81
+ <dim>1</dim>
82
+ <dim>1</dim>
83
+ <dim>640</dim>
84
+ </port>
85
+ </output>
86
+ </layer>
87
+ <layer id="6" name="Transpose_33330_compressed" type="Const" version="opset1">
88
+ <data element_type="f16" shape="640, 1024" offset="17676" size="1310720" />
89
+ <output>
90
+ <port id="0" precision="FP16">
91
+ <dim>640</dim>
92
+ <dim>1024</dim>
93
+ </port>
94
+ </output>
95
+ </layer>
96
+ <layer id="7" name="Transpose_33330" type="Convert" version="opset1">
97
+ <data destination_type="f32" />
98
+ <rt_info>
99
+ <attribute name="decompression" version="0" />
100
+ </rt_info>
101
+ <input>
102
+ <port id="0" precision="FP16">
103
+ <dim>640</dim>
104
+ <dim>1024</dim>
105
+ </port>
106
+ </input>
107
+ <output>
108
+ <port id="1" precision="FP32">
109
+ <dim>640</dim>
110
+ <dim>1024</dim>
111
+ </port>
112
+ </output>
113
+ </layer>
114
+ <layer id="8" name="/enc/MatMul" type="MatMul" version="opset1">
115
+ <data transpose_a="false" transpose_b="true" />
116
+ <input>
117
+ <port id="0" precision="FP32">
118
+ <dim>1</dim>
119
+ <dim>1</dim>
120
+ <dim>1024</dim>
121
+ </port>
122
+ <port id="1" precision="FP32">
123
+ <dim>640</dim>
124
+ <dim>1024</dim>
125
+ </port>
126
+ </input>
127
+ <output>
128
+ <port id="2" precision="FP32" names="/enc/MatMul_output_0">
129
+ <dim>1</dim>
130
+ <dim>1</dim>
131
+ <dim>640</dim>
132
+ </port>
133
+ </output>
134
+ </layer>
135
+ <layer id="9" name="/enc/Add" type="Add" version="opset1">
136
+ <data auto_broadcast="numpy" />
137
+ <input>
138
+ <port id="0" precision="FP32">
139
+ <dim>1</dim>
140
+ <dim>1</dim>
141
+ <dim>640</dim>
142
+ </port>
143
+ <port id="1" precision="FP32">
144
+ <dim>1</dim>
145
+ <dim>1</dim>
146
+ <dim>640</dim>
147
+ </port>
148
+ </input>
149
+ <output>
150
+ <port id="2" precision="FP32" names="/enc/Add_output_0">
151
+ <dim>1</dim>
152
+ <dim>1</dim>
153
+ <dim>640</dim>
154
+ </port>
155
+ </output>
156
+ </layer>
157
+ <layer id="10" name="/Constant" type="Const" version="opset1">
158
+ <data element_type="i64" shape="1" offset="1328396" size="8" />
159
+ <output>
160
+ <port id="0" precision="I64" names="/Constant_output_0">
161
+ <dim>1</dim>
162
+ </port>
163
+ </output>
164
+ </layer>
165
+ <layer id="11" name="/Unsqueeze" type="Unsqueeze" version="opset1">
166
+ <input>
167
+ <port id="0" precision="FP32">
168
+ <dim>1</dim>
169
+ <dim>1</dim>
170
+ <dim>640</dim>
171
+ </port>
172
+ <port id="1" precision="I64">
173
+ <dim>1</dim>
174
+ </port>
175
+ </input>
176
+ <output>
177
+ <port id="2" precision="FP32" names="/Unsqueeze_output_0">
178
+ <dim>1</dim>
179
+ <dim>1</dim>
180
+ <dim>1</dim>
181
+ <dim>640</dim>
182
+ </port>
183
+ </output>
184
+ </layer>
185
+ <layer id="12" name="Constant_33449_compressed" type="Const" version="opset1">
186
+ <data element_type="f16" shape="1, 1, 640" offset="1328404" size="1280" />
187
+ <output>
188
+ <port id="0" precision="FP16">
189
+ <dim>1</dim>
190
+ <dim>1</dim>
191
+ <dim>640</dim>
192
+ </port>
193
+ </output>
194
+ </layer>
195
+ <layer id="13" name="Constant_33449" type="Convert" version="opset1">
196
+ <data destination_type="f32" />
197
+ <rt_info>
198
+ <attribute name="decompression" version="0" />
199
+ </rt_info>
200
+ <input>
201
+ <port id="0" precision="FP16">
202
+ <dim>1</dim>
203
+ <dim>1</dim>
204
+ <dim>640</dim>
205
+ </port>
206
+ </input>
207
+ <output>
208
+ <port id="1" precision="FP32">
209
+ <dim>1</dim>
210
+ <dim>1</dim>
211
+ <dim>640</dim>
212
+ </port>
213
+ </output>
214
+ </layer>
215
+ <layer id="14" name="Transpose_33337_compressed" type="Const" version="opset1">
216
+ <data element_type="f16" shape="640, 640" offset="1329684" size="819200" />
217
+ <output>
218
+ <port id="0" precision="FP16">
219
+ <dim>640</dim>
220
+ <dim>640</dim>
221
+ </port>
222
+ </output>
223
+ </layer>
224
+ <layer id="15" name="Transpose_33337" type="Convert" version="opset1">
225
+ <data destination_type="f32" />
226
+ <rt_info>
227
+ <attribute name="decompression" version="0" />
228
+ </rt_info>
229
+ <input>
230
+ <port id="0" precision="FP16">
231
+ <dim>640</dim>
232
+ <dim>640</dim>
233
+ </port>
234
+ </input>
235
+ <output>
236
+ <port id="1" precision="FP32">
237
+ <dim>640</dim>
238
+ <dim>640</dim>
239
+ </port>
240
+ </output>
241
+ </layer>
242
+ <layer id="16" name="/pred/MatMul" type="MatMul" version="opset1">
243
+ <data transpose_a="false" transpose_b="true" />
244
+ <input>
245
+ <port id="0" precision="FP32">
246
+ <dim>1</dim>
247
+ <dim>1</dim>
248
+ <dim>640</dim>
249
+ </port>
250
+ <port id="1" precision="FP32">
251
+ <dim>640</dim>
252
+ <dim>640</dim>
253
+ </port>
254
+ </input>
255
+ <output>
256
+ <port id="2" precision="FP32" names="/pred/MatMul_output_0">
257
+ <dim>1</dim>
258
+ <dim>1</dim>
259
+ <dim>640</dim>
260
+ </port>
261
+ </output>
262
+ </layer>
263
+ <layer id="17" name="/pred/Add" type="Add" version="opset1">
264
+ <data auto_broadcast="numpy" />
265
+ <input>
266
+ <port id="0" precision="FP32">
267
+ <dim>1</dim>
268
+ <dim>1</dim>
269
+ <dim>640</dim>
270
+ </port>
271
+ <port id="1" precision="FP32">
272
+ <dim>1</dim>
273
+ <dim>1</dim>
274
+ <dim>640</dim>
275
+ </port>
276
+ </input>
277
+ <output>
278
+ <port id="2" precision="FP32" names="/pred/Add_output_0">
279
+ <dim>1</dim>
280
+ <dim>1</dim>
281
+ <dim>640</dim>
282
+ </port>
283
+ </output>
284
+ </layer>
285
+ <layer id="18" name="/Constant_1" type="Const" version="opset1">
286
+ <data element_type="i64" shape="1" offset="2148884" size="8" />
287
+ <output>
288
+ <port id="0" precision="I64" names="/Constant_1_output_0">
289
+ <dim>1</dim>
290
+ </port>
291
+ </output>
292
+ </layer>
293
+ <layer id="19" name="/Unsqueeze_1" type="Unsqueeze" version="opset1">
294
+ <input>
295
+ <port id="0" precision="FP32">
296
+ <dim>1</dim>
297
+ <dim>1</dim>
298
+ <dim>640</dim>
299
+ </port>
300
+ <port id="1" precision="I64">
301
+ <dim>1</dim>
302
+ </port>
303
+ </input>
304
+ <output>
305
+ <port id="2" precision="FP32" names="/Unsqueeze_1_output_0">
306
+ <dim>1</dim>
307
+ <dim>1</dim>
308
+ <dim>1</dim>
309
+ <dim>640</dim>
310
+ </port>
311
+ </output>
312
+ </layer>
313
+ <layer id="20" name="/Add" type="Add" version="opset1">
314
+ <data auto_broadcast="numpy" />
315
+ <input>
316
+ <port id="0" precision="FP32">
317
+ <dim>1</dim>
318
+ <dim>1</dim>
319
+ <dim>1</dim>
320
+ <dim>640</dim>
321
+ </port>
322
+ <port id="1" precision="FP32">
323
+ <dim>1</dim>
324
+ <dim>1</dim>
325
+ <dim>1</dim>
326
+ <dim>640</dim>
327
+ </port>
328
+ </input>
329
+ <output>
330
+ <port id="2" precision="FP32" names="/Add_output_0">
331
+ <dim>1</dim>
332
+ <dim>1</dim>
333
+ <dim>1</dim>
334
+ <dim>640</dim>
335
+ </port>
336
+ </output>
337
+ </layer>
338
+ <layer id="21" name="/joint_net/joint_net.0/Relu" type="ReLU" version="opset1">
339
+ <input>
340
+ <port id="0" precision="FP32">
341
+ <dim>1</dim>
342
+ <dim>1</dim>
343
+ <dim>1</dim>
344
+ <dim>640</dim>
345
+ </port>
346
+ </input>
347
+ <output>
348
+ <port id="1" precision="FP32" names="/joint_net/joint_net.0/Relu_output_0">
349
+ <dim>1</dim>
350
+ <dim>1</dim>
351
+ <dim>1</dim>
352
+ <dim>640</dim>
353
+ </port>
354
+ </output>
355
+ </layer>
356
+ <layer id="22" name="Transpose_33346_compressed" type="Const" version="opset1">
357
+ <data element_type="f16" shape="8198, 640" offset="2148892" size="10493440" />
358
+ <output>
359
+ <port id="0" precision="FP16">
360
+ <dim>8198</dim>
361
+ <dim>640</dim>
362
+ </port>
363
+ </output>
364
+ </layer>
365
+ <layer id="23" name="Transpose_33346" type="Convert" version="opset1">
366
+ <data destination_type="f32" />
367
+ <rt_info>
368
+ <attribute name="decompression" version="0" />
369
+ </rt_info>
370
+ <input>
371
+ <port id="0" precision="FP16">
372
+ <dim>8198</dim>
373
+ <dim>640</dim>
374
+ </port>
375
+ </input>
376
+ <output>
377
+ <port id="1" precision="FP32">
378
+ <dim>8198</dim>
379
+ <dim>640</dim>
380
+ </port>
381
+ </output>
382
+ </layer>
383
+ <layer id="24" name="/joint_net/joint_net.2/MatMul" type="MatMul" version="opset1">
384
+ <data transpose_a="false" transpose_b="true" />
385
+ <input>
386
+ <port id="0" precision="FP32">
387
+ <dim>1</dim>
388
+ <dim>1</dim>
389
+ <dim>1</dim>
390
+ <dim>640</dim>
391
+ </port>
392
+ <port id="1" precision="FP32">
393
+ <dim>8198</dim>
394
+ <dim>640</dim>
395
+ </port>
396
+ </input>
397
+ <output>
398
+ <port id="2" precision="FP32" names="/joint_net/joint_net.2/MatMul_output_0">
399
+ <dim>1</dim>
400
+ <dim>1</dim>
401
+ <dim>1</dim>
402
+ <dim>8198</dim>
403
+ </port>
404
+ </output>
405
+ </layer>
406
+ <layer id="25" name="/joint_net/joint_net.2/Add" type="Add" version="opset1">
407
+ <data auto_broadcast="numpy" />
408
+ <input>
409
+ <port id="0" precision="FP32">
410
+ <dim>1</dim>
411
+ <dim>1</dim>
412
+ <dim>1</dim>
413
+ <dim>8198</dim>
414
+ </port>
415
+ <port id="1" precision="FP32">
416
+ <dim>1</dim>
417
+ <dim>1</dim>
418
+ <dim>1</dim>
419
+ <dim>8198</dim>
420
+ </port>
421
+ </input>
422
+ <output>
423
+ <port id="2" precision="FP32" names="/joint_net/joint_net.2/Add_output_0">
424
+ <dim>1</dim>
425
+ <dim>1</dim>
426
+ <dim>1</dim>
427
+ <dim>8198</dim>
428
+ </port>
429
+ </output>
430
+ </layer>
431
+ <layer id="26" name="logits" type="LogSoftmax" version="opset5">
432
+ <data axis="-1" />
433
+ <input>
434
+ <port id="0" precision="FP32">
435
+ <dim>1</dim>
436
+ <dim>1</dim>
437
+ <dim>1</dim>
438
+ <dim>8198</dim>
439
+ </port>
440
+ </input>
441
+ <output>
442
+ <port id="1" precision="FP32" names="logits">
443
+ <dim>1</dim>
444
+ <dim>1</dim>
445
+ <dim>1</dim>
446
+ <dim>8198</dim>
447
+ </port>
448
+ </output>
449
+ </layer>
450
+ <layer id="27" name="logits/sink_port_0" type="Result" version="opset1" output_names="logits">
451
+ <input>
452
+ <port id="0" precision="FP32">
453
+ <dim>1</dim>
454
+ <dim>1</dim>
455
+ <dim>1</dim>
456
+ <dim>8198</dim>
457
+ </port>
458
+ </input>
459
+ </layer>
460
+ </layers>
461
+ <edges>
462
+ <edge from-layer="0" from-port="0" to-layer="16" to-port="0" />
463
+ <edge from-layer="1" from-port="0" to-layer="8" to-port="0" />
464
+ <edge from-layer="2" from-port="0" to-layer="3" to-port="0" />
465
+ <edge from-layer="3" from-port="1" to-layer="25" to-port="0" />
466
+ <edge from-layer="4" from-port="0" to-layer="5" to-port="0" />
467
+ <edge from-layer="5" from-port="1" to-layer="9" to-port="0" />
468
+ <edge from-layer="6" from-port="0" to-layer="7" to-port="0" />
469
+ <edge from-layer="7" from-port="1" to-layer="8" to-port="1" />
470
+ <edge from-layer="8" from-port="2" to-layer="9" to-port="1" />
471
+ <edge from-layer="9" from-port="2" to-layer="11" to-port="0" />
472
+ <edge from-layer="10" from-port="0" to-layer="11" to-port="1" />
473
+ <edge from-layer="11" from-port="2" to-layer="20" to-port="0" />
474
+ <edge from-layer="12" from-port="0" to-layer="13" to-port="0" />
475
+ <edge from-layer="13" from-port="1" to-layer="17" to-port="0" />
476
+ <edge from-layer="14" from-port="0" to-layer="15" to-port="0" />
477
+ <edge from-layer="15" from-port="1" to-layer="16" to-port="1" />
478
+ <edge from-layer="16" from-port="2" to-layer="17" to-port="1" />
479
+ <edge from-layer="17" from-port="2" to-layer="19" to-port="0" />
480
+ <edge from-layer="18" from-port="0" to-layer="19" to-port="1" />
481
+ <edge from-layer="19" from-port="2" to-layer="20" to-port="1" />
482
+ <edge from-layer="20" from-port="2" to-layer="21" to-port="0" />
483
+ <edge from-layer="21" from-port="1" to-layer="24" to-port="0" />
484
+ <edge from-layer="22" from-port="0" to-layer="23" to-port="0" />
485
+ <edge from-layer="23" from-port="1" to-layer="24" to-port="1" />
486
+ <edge from-layer="24" from-port="2" to-layer="25" to-port="1" />
487
+ <edge from-layer="25" from-port="2" to-layer="26" to-port="0" />
488
+ <edge from-layer="26" from-port="1" to-layer="27" to-port="0" />
489
+ </edges>
490
+ <rt_info>
491
+ <Runtime_version value="2025.3.0-19807-44526285f24-releases/2025/3" />
492
+ <conversion_parameters>
493
+ <input_model value="DIR\parakeet_joint.onnx" />
494
+ <is_python_object value="False" />
495
+ </conversion_parameters>
496
+ </rt_info>
497
+ </net>
parakeet_melspectogram.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a20c3299f863b03ccb0b91465fd02f1d5eddd749bc44418e3674e7a6b88b1116
3
+ size 477078
parakeet_melspectogram.xml ADDED
@@ -0,0 +1,1685 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="Model9" version="11">
3
+ <layers>
4
+ <layer id="1" name="input_signals" type="Parameter" version="opset1">
5
+ <data shape="1,240000" element_type="f32" />
6
+ <output>
7
+ <port id="0" precision="FP32" names="input_signals">
8
+ <dim>1</dim>
9
+ <dim>240000</dim>
10
+ </port>
11
+ </output>
12
+ </layer>
13
+ <layer id="0" name="input_length" type="Parameter" version="opset1">
14
+ <data shape="1" element_type="i64" />
15
+ <output>
16
+ <port id="0" precision="I64" names="input_length">
17
+ <dim>1</dim>
18
+ </port>
19
+ </output>
20
+ </layer>
21
+ <layer id="2" name="aten::to/Convert" type="Convert" version="opset1">
22
+ <data destination_type="f32" />
23
+ <input>
24
+ <port id="0" precision="I64">
25
+ <dim>1</dim>
26
+ </port>
27
+ </input>
28
+ <output>
29
+ <port id="1" precision="FP32" names="9,x.1">
30
+ <dim>1</dim>
31
+ </port>
32
+ </output>
33
+ </layer>
34
+ <layer id="3" name="Constant_36655_compressed" type="Const" version="opset1">
35
+ <data element_type="f16" shape="1" offset="0" size="2" />
36
+ <output>
37
+ <port id="0" precision="FP16">
38
+ <dim>1</dim>
39
+ </port>
40
+ </output>
41
+ </layer>
42
+ <layer id="4" name="Constant_36655" type="Convert" version="opset1">
43
+ <data destination_type="f32" />
44
+ <rt_info>
45
+ <attribute name="decompression" version="0" />
46
+ </rt_info>
47
+ <input>
48
+ <port id="0" precision="FP16">
49
+ <dim>1</dim>
50
+ </port>
51
+ </input>
52
+ <output>
53
+ <port id="1" precision="FP32">
54
+ <dim>1</dim>
55
+ </port>
56
+ </output>
57
+ </layer>
58
+ <layer id="5" name="aten::sub/Subtract" type="Add" version="opset1">
59
+ <data auto_broadcast="numpy" />
60
+ <input>
61
+ <port id="0" precision="FP32">
62
+ <dim>1</dim>
63
+ </port>
64
+ <port id="1" precision="FP32">
65
+ <dim>1</dim>
66
+ </port>
67
+ </input>
68
+ <output>
69
+ <port id="2" precision="FP32" names="12">
70
+ <dim>1</dim>
71
+ </port>
72
+ </output>
73
+ </layer>
74
+ <layer id="6" name="Constant_36656_compressed" type="Const" version="opset1">
75
+ <data element_type="f16" shape="1" offset="2" size="2" />
76
+ <output>
77
+ <port id="0" precision="FP16">
78
+ <dim>1</dim>
79
+ </port>
80
+ </output>
81
+ </layer>
82
+ <layer id="7" name="Constant_36656" type="Convert" version="opset1">
83
+ <data destination_type="f32" />
84
+ <rt_info>
85
+ <attribute name="decompression" version="0" />
86
+ </rt_info>
87
+ <input>
88
+ <port id="0" precision="FP16">
89
+ <dim>1</dim>
90
+ </port>
91
+ </input>
92
+ <output>
93
+ <port id="1" precision="FP32">
94
+ <dim>1</dim>
95
+ </port>
96
+ </output>
97
+ </layer>
98
+ <layer id="8" name="aten::div/Divide" type="Divide" version="opset1">
99
+ <data auto_broadcast="numpy" m_pythondiv="true" />
100
+ <input>
101
+ <port id="0" precision="FP32">
102
+ <dim>1</dim>
103
+ </port>
104
+ <port id="1" precision="FP32">
105
+ <dim>1</dim>
106
+ </port>
107
+ </input>
108
+ <output>
109
+ <port id="2" precision="FP32" names="14">
110
+ <dim>1</dim>
111
+ </port>
112
+ </output>
113
+ </layer>
114
+ <layer id="9" name="aten::floor/Floor" type="Floor" version="opset1">
115
+ <input>
116
+ <port id="0" precision="FP32">
117
+ <dim>1</dim>
118
+ </port>
119
+ </input>
120
+ <output>
121
+ <port id="1" precision="FP32" names="15">
122
+ <dim>1</dim>
123
+ </port>
124
+ </output>
125
+ </layer>
126
+ <layer id="10" name="Constant_36657_compressed" type="Const" version="opset1">
127
+ <data element_type="f16" shape="1" offset="4" size="2" />
128
+ <output>
129
+ <port id="0" precision="FP16">
130
+ <dim>1</dim>
131
+ </port>
132
+ </output>
133
+ </layer>
134
+ <layer id="11" name="Constant_36657" type="Convert" version="opset1">
135
+ <data destination_type="f32" />
136
+ <rt_info>
137
+ <attribute name="decompression" version="0" />
138
+ </rt_info>
139
+ <input>
140
+ <port id="0" precision="FP16">
141
+ <dim>1</dim>
142
+ </port>
143
+ </input>
144
+ <output>
145
+ <port id="1" precision="FP32">
146
+ <dim>1</dim>
147
+ </port>
148
+ </output>
149
+ </layer>
150
+ <layer id="12" name="aten::add/Add" type="Add" version="opset1">
151
+ <data auto_broadcast="numpy" />
152
+ <input>
153
+ <port id="0" precision="FP32">
154
+ <dim>1</dim>
155
+ </port>
156
+ <port id="1" precision="FP32">
157
+ <dim>1</dim>
158
+ </port>
159
+ </input>
160
+ <output>
161
+ <port id="2" precision="FP32" names="18,seq">
162
+ <dim>1</dim>
163
+ </port>
164
+ </output>
165
+ </layer>
166
+ <layer id="13" name="aten::to/Convert_1" type="Convert" version="opset1">
167
+ <data destination_type="i64" />
168
+ <input>
169
+ <port id="0" precision="FP32">
170
+ <dim>1</dim>
171
+ </port>
172
+ </input>
173
+ <output>
174
+ <port id="1" precision="I64" names="seq_len">
175
+ <dim>1</dim>
176
+ </port>
177
+ </output>
178
+ </layer>
179
+ <layer id="15" name="aten::arange/Constant" type="Const" version="opset1">
180
+ <data element_type="i32" shape="" offset="6" size="4" />
181
+ <output>
182
+ <port id="0" precision="I32" />
183
+ </output>
184
+ </layer>
185
+ <layer id="16" name="24" type="Const" version="opset1">
186
+ <data element_type="i64" shape="" offset="10" size="8" />
187
+ <output>
188
+ <port id="0" precision="I64" names="24" />
189
+ </output>
190
+ </layer>
191
+ <layer id="17" name="aten::unsqueeze/Unsqueeze" type="Unsqueeze" version="opset1">
192
+ <input>
193
+ <port id="0" precision="FP32">
194
+ <dim>1</dim>
195
+ <dim>240000</dim>
196
+ </port>
197
+ <port id="1" precision="I64" />
198
+ </input>
199
+ <output>
200
+ <port id="2" precision="FP32" names="25,input.1">
201
+ <dim>1</dim>
202
+ <dim>1</dim>
203
+ <dim>240000</dim>
204
+ </port>
205
+ </output>
206
+ </layer>
207
+ <layer id="18" name="self.preemph_kernel_compressed" type="Const" version="opset1">
208
+ <data element_type="f16" shape="1, 1, 2" offset="18" size="4" />
209
+ <output>
210
+ <port id="0" precision="FP16" names="self.preemph_kernel">
211
+ <dim>1</dim>
212
+ <dim>1</dim>
213
+ <dim>2</dim>
214
+ </port>
215
+ </output>
216
+ </layer>
217
+ <layer id="19" name="self.preemph_kernel" type="Convert" version="opset1">
218
+ <data destination_type="f32" />
219
+ <rt_info>
220
+ <attribute name="decompression" version="0" />
221
+ </rt_info>
222
+ <input>
223
+ <port id="0" precision="FP16">
224
+ <dim>1</dim>
225
+ <dim>1</dim>
226
+ <dim>2</dim>
227
+ </port>
228
+ </input>
229
+ <output>
230
+ <port id="1" precision="FP32">
231
+ <dim>1</dim>
232
+ <dim>1</dim>
233
+ <dim>2</dim>
234
+ </port>
235
+ </output>
236
+ </layer>
237
+ <layer id="20" name="aten::_convolution/Convolution" type="Convolution" version="opset1">
238
+ <data strides="1" dilations="1" pads_begin="1" pads_end="0" auto_pad="explicit" />
239
+ <input>
240
+ <port id="0" precision="FP32">
241
+ <dim>1</dim>
242
+ <dim>1</dim>
243
+ <dim>240000</dim>
244
+ </port>
245
+ <port id="1" precision="FP32">
246
+ <dim>1</dim>
247
+ <dim>1</dim>
248
+ <dim>2</dim>
249
+ </port>
250
+ </input>
251
+ <output>
252
+ <port id="2" precision="FP32" names="47">
253
+ <dim>1</dim>
254
+ <dim>1</dim>
255
+ <dim>240000</dim>
256
+ </port>
257
+ </output>
258
+ </layer>
259
+ <layer id="21" name="48" type="Const" version="opset1">
260
+ <data element_type="i64" shape="" offset="10" size="8" />
261
+ <output>
262
+ <port id="0" precision="I64" names="48" />
263
+ </output>
264
+ </layer>
265
+ <layer id="22" name="aten::squeeze/Squeeze" type="Squeeze" version="opset1">
266
+ <input>
267
+ <port id="0" precision="FP32">
268
+ <dim>1</dim>
269
+ <dim>1</dim>
270
+ <dim>240000</dim>
271
+ </port>
272
+ <port id="1" precision="I64" />
273
+ </input>
274
+ <output>
275
+ <port id="2" precision="FP32" names="49,x.5">
276
+ <dim>1</dim>
277
+ <dim>240000</dim>
278
+ </port>
279
+ </output>
280
+ </layer>
281
+ <layer id="23" name="50" type="Const" version="opset1">
282
+ <data element_type="i64" shape="" offset="10" size="8" />
283
+ <output>
284
+ <port id="0" precision="I64" names="50" />
285
+ </output>
286
+ </layer>
287
+ <layer id="24" name="aten::unsqueeze/Unsqueeze_1" type="Unsqueeze" version="opset1">
288
+ <input>
289
+ <port id="0" precision="FP32">
290
+ <dim>1</dim>
291
+ <dim>240000</dim>
292
+ </port>
293
+ <port id="1" precision="I64" />
294
+ </input>
295
+ <output>
296
+ <port id="2" precision="FP32" names="51,input">
297
+ <dim>1</dim>
298
+ <dim>1</dim>
299
+ <dim>240000</dim>
300
+ </port>
301
+ </output>
302
+ </layer>
303
+ <layer id="25" name="aten::to/Convert_3_compressed" type="Const" version="opset1">
304
+ <data element_type="f16" shape="514, 1, 400" offset="22" size="411200" />
305
+ <output>
306
+ <port id="0" precision="FP16" names="65">
307
+ <dim>514</dim>
308
+ <dim>1</dim>
309
+ <dim>400</dim>
310
+ </port>
311
+ </output>
312
+ </layer>
313
+ <layer id="26" name="aten::to/Convert_3" type="Convert" version="opset1">
314
+ <data destination_type="f32" />
315
+ <rt_info>
316
+ <attribute name="decompression" version="0" />
317
+ </rt_info>
318
+ <input>
319
+ <port id="0" precision="FP16">
320
+ <dim>514</dim>
321
+ <dim>1</dim>
322
+ <dim>400</dim>
323
+ </port>
324
+ </input>
325
+ <output>
326
+ <port id="1" precision="FP32">
327
+ <dim>514</dim>
328
+ <dim>1</dim>
329
+ <dim>400</dim>
330
+ </port>
331
+ </output>
332
+ </layer>
333
+ <layer id="27" name="aten::_convolution/Convolution_1" type="Convolution" version="opset1">
334
+ <data strides="160" dilations="1" pads_begin="200" pads_end="200" auto_pad="explicit" />
335
+ <input>
336
+ <port id="0" precision="FP32">
337
+ <dim>1</dim>
338
+ <dim>1</dim>
339
+ <dim>240000</dim>
340
+ </port>
341
+ <port id="1" precision="FP32">
342
+ <dim>514</dim>
343
+ <dim>1</dim>
344
+ <dim>400</dim>
345
+ </port>
346
+ </input>
347
+ <output>
348
+ <port id="2" precision="FP32" names="81,y.1">
349
+ <dim>1</dim>
350
+ <dim>514</dim>
351
+ <dim>1501</dim>
352
+ </port>
353
+ </output>
354
+ </layer>
355
+ <layer id="28" name="ShapeOf_36680" type="ShapeOf" version="opset3">
356
+ <data output_type="i64" />
357
+ <input>
358
+ <port id="0" precision="FP32">
359
+ <dim>1</dim>
360
+ <dim>514</dim>
361
+ <dim>1501</dim>
362
+ </port>
363
+ </input>
364
+ <output>
365
+ <port id="1" precision="I64">
366
+ <dim>3</dim>
367
+ </port>
368
+ </output>
369
+ </layer>
370
+ <layer id="29" name="Constant_36681" type="Const" version="opset1">
371
+ <data element_type="i64" shape="1" offset="411222" size="8" />
372
+ <output>
373
+ <port id="0" precision="I64">
374
+ <dim>1</dim>
375
+ </port>
376
+ </output>
377
+ </layer>
378
+ <layer id="30" name="Constant_36682" type="Const" version="opset1">
379
+ <data element_type="i64" shape="" offset="411230" size="8" />
380
+ <output>
381
+ <port id="0" precision="I64" />
382
+ </output>
383
+ </layer>
384
+ <layer id="31" name="Gather_36683" type="Gather" version="opset8">
385
+ <data batch_dims="0" />
386
+ <input>
387
+ <port id="0" precision="I64">
388
+ <dim>3</dim>
389
+ </port>
390
+ <port id="1" precision="I64">
391
+ <dim>1</dim>
392
+ </port>
393
+ <port id="2" precision="I64" />
394
+ </input>
395
+ <output>
396
+ <port id="3" precision="I64" names="87,89">
397
+ <dim>1</dim>
398
+ </port>
399
+ </output>
400
+ </layer>
401
+ <layer id="32" name="Squeeze_36691" type="Squeeze" version="opset1">
402
+ <input>
403
+ <port id="0" precision="I64">
404
+ <dim>1</dim>
405
+ </port>
406
+ </input>
407
+ <output>
408
+ <port id="1" precision="I64" names="156,219" />
409
+ </output>
410
+ </layer>
411
+ <layer id="33" name="aten::arange/Constant_3" type="Const" version="opset1">
412
+ <data element_type="i32" shape="" offset="411238" size="4" />
413
+ <output>
414
+ <port id="0" precision="I32" />
415
+ </output>
416
+ </layer>
417
+ <layer id="34" name="aten::arange/Range_1" type="Range" version="opset4">
418
+ <data output_type="f32" />
419
+ <input>
420
+ <port id="0" precision="I32" />
421
+ <port id="1" precision="I64" />
422
+ <port id="2" precision="I32" />
423
+ </input>
424
+ <output>
425
+ <port id="3" precision="FP32">
426
+ <dim>1501</dim>
427
+ </port>
428
+ </output>
429
+ </layer>
430
+ <layer id="35" name="aten::arange/ConvertLike_1" type="Convert" version="opset1">
431
+ <data destination_type="i64" />
432
+ <input>
433
+ <port id="0" precision="FP32">
434
+ <dim>1501</dim>
435
+ </port>
436
+ </input>
437
+ <output>
438
+ <port id="1" precision="I64" names="163,226">
439
+ <dim>1501</dim>
440
+ </port>
441
+ </output>
442
+ </layer>
443
+ <layer id="36" name="aten::to/Convert_2_compressed" type="Const" version="opset1">
444
+ <data element_type="f16" shape="128, 257" offset="411242" size="65792" />
445
+ <output>
446
+ <port id="0" precision="FP16" names="149">
447
+ <dim>128</dim>
448
+ <dim>257</dim>
449
+ </port>
450
+ </output>
451
+ </layer>
452
+ <layer id="37" name="aten::to/Convert_2" type="Convert" version="opset1">
453
+ <data destination_type="f32" />
454
+ <rt_info>
455
+ <attribute name="decompression" version="0" />
456
+ </rt_info>
457
+ <input>
458
+ <port id="0" precision="FP16">
459
+ <dim>128</dim>
460
+ <dim>257</dim>
461
+ </port>
462
+ </input>
463
+ <output>
464
+ <port id="1" precision="FP32">
465
+ <dim>128</dim>
466
+ <dim>257</dim>
467
+ </port>
468
+ </output>
469
+ </layer>
470
+ <layer id="38" name="Constant_36698" type="Const" version="opset1">
471
+ <data element_type="i64" shape="4" offset="477034" size="32" />
472
+ <rt_info>
473
+ <attribute name="precise" version="0" />
474
+ </rt_info>
475
+ <output>
476
+ <port id="0" precision="I64">
477
+ <dim>4</dim>
478
+ </port>
479
+ </output>
480
+ </layer>
481
+ <layer id="39" name="aten::view/Reshape" type="Reshape" version="opset1">
482
+ <data special_zero="true" />
483
+ <input>
484
+ <port id="0" precision="FP32">
485
+ <dim>1</dim>
486
+ <dim>514</dim>
487
+ <dim>1501</dim>
488
+ </port>
489
+ <port id="1" precision="I64">
490
+ <dim>4</dim>
491
+ </port>
492
+ </input>
493
+ <output>
494
+ <port id="2" precision="FP32" names="103,116,121,93,98,y">
495
+ <dim>1</dim>
496
+ <dim>257</dim>
497
+ <dim>2</dim>
498
+ <dim>1501</dim>
499
+ </port>
500
+ </output>
501
+ </layer>
502
+ <layer id="40" name="105" type="Const" version="opset1">
503
+ <data element_type="i64" shape="" offset="411230" size="8" />
504
+ <output>
505
+ <port id="0" precision="I64" names="105" />
506
+ </output>
507
+ </layer>
508
+ <layer id="41" name="104" type="Const" version="opset1">
509
+ <data element_type="i64" shape="" offset="411222" size="8" />
510
+ <output>
511
+ <port id="0" precision="I64" names="104" />
512
+ </output>
513
+ </layer>
514
+ <layer id="42" name="aten::select/Gather" type="Gather" version="opset8">
515
+ <data batch_dims="0" />
516
+ <input>
517
+ <port id="0" precision="FP32">
518
+ <dim>1</dim>
519
+ <dim>257</dim>
520
+ <dim>2</dim>
521
+ <dim>1501</dim>
522
+ </port>
523
+ <port id="1" precision="I64" />
524
+ <port id="2" precision="I64" />
525
+ </input>
526
+ <output>
527
+ <port id="3" precision="FP32" names="106,111,real">
528
+ <dim>1</dim>
529
+ <dim>257</dim>
530
+ <dim>1501</dim>
531
+ </port>
532
+ </output>
533
+ </layer>
534
+ <layer id="43" name="aten::mul/Multiply" type="Multiply" version="opset1">
535
+ <data auto_broadcast="numpy" />
536
+ <input>
537
+ <port id="0" precision="FP32">
538
+ <dim>1</dim>
539
+ <dim>257</dim>
540
+ <dim>1501</dim>
541
+ </port>
542
+ <port id="1" precision="FP32">
543
+ <dim>1</dim>
544
+ <dim>257</dim>
545
+ <dim>1501</dim>
546
+ </port>
547
+ </input>
548
+ <output>
549
+ <port id="2" precision="FP32" names="130">
550
+ <dim>1</dim>
551
+ <dim>257</dim>
552
+ <dim>1501</dim>
553
+ </port>
554
+ </output>
555
+ </layer>
556
+ <layer id="44" name="123" type="Const" version="opset1">
557
+ <data element_type="i64" shape="" offset="10" size="8" />
558
+ <output>
559
+ <port id="0" precision="I64" names="123" />
560
+ </output>
561
+ </layer>
562
+ <layer id="45" name="122" type="Const" version="opset1">
563
+ <data element_type="i64" shape="" offset="411222" size="8" />
564
+ <output>
565
+ <port id="0" precision="I64" names="122" />
566
+ </output>
567
+ </layer>
568
+ <layer id="46" name="aten::select/Gather_1" type="Gather" version="opset8">
569
+ <data batch_dims="0" />
570
+ <input>
571
+ <port id="0" precision="FP32">
572
+ <dim>1</dim>
573
+ <dim>257</dim>
574
+ <dim>2</dim>
575
+ <dim>1501</dim>
576
+ </port>
577
+ <port id="1" precision="I64" />
578
+ <port id="2" precision="I64" />
579
+ </input>
580
+ <output>
581
+ <port id="3" precision="FP32" names="124,129,imag">
582
+ <dim>1</dim>
583
+ <dim>257</dim>
584
+ <dim>1501</dim>
585
+ </port>
586
+ </output>
587
+ </layer>
588
+ <layer id="47" name="aten::mul/Multiply_1" type="Multiply" version="opset1">
589
+ <data auto_broadcast="numpy" />
590
+ <input>
591
+ <port id="0" precision="FP32">
592
+ <dim>1</dim>
593
+ <dim>257</dim>
594
+ <dim>1501</dim>
595
+ </port>
596
+ <port id="1" precision="FP32">
597
+ <dim>1</dim>
598
+ <dim>257</dim>
599
+ <dim>1501</dim>
600
+ </port>
601
+ </input>
602
+ <output>
603
+ <port id="2" precision="FP32" names="131">
604
+ <dim>1</dim>
605
+ <dim>257</dim>
606
+ <dim>1501</dim>
607
+ </port>
608
+ </output>
609
+ </layer>
610
+ <layer id="48" name="aten::add/Add_1" type="Add" version="opset1">
611
+ <data auto_broadcast="numpy" />
612
+ <input>
613
+ <port id="0" precision="FP32">
614
+ <dim>1</dim>
615
+ <dim>257</dim>
616
+ <dim>1501</dim>
617
+ </port>
618
+ <port id="1" precision="FP32">
619
+ <dim>1</dim>
620
+ <dim>257</dim>
621
+ <dim>1501</dim>
622
+ </port>
623
+ </input>
624
+ <output>
625
+ <port id="2" precision="FP32" names="133,136">
626
+ <dim>1</dim>
627
+ <dim>257</dim>
628
+ <dim>1501</dim>
629
+ </port>
630
+ </output>
631
+ </layer>
632
+ <layer id="49" name="Constant_36658_compressed" type="Const" version="opset1">
633
+ <data element_type="f16" shape="1, 1, 1" offset="477066" size="2" />
634
+ <output>
635
+ <port id="0" precision="FP16">
636
+ <dim>1</dim>
637
+ <dim>1</dim>
638
+ <dim>1</dim>
639
+ </port>
640
+ </output>
641
+ </layer>
642
+ <layer id="50" name="Constant_36658" type="Convert" version="opset1">
643
+ <data destination_type="f32" />
644
+ <rt_info>
645
+ <attribute name="decompression" version="0" />
646
+ </rt_info>
647
+ <input>
648
+ <port id="0" precision="FP16">
649
+ <dim>1</dim>
650
+ <dim>1</dim>
651
+ <dim>1</dim>
652
+ </port>
653
+ </input>
654
+ <output>
655
+ <port id="1" precision="FP32">
656
+ <dim>1</dim>
657
+ <dim>1</dim>
658
+ <dim>1</dim>
659
+ </port>
660
+ </output>
661
+ </layer>
662
+ <layer id="51" name="aten::pow/Power" type="Power" version="opset1">
663
+ <data auto_broadcast="numpy" />
664
+ <input>
665
+ <port id="0" precision="FP32">
666
+ <dim>1</dim>
667
+ <dim>257</dim>
668
+ <dim>1501</dim>
669
+ </port>
670
+ <port id="1" precision="FP32">
671
+ <dim>1</dim>
672
+ <dim>1</dim>
673
+ <dim>1</dim>
674
+ </port>
675
+ </input>
676
+ <output>
677
+ <port id="2" precision="FP32" names="138">
678
+ <dim>1</dim>
679
+ <dim>257</dim>
680
+ <dim>1501</dim>
681
+ </port>
682
+ </output>
683
+ </layer>
684
+ <layer id="52" name="Constant_36659_compressed" type="Const" version="opset1">
685
+ <data element_type="f16" shape="1, 1, 1" offset="477068" size="2" />
686
+ <output>
687
+ <port id="0" precision="FP16">
688
+ <dim>1</dim>
689
+ <dim>1</dim>
690
+ <dim>1</dim>
691
+ </port>
692
+ </output>
693
+ </layer>
694
+ <layer id="53" name="Constant_36659" type="Convert" version="opset1">
695
+ <data destination_type="f32" />
696
+ <rt_info>
697
+ <attribute name="decompression" version="0" />
698
+ </rt_info>
699
+ <input>
700
+ <port id="0" precision="FP16">
701
+ <dim>1</dim>
702
+ <dim>1</dim>
703
+ <dim>1</dim>
704
+ </port>
705
+ </input>
706
+ <output>
707
+ <port id="1" precision="FP32">
708
+ <dim>1</dim>
709
+ <dim>1</dim>
710
+ <dim>1</dim>
711
+ </port>
712
+ </output>
713
+ </layer>
714
+ <layer id="54" name="aten::pow/Power_1" type="Power" version="opset1">
715
+ <data auto_broadcast="numpy" />
716
+ <input>
717
+ <port id="0" precision="FP32">
718
+ <dim>1</dim>
719
+ <dim>257</dim>
720
+ <dim>1501</dim>
721
+ </port>
722
+ <port id="1" precision="FP32">
723
+ <dim>1</dim>
724
+ <dim>1</dim>
725
+ <dim>1</dim>
726
+ </port>
727
+ </input>
728
+ <output>
729
+ <port id="2" precision="FP32" names="140,x.7">
730
+ <dim>1</dim>
731
+ <dim>257</dim>
732
+ <dim>1501</dim>
733
+ </port>
734
+ </output>
735
+ </layer>
736
+ <layer id="55" name="aten::matmul/MatMul" type="MatMul" version="opset1">
737
+ <data transpose_a="false" transpose_b="false" />
738
+ <input>
739
+ <port id="0" precision="FP32">
740
+ <dim>128</dim>
741
+ <dim>257</dim>
742
+ </port>
743
+ <port id="1" precision="FP32">
744
+ <dim>1</dim>
745
+ <dim>257</dim>
746
+ <dim>1501</dim>
747
+ </port>
748
+ </input>
749
+ <output>
750
+ <port id="2" precision="FP32" names="150,x.9">
751
+ <dim>1</dim>
752
+ <dim>128</dim>
753
+ <dim>1501</dim>
754
+ </port>
755
+ </output>
756
+ </layer>
757
+ <layer id="56" name="Constant_36660_compressed" type="Const" version="opset1">
758
+ <data element_type="f16" shape="1, 1, 1" offset="477070" size="2" />
759
+ <output>
760
+ <port id="0" precision="FP16">
761
+ <dim>1</dim>
762
+ <dim>1</dim>
763
+ <dim>1</dim>
764
+ </port>
765
+ </output>
766
+ </layer>
767
+ <layer id="57" name="Constant_36660" type="Convert" version="opset1">
768
+ <data destination_type="f32" />
769
+ <rt_info>
770
+ <attribute name="decompression" version="0" />
771
+ </rt_info>
772
+ <input>
773
+ <port id="0" precision="FP16">
774
+ <dim>1</dim>
775
+ <dim>1</dim>
776
+ <dim>1</dim>
777
+ </port>
778
+ </input>
779
+ <output>
780
+ <port id="1" precision="FP32">
781
+ <dim>1</dim>
782
+ <dim>1</dim>
783
+ <dim>1</dim>
784
+ </port>
785
+ </output>
786
+ </layer>
787
+ <layer id="58" name="aten::add/Add_3" type="Add" version="opset1">
788
+ <data auto_broadcast="numpy" />
789
+ <input>
790
+ <port id="0" precision="FP32">
791
+ <dim>1</dim>
792
+ <dim>128</dim>
793
+ <dim>1501</dim>
794
+ </port>
795
+ <port id="1" precision="FP32">
796
+ <dim>1</dim>
797
+ <dim>1</dim>
798
+ <dim>1</dim>
799
+ </port>
800
+ </input>
801
+ <output>
802
+ <port id="2" precision="FP32" names="153">
803
+ <dim>1</dim>
804
+ <dim>128</dim>
805
+ <dim>1501</dim>
806
+ </port>
807
+ </output>
808
+ </layer>
809
+ <layer id="59" name="aten::log/Log" type="Log" version="opset1">
810
+ <input>
811
+ <port id="0" precision="FP32">
812
+ <dim>1</dim>
813
+ <dim>128</dim>
814
+ <dim>1501</dim>
815
+ </port>
816
+ </input>
817
+ <output>
818
+ <port id="1" precision="FP32" names="154,x.11">
819
+ <dim>1</dim>
820
+ <dim>128</dim>
821
+ <dim>1501</dim>
822
+ </port>
823
+ </output>
824
+ </layer>
825
+ <layer id="60" name="164" type="Const" version="opset1">
826
+ <data element_type="i64" shape="" offset="411230" size="8" />
827
+ <output>
828
+ <port id="0" precision="I64" names="164" />
829
+ </output>
830
+ </layer>
831
+ <layer id="61" name="aten::unsqueeze/Unsqueeze_2" type="Unsqueeze" version="opset1">
832
+ <input>
833
+ <port id="0" precision="I64">
834
+ <dim>1501</dim>
835
+ </port>
836
+ <port id="1" precision="I64" />
837
+ </input>
838
+ <output>
839
+ <port id="2" precision="I64" names="165">
840
+ <dim>1</dim>
841
+ <dim>1501</dim>
842
+ </port>
843
+ </output>
844
+ </layer>
845
+ <layer id="62" name="166" type="Const" version="opset1">
846
+ <data element_type="i64" shape="" offset="10" size="8" />
847
+ <output>
848
+ <port id="0" precision="I64" names="166" />
849
+ </output>
850
+ </layer>
851
+ <layer id="63" name="aten::unsqueeze/Unsqueeze_3" type="Unsqueeze" version="opset1">
852
+ <input>
853
+ <port id="0" precision="I64">
854
+ <dim>1</dim>
855
+ </port>
856
+ <port id="1" precision="I64" />
857
+ </input>
858
+ <output>
859
+ <port id="2" precision="I64" names="167,235">
860
+ <dim>1</dim>
861
+ <dim>1</dim>
862
+ </port>
863
+ </output>
864
+ </layer>
865
+ <layer id="64" name="aten::lt/Less" type="Less" version="opset1">
866
+ <data auto_broadcast="numpy" />
867
+ <input>
868
+ <port id="0" precision="I64">
869
+ <dim>1</dim>
870
+ <dim>1501</dim>
871
+ </port>
872
+ <port id="1" precision="I64">
873
+ <dim>1</dim>
874
+ <dim>1</dim>
875
+ </port>
876
+ </input>
877
+ <output>
878
+ <port id="2" precision="BOOL" names="168,valid_mask">
879
+ <dim>1</dim>
880
+ <dim>1501</dim>
881
+ </port>
882
+ </output>
883
+ </layer>
884
+ <layer id="65" name="169" type="Const" version="opset1">
885
+ <data element_type="i64" shape="" offset="10" size="8" />
886
+ <output>
887
+ <port id="0" precision="I64" names="169" />
888
+ </output>
889
+ </layer>
890
+ <layer id="66" name="aten::unsqueeze/Unsqueeze_4" type="Unsqueeze" version="opset1">
891
+ <input>
892
+ <port id="0" precision="BOOL">
893
+ <dim>1</dim>
894
+ <dim>1501</dim>
895
+ </port>
896
+ <port id="1" precision="I64" />
897
+ </input>
898
+ <output>
899
+ <port id="2" precision="BOOL" names="170,187">
900
+ <dim>1</dim>
901
+ <dim>1</dim>
902
+ <dim>1501</dim>
903
+ </port>
904
+ </output>
905
+ </layer>
906
+ <layer id="67" name="171_compressed" type="Const" version="opset1">
907
+ <data element_type="f16" shape="" offset="477072" size="2" />
908
+ <output>
909
+ <port id="0" precision="FP16" names="171" />
910
+ </output>
911
+ </layer>
912
+ <layer id="68" name="171" type="Convert" version="opset1">
913
+ <data destination_type="f32" />
914
+ <rt_info>
915
+ <attribute name="decompression" version="0" />
916
+ </rt_info>
917
+ <input>
918
+ <port id="0" precision="FP16" />
919
+ </input>
920
+ <output>
921
+ <port id="1" precision="FP32" />
922
+ </output>
923
+ </layer>
924
+ <layer id="69" name="aten::where/Select" type="Select" version="opset1">
925
+ <data auto_broadcast="numpy" />
926
+ <input>
927
+ <port id="0" precision="BOOL">
928
+ <dim>1</dim>
929
+ <dim>1</dim>
930
+ <dim>1501</dim>
931
+ </port>
932
+ <port id="1" precision="FP32">
933
+ <dim>1</dim>
934
+ <dim>128</dim>
935
+ <dim>1501</dim>
936
+ </port>
937
+ <port id="2" precision="FP32" />
938
+ </input>
939
+ <output>
940
+ <port id="3" precision="FP32" names="172">
941
+ <dim>1</dim>
942
+ <dim>128</dim>
943
+ <dim>1501</dim>
944
+ </port>
945
+ </output>
946
+ </layer>
947
+ <layer id="70" name="Constant_34003" type="Const" version="opset1">
948
+ <data element_type="i64" shape="1" offset="411222" size="8" />
949
+ <output>
950
+ <port id="0" precision="I64" names="174">
951
+ <dim>1</dim>
952
+ </port>
953
+ </output>
954
+ </layer>
955
+ <layer id="71" name="aten::sum/ReduceSum" type="ReduceSum" version="opset1">
956
+ <data keep_dims="false" />
957
+ <input>
958
+ <port id="0" precision="FP32">
959
+ <dim>1</dim>
960
+ <dim>128</dim>
961
+ <dim>1501</dim>
962
+ </port>
963
+ <port id="1" precision="I64">
964
+ <dim>1</dim>
965
+ </port>
966
+ </input>
967
+ <output>
968
+ <port id="2" precision="FP32" names="177,x_sum">
969
+ <dim>1</dim>
970
+ <dim>128</dim>
971
+ </port>
972
+ </output>
973
+ </layer>
974
+ <layer id="72" name="aten::to/Convert_4" type="Convert" version="opset1">
975
+ <data destination_type="f32" />
976
+ <input>
977
+ <port id="0" precision="I64">
978
+ <dim>1</dim>
979
+ </port>
980
+ </input>
981
+ <output>
982
+ <port id="1" precision="FP32" names="182,denom">
983
+ <dim>1</dim>
984
+ </port>
985
+ </output>
986
+ </layer>
987
+ <layer id="73" name="183" type="Const" version="opset1">
988
+ <data element_type="i64" shape="" offset="10" size="8" />
989
+ <output>
990
+ <port id="0" precision="I64" names="183" />
991
+ </output>
992
+ </layer>
993
+ <layer id="74" name="aten::unsqueeze/Unsqueeze_5" type="Unsqueeze" version="opset1">
994
+ <input>
995
+ <port id="0" precision="FP32">
996
+ <dim>1</dim>
997
+ </port>
998
+ <port id="1" precision="I64" />
999
+ </input>
1000
+ <output>
1001
+ <port id="2" precision="FP32" names="184,202">
1002
+ <dim>1</dim>
1003
+ <dim>1</dim>
1004
+ </port>
1005
+ </output>
1006
+ </layer>
1007
+ <layer id="75" name="aten::div/Divide_1" type="Divide" version="opset1">
1008
+ <data auto_broadcast="numpy" m_pythondiv="true" />
1009
+ <input>
1010
+ <port id="0" precision="FP32">
1011
+ <dim>1</dim>
1012
+ <dim>128</dim>
1013
+ </port>
1014
+ <port id="1" precision="FP32">
1015
+ <dim>1</dim>
1016
+ <dim>1</dim>
1017
+ </port>
1018
+ </input>
1019
+ <output>
1020
+ <port id="2" precision="FP32" names="185,x_mean">
1021
+ <dim>1</dim>
1022
+ <dim>128</dim>
1023
+ </port>
1024
+ </output>
1025
+ </layer>
1026
+ <layer id="76" name="211" type="Const" version="opset1">
1027
+ <data element_type="i64" shape="" offset="411222" size="8" />
1028
+ <output>
1029
+ <port id="0" precision="I64" names="211" />
1030
+ </output>
1031
+ </layer>
1032
+ <layer id="77" name="aten::unsqueeze/Unsqueeze_6" type="Unsqueeze" version="opset1">
1033
+ <input>
1034
+ <port id="0" precision="FP32">
1035
+ <dim>1</dim>
1036
+ <dim>128</dim>
1037
+ </port>
1038
+ <port id="1" precision="I64" />
1039
+ </input>
1040
+ <output>
1041
+ <port id="2" precision="FP32" names="189,212">
1042
+ <dim>1</dim>
1043
+ <dim>128</dim>
1044
+ <dim>1</dim>
1045
+ </port>
1046
+ </output>
1047
+ </layer>
1048
+ <layer id="78" name="aten::sub/Subtract_1" type="Subtract" version="opset1">
1049
+ <data auto_broadcast="numpy" />
1050
+ <input>
1051
+ <port id="0" precision="FP32">
1052
+ <dim>1</dim>
1053
+ <dim>128</dim>
1054
+ <dim>1501</dim>
1055
+ </port>
1056
+ <port id="1" precision="FP32">
1057
+ <dim>1</dim>
1058
+ <dim>128</dim>
1059
+ <dim>1</dim>
1060
+ </port>
1061
+ </input>
1062
+ <output>
1063
+ <port id="2" precision="FP32" names="191,214">
1064
+ <dim>1</dim>
1065
+ <dim>128</dim>
1066
+ <dim>1501</dim>
1067
+ </port>
1068
+ </output>
1069
+ </layer>
1070
+ <layer id="79" name="192_compressed" type="Const" version="opset1">
1071
+ <data element_type="f16" shape="" offset="477072" size="2" />
1072
+ <output>
1073
+ <port id="0" precision="FP16" names="192" />
1074
+ </output>
1075
+ </layer>
1076
+ <layer id="80" name="192" type="Convert" version="opset1">
1077
+ <data destination_type="f32" />
1078
+ <rt_info>
1079
+ <attribute name="decompression" version="0" />
1080
+ </rt_info>
1081
+ <input>
1082
+ <port id="0" precision="FP16" />
1083
+ </input>
1084
+ <output>
1085
+ <port id="1" precision="FP32" />
1086
+ </output>
1087
+ </layer>
1088
+ <layer id="81" name="aten::where/Select_1" type="Select" version="opset1">
1089
+ <data auto_broadcast="numpy" />
1090
+ <input>
1091
+ <port id="0" precision="BOOL">
1092
+ <dim>1</dim>
1093
+ <dim>1</dim>
1094
+ <dim>1501</dim>
1095
+ </port>
1096
+ <port id="1" precision="FP32">
1097
+ <dim>1</dim>
1098
+ <dim>128</dim>
1099
+ <dim>1501</dim>
1100
+ </port>
1101
+ <port id="2" precision="FP32" />
1102
+ </input>
1103
+ <output>
1104
+ <port id="3" precision="FP32" names="193">
1105
+ <dim>1</dim>
1106
+ <dim>128</dim>
1107
+ <dim>1501</dim>
1108
+ </port>
1109
+ </output>
1110
+ </layer>
1111
+ <layer id="82" name="Constant_36661_compressed" type="Const" version="opset1">
1112
+ <data element_type="f16" shape="1, 1, 1" offset="477068" size="2" />
1113
+ <output>
1114
+ <port id="0" precision="FP16">
1115
+ <dim>1</dim>
1116
+ <dim>1</dim>
1117
+ <dim>1</dim>
1118
+ </port>
1119
+ </output>
1120
+ </layer>
1121
+ <layer id="83" name="Constant_36661" type="Convert" version="opset1">
1122
+ <data destination_type="f32" />
1123
+ <rt_info>
1124
+ <attribute name="decompression" version="0" />
1125
+ </rt_info>
1126
+ <input>
1127
+ <port id="0" precision="FP16">
1128
+ <dim>1</dim>
1129
+ <dim>1</dim>
1130
+ <dim>1</dim>
1131
+ </port>
1132
+ </input>
1133
+ <output>
1134
+ <port id="1" precision="FP32">
1135
+ <dim>1</dim>
1136
+ <dim>1</dim>
1137
+ <dim>1</dim>
1138
+ </port>
1139
+ </output>
1140
+ </layer>
1141
+ <layer id="84" name="aten::pow/Power_2" type="Power" version="opset1">
1142
+ <data auto_broadcast="numpy" />
1143
+ <input>
1144
+ <port id="0" precision="FP32">
1145
+ <dim>1</dim>
1146
+ <dim>128</dim>
1147
+ <dim>1501</dim>
1148
+ </port>
1149
+ <port id="1" precision="FP32">
1150
+ <dim>1</dim>
1151
+ <dim>1</dim>
1152
+ <dim>1</dim>
1153
+ </port>
1154
+ </input>
1155
+ <output>
1156
+ <port id="2" precision="FP32" names="195">
1157
+ <dim>1</dim>
1158
+ <dim>128</dim>
1159
+ <dim>1501</dim>
1160
+ </port>
1161
+ </output>
1162
+ </layer>
1163
+ <layer id="85" name="Constant_34049" type="Const" version="opset1">
1164
+ <data element_type="i64" shape="1" offset="411222" size="8" />
1165
+ <output>
1166
+ <port id="0" precision="I64" names="197">
1167
+ <dim>1</dim>
1168
+ </port>
1169
+ </output>
1170
+ </layer>
1171
+ <layer id="86" name="aten::sum/ReduceSum_1" type="ReduceSum" version="opset1">
1172
+ <data keep_dims="false" />
1173
+ <input>
1174
+ <port id="0" precision="FP32">
1175
+ <dim>1</dim>
1176
+ <dim>128</dim>
1177
+ <dim>1501</dim>
1178
+ </port>
1179
+ <port id="1" precision="I64">
1180
+ <dim>1</dim>
1181
+ </port>
1182
+ </input>
1183
+ <output>
1184
+ <port id="2" precision="FP32" names="200">
1185
+ <dim>1</dim>
1186
+ <dim>128</dim>
1187
+ </port>
1188
+ </output>
1189
+ </layer>
1190
+ <layer id="87" name="Constant_36662_compressed" type="Const" version="opset1">
1191
+ <data element_type="f16" shape="1, 1" offset="477074" size="2" />
1192
+ <output>
1193
+ <port id="0" precision="FP16">
1194
+ <dim>1</dim>
1195
+ <dim>1</dim>
1196
+ </port>
1197
+ </output>
1198
+ </layer>
1199
+ <layer id="88" name="Constant_36662" type="Convert" version="opset1">
1200
+ <data destination_type="f32" />
1201
+ <rt_info>
1202
+ <attribute name="decompression" version="0" />
1203
+ </rt_info>
1204
+ <input>
1205
+ <port id="0" precision="FP16">
1206
+ <dim>1</dim>
1207
+ <dim>1</dim>
1208
+ </port>
1209
+ </input>
1210
+ <output>
1211
+ <port id="1" precision="FP32">
1212
+ <dim>1</dim>
1213
+ <dim>1</dim>
1214
+ </port>
1215
+ </output>
1216
+ </layer>
1217
+ <layer id="89" name="aten::sub/Subtract_3" type="Add" version="opset1">
1218
+ <data auto_broadcast="numpy" />
1219
+ <input>
1220
+ <port id="0" precision="FP32">
1221
+ <dim>1</dim>
1222
+ <dim>1</dim>
1223
+ </port>
1224
+ <port id="1" precision="FP32">
1225
+ <dim>1</dim>
1226
+ <dim>1</dim>
1227
+ </port>
1228
+ </input>
1229
+ <output>
1230
+ <port id="2" precision="FP32" names="205">
1231
+ <dim>1</dim>
1232
+ <dim>1</dim>
1233
+ </port>
1234
+ </output>
1235
+ </layer>
1236
+ <layer id="90" name="aten::div/Divide_2" type="Divide" version="opset1">
1237
+ <data auto_broadcast="numpy" m_pythondiv="true" />
1238
+ <input>
1239
+ <port id="0" precision="FP32">
1240
+ <dim>1</dim>
1241
+ <dim>128</dim>
1242
+ </port>
1243
+ <port id="1" precision="FP32">
1244
+ <dim>1</dim>
1245
+ <dim>1</dim>
1246
+ </port>
1247
+ </input>
1248
+ <output>
1249
+ <port id="2" precision="FP32" names="206">
1250
+ <dim>1</dim>
1251
+ <dim>128</dim>
1252
+ </port>
1253
+ </output>
1254
+ </layer>
1255
+ <layer id="91" name="aten::sqrt/Sqrt" type="Sqrt" version="opset1">
1256
+ <input>
1257
+ <port id="0" precision="FP32">
1258
+ <dim>1</dim>
1259
+ <dim>128</dim>
1260
+ </port>
1261
+ </input>
1262
+ <output>
1263
+ <port id="1" precision="FP32" names="207,x_std.1">
1264
+ <dim>1</dim>
1265
+ <dim>128</dim>
1266
+ </port>
1267
+ </output>
1268
+ </layer>
1269
+ <layer id="92" name="Constant_36663_compressed" type="Const" version="opset1">
1270
+ <data element_type="f16" shape="1, 1" offset="477076" size="2" />
1271
+ <output>
1272
+ <port id="0" precision="FP16">
1273
+ <dim>1</dim>
1274
+ <dim>1</dim>
1275
+ </port>
1276
+ </output>
1277
+ </layer>
1278
+ <layer id="93" name="Constant_36663" type="Convert" version="opset1">
1279
+ <data destination_type="f32" />
1280
+ <rt_info>
1281
+ <attribute name="decompression" version="0" />
1282
+ </rt_info>
1283
+ <input>
1284
+ <port id="0" precision="FP16">
1285
+ <dim>1</dim>
1286
+ <dim>1</dim>
1287
+ </port>
1288
+ </input>
1289
+ <output>
1290
+ <port id="1" precision="FP32">
1291
+ <dim>1</dim>
1292
+ <dim>1</dim>
1293
+ </port>
1294
+ </output>
1295
+ </layer>
1296
+ <layer id="94" name="aten::add/Add_4" type="Add" version="opset1">
1297
+ <data auto_broadcast="numpy" />
1298
+ <input>
1299
+ <port id="0" precision="FP32">
1300
+ <dim>1</dim>
1301
+ <dim>128</dim>
1302
+ </port>
1303
+ <port id="1" precision="FP32">
1304
+ <dim>1</dim>
1305
+ <dim>1</dim>
1306
+ </port>
1307
+ </input>
1308
+ <output>
1309
+ <port id="2" precision="FP32" names="210,x_std">
1310
+ <dim>1</dim>
1311
+ <dim>128</dim>
1312
+ </port>
1313
+ </output>
1314
+ </layer>
1315
+ <layer id="95" name="215" type="Const" version="opset1">
1316
+ <data element_type="i64" shape="" offset="411222" size="8" />
1317
+ <output>
1318
+ <port id="0" precision="I64" names="215" />
1319
+ </output>
1320
+ </layer>
1321
+ <layer id="96" name="aten::unsqueeze/Unsqueeze_10" type="Unsqueeze" version="opset1">
1322
+ <input>
1323
+ <port id="0" precision="FP32">
1324
+ <dim>1</dim>
1325
+ <dim>128</dim>
1326
+ </port>
1327
+ <port id="1" precision="I64" />
1328
+ </input>
1329
+ <output>
1330
+ <port id="2" precision="FP32" names="216">
1331
+ <dim>1</dim>
1332
+ <dim>128</dim>
1333
+ <dim>1</dim>
1334
+ </port>
1335
+ </output>
1336
+ </layer>
1337
+ <layer id="97" name="aten::div/Divide_3" type="Divide" version="opset1">
1338
+ <data auto_broadcast="numpy" m_pythondiv="true" />
1339
+ <input>
1340
+ <port id="0" precision="FP32">
1341
+ <dim>1</dim>
1342
+ <dim>128</dim>
1343
+ <dim>1501</dim>
1344
+ </port>
1345
+ <port id="1" precision="FP32">
1346
+ <dim>1</dim>
1347
+ <dim>128</dim>
1348
+ <dim>1</dim>
1349
+ </port>
1350
+ </input>
1351
+ <output>
1352
+ <port id="2" precision="FP32" names="217,x">
1353
+ <dim>1</dim>
1354
+ <dim>128</dim>
1355
+ <dim>1501</dim>
1356
+ </port>
1357
+ </output>
1358
+ </layer>
1359
+ <layer id="98" name="ShapeOf_36692" type="ShapeOf" version="opset3">
1360
+ <data output_type="i64" />
1361
+ <input>
1362
+ <port id="0" precision="FP32">
1363
+ <dim>1</dim>
1364
+ <dim>128</dim>
1365
+ <dim>1501</dim>
1366
+ </port>
1367
+ </input>
1368
+ <output>
1369
+ <port id="1" precision="I64">
1370
+ <dim>3</dim>
1371
+ </port>
1372
+ </output>
1373
+ </layer>
1374
+ <layer id="99" name="Constant_36693" type="Const" version="opset1">
1375
+ <data element_type="i64" shape="1" offset="411230" size="8" />
1376
+ <rt_info>
1377
+ <attribute name="precise" version="0" />
1378
+ </rt_info>
1379
+ <output>
1380
+ <port id="0" precision="I64">
1381
+ <dim>1</dim>
1382
+ </port>
1383
+ </output>
1384
+ </layer>
1385
+ <layer id="100" name="Constant_36694" type="Const" version="opset1">
1386
+ <data element_type="i64" shape="" offset="411230" size="8" />
1387
+ <rt_info>
1388
+ <attribute name="precise" version="0" />
1389
+ </rt_info>
1390
+ <output>
1391
+ <port id="0" precision="I64" />
1392
+ </output>
1393
+ </layer>
1394
+ <layer id="101" name="Gather_36695" type="Gather" version="opset8">
1395
+ <data batch_dims="0" />
1396
+ <input>
1397
+ <port id="0" precision="I64">
1398
+ <dim>3</dim>
1399
+ </port>
1400
+ <port id="1" precision="I64">
1401
+ <dim>1</dim>
1402
+ </port>
1403
+ <port id="2" precision="I64" />
1404
+ </input>
1405
+ <output>
1406
+ <port id="3" precision="I64" names="228,230">
1407
+ <dim>1</dim>
1408
+ </port>
1409
+ </output>
1410
+ </layer>
1411
+ <layer id="102" name="Constant_34276" type="Const" version="opset1">
1412
+ <data element_type="i64" shape="1" offset="10" size="8" />
1413
+ <rt_info>
1414
+ <attribute name="precise" version="0" />
1415
+ </rt_info>
1416
+ <output>
1417
+ <port id="0" precision="I64">
1418
+ <dim>1</dim>
1419
+ </port>
1420
+ </output>
1421
+ </layer>
1422
+ <layer id="103" name="prim::ListConstruct/Concat" type="Concat" version="opset1">
1423
+ <data axis="0" />
1424
+ <input>
1425
+ <port id="0" precision="I64">
1426
+ <dim>1</dim>
1427
+ </port>
1428
+ <port id="1" precision="I64">
1429
+ <dim>1</dim>
1430
+ </port>
1431
+ </input>
1432
+ <output>
1433
+ <port id="2" precision="I64" names="232">
1434
+ <dim>2</dim>
1435
+ </port>
1436
+ </output>
1437
+ </layer>
1438
+ <layer id="104" name="aten::repeat/Tile" type="Tile" version="opset1">
1439
+ <input>
1440
+ <port id="0" precision="I64">
1441
+ <dim>1501</dim>
1442
+ </port>
1443
+ <port id="1" precision="I64">
1444
+ <dim>2</dim>
1445
+ </port>
1446
+ </input>
1447
+ <output>
1448
+ <port id="2" precision="I64" names="233">
1449
+ <dim>1</dim>
1450
+ <dim>1501</dim>
1451
+ </port>
1452
+ </output>
1453
+ </layer>
1454
+ <layer id="105" name="aten::ge/GreaterEqual" type="GreaterEqual" version="opset1">
1455
+ <data auto_broadcast="numpy" />
1456
+ <input>
1457
+ <port id="0" precision="I64">
1458
+ <dim>1</dim>
1459
+ <dim>1501</dim>
1460
+ </port>
1461
+ <port id="1" precision="I64">
1462
+ <dim>1</dim>
1463
+ <dim>1</dim>
1464
+ </port>
1465
+ </input>
1466
+ <output>
1467
+ <port id="2" precision="BOOL" names="236,mask">
1468
+ <dim>1</dim>
1469
+ <dim>1501</dim>
1470
+ </port>
1471
+ </output>
1472
+ </layer>
1473
+ <layer id="106" name="237" type="Const" version="opset1">
1474
+ <data element_type="i64" shape="" offset="10" size="8" />
1475
+ <output>
1476
+ <port id="0" precision="I64" names="237" />
1477
+ </output>
1478
+ </layer>
1479
+ <layer id="107" name="aten::unsqueeze/Unsqueeze_12" type="Unsqueeze" version="opset1">
1480
+ <input>
1481
+ <port id="0" precision="BOOL">
1482
+ <dim>1</dim>
1483
+ <dim>1501</dim>
1484
+ </port>
1485
+ <port id="1" precision="I64" />
1486
+ </input>
1487
+ <output>
1488
+ <port id="2" precision="BOOL" names="238">
1489
+ <dim>1</dim>
1490
+ <dim>1</dim>
1491
+ <dim>1501</dim>
1492
+ </port>
1493
+ </output>
1494
+ </layer>
1495
+ <layer id="108" name="aten::masked_fill/ConvertLike_compressed" type="Const" version="opset1">
1496
+ <data element_type="f16" shape="" offset="477072" size="2" />
1497
+ <output>
1498
+ <port id="0" precision="FP16" />
1499
+ </output>
1500
+ </layer>
1501
+ <layer id="109" name="aten::masked_fill/ConvertLike" type="Convert" version="opset1">
1502
+ <data destination_type="f32" />
1503
+ <rt_info>
1504
+ <attribute name="decompression" version="0" />
1505
+ </rt_info>
1506
+ <input>
1507
+ <port id="0" precision="FP16" />
1508
+ </input>
1509
+ <output>
1510
+ <port id="1" precision="FP32" />
1511
+ </output>
1512
+ </layer>
1513
+ <layer id="110" name="aten::masked_fill/Select" type="Select" version="opset1">
1514
+ <data auto_broadcast="numpy" />
1515
+ <input>
1516
+ <port id="0" precision="BOOL">
1517
+ <dim>1</dim>
1518
+ <dim>1</dim>
1519
+ <dim>1501</dim>
1520
+ </port>
1521
+ <port id="1" precision="FP32" />
1522
+ <port id="2" precision="FP32">
1523
+ <dim>1</dim>
1524
+ <dim>128</dim>
1525
+ <dim>1501</dim>
1526
+ </port>
1527
+ </input>
1528
+ <output>
1529
+ <port id="3" precision="FP32">
1530
+ <dim>1</dim>
1531
+ <dim>128</dim>
1532
+ <dim>1501</dim>
1533
+ </port>
1534
+ </output>
1535
+ </layer>
1536
+ <layer id="111" name="Result_34174" type="Result" version="opset1">
1537
+ <input>
1538
+ <port id="0" precision="FP32">
1539
+ <dim>1</dim>
1540
+ <dim>128</dim>
1541
+ <dim>1501</dim>
1542
+ </port>
1543
+ </input>
1544
+ </layer>
1545
+ <layer id="14" name="Result_34173" type="Result" version="opset1" output_names="seq_len">
1546
+ <input>
1547
+ <port id="0" precision="I64">
1548
+ <dim>1</dim>
1549
+ </port>
1550
+ </input>
1551
+ </layer>
1552
+ </layers>
1553
+ <edges>
1554
+ <edge from-layer="0" from-port="0" to-layer="2" to-port="0" />
1555
+ <edge from-layer="1" from-port="0" to-layer="17" to-port="0" />
1556
+ <edge from-layer="2" from-port="1" to-layer="5" to-port="0" />
1557
+ <edge from-layer="3" from-port="0" to-layer="4" to-port="0" />
1558
+ <edge from-layer="4" from-port="1" to-layer="5" to-port="1" />
1559
+ <edge from-layer="5" from-port="2" to-layer="8" to-port="0" />
1560
+ <edge from-layer="6" from-port="0" to-layer="7" to-port="0" />
1561
+ <edge from-layer="7" from-port="1" to-layer="8" to-port="1" />
1562
+ <edge from-layer="8" from-port="2" to-layer="9" to-port="0" />
1563
+ <edge from-layer="9" from-port="1" to-layer="12" to-port="0" />
1564
+ <edge from-layer="10" from-port="0" to-layer="11" to-port="0" />
1565
+ <edge from-layer="11" from-port="1" to-layer="12" to-port="1" />
1566
+ <edge from-layer="12" from-port="2" to-layer="13" to-port="0" />
1567
+ <edge from-layer="13" from-port="1" to-layer="14" to-port="0" />
1568
+ <edge from-layer="13" from-port="1" to-layer="72" to-port="0" />
1569
+ <edge from-layer="13" from-port="1" to-layer="63" to-port="0" />
1570
+ <edge from-layer="15" from-port="0" to-layer="34" to-port="0" />
1571
+ <edge from-layer="16" from-port="0" to-layer="17" to-port="1" />
1572
+ <edge from-layer="17" from-port="2" to-layer="20" to-port="0" />
1573
+ <edge from-layer="18" from-port="0" to-layer="19" to-port="0" />
1574
+ <edge from-layer="19" from-port="1" to-layer="20" to-port="1" />
1575
+ <edge from-layer="20" from-port="2" to-layer="22" to-port="0" />
1576
+ <edge from-layer="21" from-port="0" to-layer="22" to-port="1" />
1577
+ <edge from-layer="22" from-port="2" to-layer="24" to-port="0" />
1578
+ <edge from-layer="23" from-port="0" to-layer="24" to-port="1" />
1579
+ <edge from-layer="24" from-port="2" to-layer="27" to-port="0" />
1580
+ <edge from-layer="25" from-port="0" to-layer="26" to-port="0" />
1581
+ <edge from-layer="26" from-port="1" to-layer="27" to-port="1" />
1582
+ <edge from-layer="27" from-port="2" to-layer="28" to-port="0" />
1583
+ <edge from-layer="27" from-port="2" to-layer="39" to-port="0" />
1584
+ <edge from-layer="28" from-port="1" to-layer="31" to-port="0" />
1585
+ <edge from-layer="29" from-port="0" to-layer="31" to-port="1" />
1586
+ <edge from-layer="30" from-port="0" to-layer="31" to-port="2" />
1587
+ <edge from-layer="31" from-port="3" to-layer="32" to-port="0" />
1588
+ <edge from-layer="32" from-port="1" to-layer="34" to-port="1" />
1589
+ <edge from-layer="33" from-port="0" to-layer="34" to-port="2" />
1590
+ <edge from-layer="34" from-port="3" to-layer="35" to-port="0" />
1591
+ <edge from-layer="35" from-port="1" to-layer="104" to-port="0" />
1592
+ <edge from-layer="35" from-port="1" to-layer="61" to-port="0" />
1593
+ <edge from-layer="36" from-port="0" to-layer="37" to-port="0" />
1594
+ <edge from-layer="37" from-port="1" to-layer="55" to-port="0" />
1595
+ <edge from-layer="38" from-port="0" to-layer="39" to-port="1" />
1596
+ <edge from-layer="39" from-port="2" to-layer="42" to-port="0" />
1597
+ <edge from-layer="39" from-port="2" to-layer="46" to-port="0" />
1598
+ <edge from-layer="40" from-port="0" to-layer="42" to-port="1" />
1599
+ <edge from-layer="41" from-port="0" to-layer="42" to-port="2" />
1600
+ <edge from-layer="42" from-port="3" to-layer="43" to-port="0" />
1601
+ <edge from-layer="42" from-port="3" to-layer="43" to-port="1" />
1602
+ <edge from-layer="43" from-port="2" to-layer="48" to-port="0" />
1603
+ <edge from-layer="44" from-port="0" to-layer="46" to-port="1" />
1604
+ <edge from-layer="45" from-port="0" to-layer="46" to-port="2" />
1605
+ <edge from-layer="46" from-port="3" to-layer="47" to-port="0" />
1606
+ <edge from-layer="46" from-port="3" to-layer="47" to-port="1" />
1607
+ <edge from-layer="47" from-port="2" to-layer="48" to-port="1" />
1608
+ <edge from-layer="48" from-port="2" to-layer="51" to-port="0" />
1609
+ <edge from-layer="49" from-port="0" to-layer="50" to-port="0" />
1610
+ <edge from-layer="50" from-port="1" to-layer="51" to-port="1" />
1611
+ <edge from-layer="51" from-port="2" to-layer="54" to-port="0" />
1612
+ <edge from-layer="52" from-port="0" to-layer="53" to-port="0" />
1613
+ <edge from-layer="53" from-port="1" to-layer="54" to-port="1" />
1614
+ <edge from-layer="54" from-port="2" to-layer="55" to-port="1" />
1615
+ <edge from-layer="55" from-port="2" to-layer="58" to-port="0" />
1616
+ <edge from-layer="56" from-port="0" to-layer="57" to-port="0" />
1617
+ <edge from-layer="57" from-port="1" to-layer="58" to-port="1" />
1618
+ <edge from-layer="58" from-port="2" to-layer="59" to-port="0" />
1619
+ <edge from-layer="59" from-port="1" to-layer="78" to-port="0" />
1620
+ <edge from-layer="59" from-port="1" to-layer="69" to-port="1" />
1621
+ <edge from-layer="60" from-port="0" to-layer="61" to-port="1" />
1622
+ <edge from-layer="61" from-port="2" to-layer="64" to-port="0" />
1623
+ <edge from-layer="62" from-port="0" to-layer="63" to-port="1" />
1624
+ <edge from-layer="63" from-port="2" to-layer="64" to-port="1" />
1625
+ <edge from-layer="63" from-port="2" to-layer="105" to-port="1" />
1626
+ <edge from-layer="64" from-port="2" to-layer="66" to-port="0" />
1627
+ <edge from-layer="65" from-port="0" to-layer="66" to-port="1" />
1628
+ <edge from-layer="66" from-port="2" to-layer="69" to-port="0" />
1629
+ <edge from-layer="66" from-port="2" to-layer="81" to-port="0" />
1630
+ <edge from-layer="67" from-port="0" to-layer="68" to-port="0" />
1631
+ <edge from-layer="68" from-port="1" to-layer="69" to-port="2" />
1632
+ <edge from-layer="69" from-port="3" to-layer="71" to-port="0" />
1633
+ <edge from-layer="70" from-port="0" to-layer="71" to-port="1" />
1634
+ <edge from-layer="71" from-port="2" to-layer="75" to-port="0" />
1635
+ <edge from-layer="72" from-port="1" to-layer="74" to-port="0" />
1636
+ <edge from-layer="73" from-port="0" to-layer="74" to-port="1" />
1637
+ <edge from-layer="74" from-port="2" to-layer="75" to-port="1" />
1638
+ <edge from-layer="74" from-port="2" to-layer="89" to-port="0" />
1639
+ <edge from-layer="75" from-port="2" to-layer="77" to-port="0" />
1640
+ <edge from-layer="76" from-port="0" to-layer="77" to-port="1" />
1641
+ <edge from-layer="77" from-port="2" to-layer="78" to-port="1" />
1642
+ <edge from-layer="78" from-port="2" to-layer="81" to-port="1" />
1643
+ <edge from-layer="78" from-port="2" to-layer="97" to-port="0" />
1644
+ <edge from-layer="79" from-port="0" to-layer="80" to-port="0" />
1645
+ <edge from-layer="80" from-port="1" to-layer="81" to-port="2" />
1646
+ <edge from-layer="81" from-port="3" to-layer="84" to-port="0" />
1647
+ <edge from-layer="82" from-port="0" to-layer="83" to-port="0" />
1648
+ <edge from-layer="83" from-port="1" to-layer="84" to-port="1" />
1649
+ <edge from-layer="84" from-port="2" to-layer="86" to-port="0" />
1650
+ <edge from-layer="85" from-port="0" to-layer="86" to-port="1" />
1651
+ <edge from-layer="86" from-port="2" to-layer="90" to-port="0" />
1652
+ <edge from-layer="87" from-port="0" to-layer="88" to-port="0" />
1653
+ <edge from-layer="88" from-port="1" to-layer="89" to-port="1" />
1654
+ <edge from-layer="89" from-port="2" to-layer="90" to-port="1" />
1655
+ <edge from-layer="90" from-port="2" to-layer="91" to-port="0" />
1656
+ <edge from-layer="91" from-port="1" to-layer="94" to-port="0" />
1657
+ <edge from-layer="92" from-port="0" to-layer="93" to-port="0" />
1658
+ <edge from-layer="93" from-port="1" to-layer="94" to-port="1" />
1659
+ <edge from-layer="94" from-port="2" to-layer="96" to-port="0" />
1660
+ <edge from-layer="95" from-port="0" to-layer="96" to-port="1" />
1661
+ <edge from-layer="96" from-port="2" to-layer="97" to-port="1" />
1662
+ <edge from-layer="97" from-port="2" to-layer="98" to-port="0" />
1663
+ <edge from-layer="97" from-port="2" to-layer="110" to-port="2" />
1664
+ <edge from-layer="98" from-port="1" to-layer="101" to-port="0" />
1665
+ <edge from-layer="99" from-port="0" to-layer="101" to-port="1" />
1666
+ <edge from-layer="100" from-port="0" to-layer="101" to-port="2" />
1667
+ <edge from-layer="101" from-port="3" to-layer="103" to-port="0" />
1668
+ <edge from-layer="102" from-port="0" to-layer="103" to-port="1" />
1669
+ <edge from-layer="103" from-port="2" to-layer="104" to-port="1" />
1670
+ <edge from-layer="104" from-port="2" to-layer="105" to-port="0" />
1671
+ <edge from-layer="105" from-port="2" to-layer="107" to-port="0" />
1672
+ <edge from-layer="106" from-port="0" to-layer="107" to-port="1" />
1673
+ <edge from-layer="107" from-port="2" to-layer="110" to-port="0" />
1674
+ <edge from-layer="108" from-port="0" to-layer="109" to-port="0" />
1675
+ <edge from-layer="109" from-port="1" to-layer="110" to-port="1" />
1676
+ <edge from-layer="110" from-port="3" to-layer="111" to-port="0" />
1677
+ </edges>
1678
+ <rt_info>
1679
+ <Runtime_version value="2025.3.0-19807-44526285f24-releases/2025/3" />
1680
+ <conversion_parameters>
1681
+ <framework value="pytorch" />
1682
+ <is_python_object value="True" />
1683
+ </conversion_parameters>
1684
+ </rt_info>
1685
+ </net>
parakeet_v3_vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
parakeet_vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
voicescribe-model-layout.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": 1,
3
+ "generated_at": "2026-04-23T22:20:54Z",
4
+ "layout_key": "parakeet",
5
+ "target_dir": "parakeet-v3-ov",
6
+ "upstream_repo": "FluidInference/parakeet-tdt-0.6b-v3-ov",
7
+ "upstream_revision": "dfd55eb6c85a9a8546a162bed84784245d5743c2",
8
+ "upstream_sha": "dfd55eb6c85a9a8546a162bed84784245d5743c2",
9
+ "description": "Parakeet-TDT 0.6B v3 Intel/OpenVINO layout.",
10
+ "required_files": [
11
+ "config.json",
12
+ "parakeet_encoder.xml",
13
+ "parakeet_encoder.bin",
14
+ "parakeet_decoder.xml",
15
+ "parakeet_decoder.bin",
16
+ "parakeet_joint.xml",
17
+ "parakeet_joint.bin",
18
+ "parakeet_melspectogram.xml",
19
+ "parakeet_melspectogram.bin",
20
+ "parakeet_vocab.json"
21
+ ],
22
+ "allow_patterns": [],
23
+ "license_metadata": {
24
+ "license": "cc-by-4.0",
25
+ "license_tags": [
26
+ "license:cc-by-4.0"
27
+ ],
28
+ "license_files": []
29
+ }
30
+ }