Paradee-8M v1.0 Core ML: int8 + fp32 (text + acoustic)

.gitattributes

@@ -33,3 +33,8 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+samples/1_paradee_coreml_int8.wav filter=lfs diff=lfs merge=lfs -text
+samples/2_paradee_coreml_int8.wav filter=lfs diff=lfs merge=lfs -text
+samples/3_paradee_coreml_int8.wav filter=lfs diff=lfs merge=lfs -text
+samples/4_paradee_coreml_int8.wav filter=lfs diff=lfs merge=lfs -text
+samples/5_paradee_coreml_int8.wav filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -0,0 +1,86 @@
+---
+base_model: sahilmahendrakar/Paradee-8M-v1.0
+language:
+- en
+library_name: coreml
+license: apache-2.0
+pipeline_tag: text-to-speech
+tags:
+- text-to-speech
+- tts
+- coreml
+- kokoro
+- distillation
+- apple-silicon
+---
+
+# Paradee-8M Core ML
+
+Core ML conversion of [Paradee-8M v1.0](https://huggingface.co/sahilmahendrakar/Paradee-8M-v1.0) by Sahil Mahendrakar:
+Kokoro-82M distilled into an 8.07M-parameter, single-voice (`af_heart`) English TTS model, 24 kHz.
+Paper: [arXiv 2610.06817](https://arxiv.org/abs/2610.06817).
+
+Runs **~150× real time on the CPU of an M5 Pro** (93 s of audio in 0.62 s, text side + acoustic side), against 20×
+for the upstream fp32 ONNX on one onnxruntime thread. The int8 build has a **12 MB weight footprint**.
+
+Used by [FluidAudio](https://github.com/FluidInference/FluidAudio) (`ParadeeManager`).
+
+## Files
+
+| Path | What |
+|---|---|
+| `int8/` | Default. `ParadeeText.mlmodelc` (3.1 MB) + `ParadeeAcoustic.mlmodelc` (9.1 MB), int8 per-channel weights; the phase-lock filter's DFT bases stay fp32 |
+| `fp32/` | Same graphs in fp32 (12 + 22 MB). Matches PyTorch to rounding |
+| `*/vocab.json` | Phoneme → id map (identical to Kokoro's) |
+| `mlpackage/` | Source `.mlpackage`s for both variants |
+| `config.json` | Vocab, sample rate, shape limits |
+| `samples/` | The model card's five held-out sentences, rendered by `int8/` |
+
+## Pipeline
+
+```
+text -> misaki-style en-US phonemes -> ids = [0, ...vocab ids..., 0]       (<= 512 total)
+ParadeeText      input_ids [1,T] int32 -> duration [1,T], d [1,224,T], asr_tok [1,512,T]
+host             n_i = max(1, round(duration_i / speed));  F = sum(n)
+                 en  = d       with column i repeated n_i times   [1,224,F]
+                 asr = asr_tok with column i repeated n_i times   [1,512,F]
+                 noise ~ N(0,1)                                    [1,1,600F]
+ParadeeAcoustic  en, asr, noise -> audio [1,600F] float32, 24 kHz
+```
+
+- `F` may be 1…4000 (100 s at speed 1). Split longer input into chunks of ≤ 510 phonemes.
+- `noise` is the harmonic source's Gaussian noise, passed in so the graph is deterministic. It is equivalent to the
+  upstream in-graph noise (9 per-harmonic channels mixed by a linear layer = one scaled channel).
+- Phonemes must be written the way [misaki](https://github.com/hexgrad/misaki) writes them (Kokoro's G2P); that is
+  all Paradee saw in training.
+- Use `CPU_ONLY` or `CPU_AND_NE`. **Do not use `ALL` / `CPU_AND_GPU`**: the LSTMs abort in MPSGraph
+  (`GPURNNOps … JIT not supported`), the same failure Kokoro's prosody stage has.
+
+## Accuracy (M5 Pro, macOS 27)
+
+10 inputs: the README quick-start line, the five held-out sentences, a numbers/abbreviation sentence, a 28.6 s
+paragraph, and one sentence at speed 0.8 and 1.3.
+
+- fp32 vs PyTorch with the same noise: 0 duration rounding differences, identical lengths. Log-mel L1 against the
+  upstream ONNX equals the ONNX's own run-to-run difference in every case (e.g. 0.123 vs 0.123).
+- Whisper large-v3-turbo transcripts of Core ML fp32 and upstream ONNX fp32 are identical (WER 3.81 %; every miss is
+  text normalization: "Ia", "favourite", "Zzyzx").
+- int8: log-mel L1 to fp32 ONNX 0.18–0.20 vs 0.21–0.31 for upstream `paradee_int8.onnx`; WER 3.81 % vs 2.97 %
+  (one word of ~236).
+
+Conversion notes: the 1024-point phase-lock STFT is a gather + matmul + overlap-add instead of strided convolutions
+(the conv form ran ~40× slower on the Core ML CPU), and kokoro's random harmonic start phases are omitted because the
+upstream graph never reads them.
+
+## License
+
+Apache-2.0, same as Paradee and Kokoro-82M. Paradee © Sahil Mahendrakar.
+
+```bibtex
+@misc{mahendrakar2026paradee,
+  title  = {Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice Text-to-Speech Model},
+  author = {Mahendrakar, Sahil},
+  year   = {2026},
+  url    = {https://huggingface.co/sahilmahendrakar/Paradee-8M-v1.0}
+}
+```

config.json

@@ -0,0 +1,131 @@
+{
+ "base_model": "sahilmahendrakar/Paradee-8M-v1.0",
+ "format": "coreml",
+ "model_type": "paradee",
+ "sample_rate": 24000,
+ "n_token": 178,
+ "max_phonemes": 510,
+ "samples_per_frame": 600,
+ "max_frames": 4000,
+ "default_variant": "int8",
+ "variants": [
+  "int8",
+  "fp32"
+ ],
+ "vocab": {
+  ";": 1,
+  ":": 2,
+  ",": 3,
+  ".": 4,
+  "!": 5,
+  "?": 6,
+  "—": 9,
+  "…": 10,
+  "\"": 11,
+  "(": 12,
+  ")": 13,
+  "“": 14,
+  "”": 15,
+  " ": 16,
+  "̃": 17,
+  "ʣ": 18,
+  "ʥ": 19,
+  "ʦ": 20,
+  "ʨ": 21,
+  "ᵝ": 22,
+  "ꭧ": 23,
+  "A": 24,
+  "I": 25,
+  "O": 31,
+  "Q": 33,
+  "S": 35,
+  "T": 36,
+  "W": 39,
+  "Y": 41,
+  "ᵊ": 42,
+  "a": 43,
+  "b": 44,
+  "c": 45,
+  "d": 46,
+  "e": 47,
+  "f": 48,
+  "h": 50,
+  "i": 51,
+  "j": 52,
+  "k": 53,
+  "l": 54,
+  "m": 55,
+  "n": 56,
+  "o": 57,
+  "p": 58,
+  "q": 59,
+  "r": 60,
+  "s": 61,
+  "t": 62,
+  "u": 63,
+  "v": 64,
+  "w": 65,
+  "x": 66,
+  "y": 67,
+  "z": 68,
+  "ɑ": 69,
+  "ɐ": 70,
+  "ɒ": 71,
+  "æ": 72,
+  "β": 75,
+  "ɔ": 76,
+  "ɕ": 77,
+  "ç": 78,
+  "ɖ": 80,
+  "ð": 81,
+  "ʤ": 82,
+  "ə": 83,
+  "ɚ": 85,
+  "ɛ": 86,
+  "ɜ": 87,
+  "ɟ": 90,
+  "ɡ": 92,
+  "ɥ": 99,
+  "ɨ": 101,
+  "ɪ": 102,
+  "ʝ": 103,
+  "ɯ": 110,
+  "ɰ": 111,
+  "ŋ": 112,
+  "ɳ": 113,
+  "ɲ": 114,
+  "ɴ": 115,
+  "ø": 116,
+  "ɸ": 118,
+  "θ": 119,
+  "œ": 120,
+  "ɹ": 123,
+  "ɾ": 125,
+  "ɻ": 126,
+  "ʁ": 128,
+  "ɽ": 129,
+  "ʂ": 130,
+  "ʃ": 131,
+  "ʈ": 132,
+  "ʧ": 133,
+  "ʊ": 135,
+  "ʋ": 136,
+  "ʌ": 138,
+  "ɣ": 139,
+  "ɤ": 140,
+  "χ": 142,
+  "ʎ": 143,
+  "ʒ": 147,
+  "ʔ": 148,
+  "ˈ": 156,
+  "ˌ": 157,
+  "ː": 158,
+  "ʰ": 162,
+  "ʲ": 164,
+  "↓": 169,
+  "→": 171,
+  "↗": 172,
+  "↘": 173,
+  "ᵻ": 177
+ }
+}

fp32/ParadeeAcoustic.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c954ed240ee6aea1844c5d7bda8405b0856aa9c54663a159f8ee7b4f9e0a6db
+size 243

fp32/ParadeeAcoustic.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:445a1c9b70d679b44b305a0518e288a26ac356f017d6c0a6228382ac025d72d1
+size 560

fp32/ParadeeAcoustic.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5ddbd3cde24b2026c85c6cb8e7bdb53c80bb398f6b63e26f7cd16df7546b282
+size 23005056

fp32/ParadeeText.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3d8b39b7e4577111ed93b5296205a5e53b600f7e78a07a52872c85349093be6
+size 243

fp32/ParadeeText.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6279264658a74a4985fc0954a1ff2663643045a07a0ada40be83863c8d0edc61
+size 503

fp32/ParadeeText.mlmodelc/model.mil

@@ -0,0 +1,525 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.2"}, {"coremltools-component-torch", "2.7.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
+{
+    func main<ios17>(tensor<int32, [1, ?]> input_ids) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"input_ids", [1, 45]}}), ("RangeDims", {{"input_ids", [[1, 1], [3, 512]]}})))] {
+            tensor<fp32, [256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_bias = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp32, [256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_weight = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_weight"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1152)))];
+            tensor<fp32, [256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_bias = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2240)))];
+            tensor<fp32, [256, 768]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_weight = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_weight"), val = tensor<fp32, [256, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3328)))];
+            tensor<fp32, [768]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_bias = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(789824)))];
+            tensor<fp32, [768, 256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_weight = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_weight"), val = tensor<fp32, [768, 256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792960)))];
+            tensor<fp32, [256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_bias = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1579456)))];
+            tensor<fp32, [256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_weight = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_weight"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1580544)))];
+            tensor<fp32, [256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_bias = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1581632)))];
+            tensor<fp32, [256, 256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_weight = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_weight"), val = tensor<fp32, [256, 256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582720)))];
+            tensor<fp32, [256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_bias = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1844928)))];
+            tensor<fp32, [256, 256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_weight = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_weight"), val = tensor<fp32, [256, 256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1846016)))];
+            tensor<fp32, [256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_bias = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2108224)))];
+            tensor<fp32, [256, 256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_weight = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_weight"), val = tensor<fp32, [256, 256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2109312)))];
+            tensor<fp32, [256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_bias = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2371520)))];
+            tensor<fp32, [256, 256]> ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_weight = const()[name = tensor<string, []>("ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_weight"), val = tensor<fp32, [256, 256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2372608)))];
+            tensor<fp32, [256]> ts_bert_encoder_embedding_hidden_mapping_in_bias = const()[name = tensor<string, []>("ts_bert_encoder_embedding_hidden_mapping_in_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2634816)))];
+            tensor<fp32, [256, 128]> ts_bert_encoder_embedding_hidden_mapping_in_weight = const()[name = tensor<string, []>("ts_bert_encoder_embedding_hidden_mapping_in_weight"), val = tensor<fp32, [256, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2635904)))];
+            tensor<fp32, [128]> ts_bert_embeddings_LayerNorm_bias = const()[name = tensor<string, []>("ts_bert_embeddings_LayerNorm_bias"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2767040)))];
+            tensor<fp32, [128]> ts_bert_embeddings_LayerNorm_weight = const()[name = tensor<string, []>("ts_bert_embeddings_LayerNorm_weight"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2767616)))];
+            tensor<fp32, [2, 128]> ts_bert_embeddings_token_type_embeddings_weight = const()[name = tensor<string, []>("ts_bert_embeddings_token_type_embeddings_weight"), val = tensor<fp32, [2, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2768192)))];
+            tensor<fp32, [512, 128]> ts_bert_embeddings_position_embeddings_weight = const()[name = tensor<string, []>("ts_bert_embeddings_position_embeddings_weight"), val = tensor<fp32, [512, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2769280)))];
+            tensor<fp32, [178, 128]> ts_bert_embeddings_word_embeddings_weight = const()[name = tensor<string, []>("ts_bert_embeddings_word_embeddings_weight"), val = tensor<fp32, [178, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3031488)))];
+            tensor<fp32, [192]> ts_bert_encoder_bias = const()[name = tensor<string, []>("ts_bert_encoder_bias"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3122688)))];
+            tensor<fp32, [192, 256]> ts_bert_encoder_weight = const()[name = tensor<string, []>("ts_bert_encoder_weight"), val = tensor<fp32, [192, 256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3123520)))];
+            tensor<fp32, [50]> ts_predictor_duration_proj_linear_layer_bias = const()[name = tensor<string, []>("ts_predictor_duration_proj_linear_layer_bias"), val = tensor<fp32, [50]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3320192)))];
+            tensor<fp32, [50, 192]> ts_predictor_duration_proj_linear_layer_weight = const()[name = tensor<string, []>("ts_predictor_duration_proj_linear_layer_weight"), val = tensor<fp32, [50, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3320512)))];
+            tensor<fp32, [178, 192]> ts_text_encoder_embedding_weight = const()[name = tensor<string, []>("ts_text_encoder_embedding_weight"), val = tensor<fp32, [178, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3358976)))];
+            tensor<fp32, [192]> ts_text_encoder_cnn_0_0_bias = const()[name = tensor<string, []>("ts_text_encoder_cnn_0_0_bias"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3495744)))];
+            tensor<fp32, [192, 192, 5]> ts_text_encoder_cnn_0_0_weight = const()[name = tensor<string, []>("ts_text_encoder_cnn_0_0_weight"), val = tensor<fp32, [192, 192, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3496576)))];
+            tensor<fp32, [192]> ts_text_encoder_cnn_0_1_beta = const()[name = tensor<string, []>("ts_text_encoder_cnn_0_1_beta"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4233920)))];
+            tensor<fp32, [192]> ts_text_encoder_cnn_0_1_gamma = const()[name = tensor<string, []>("ts_text_encoder_cnn_0_1_gamma"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4234752)))];
+            tensor<fp32, [192]> ts_text_encoder_cnn_1_0_bias = const()[name = tensor<string, []>("ts_text_encoder_cnn_1_0_bias"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4235584)))];
+            tensor<fp32, [192, 192, 5]> ts_text_encoder_cnn_1_0_weight = const()[name = tensor<string, []>("ts_text_encoder_cnn_1_0_weight"), val = tensor<fp32, [192, 192, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4236416)))];
+            tensor<fp32, [192]> ts_text_encoder_cnn_1_1_beta = const()[name = tensor<string, []>("ts_text_encoder_cnn_1_1_beta"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4973760)))];
+            tensor<fp32, [192]> ts_text_encoder_cnn_1_1_gamma = const()[name = tensor<string, []>("ts_text_encoder_cnn_1_1_gamma"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4974592)))];
+            tensor<fp32, [192]> ts_text_encoder_cnn_2_0_bias = const()[name = tensor<string, []>("ts_text_encoder_cnn_2_0_bias"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4975424)))];
+            tensor<fp32, [192, 192, 5]> ts_text_encoder_cnn_2_0_weight = const()[name = tensor<string, []>("ts_text_encoder_cnn_2_0_weight"), val = tensor<fp32, [192, 192, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4976256)))];
+            tensor<fp32, [192]> ts_text_encoder_cnn_2_1_beta = const()[name = tensor<string, []>("ts_text_encoder_cnn_2_1_beta"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5713600)))];
+            tensor<fp32, [192]> ts_text_encoder_cnn_2_1_gamma = const()[name = tensor<string, []>("ts_text_encoder_cnn_2_1_gamma"), val = tensor<fp32, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5714432)))];
+            tensor<fp32, [512]> ts_asr_proj_0_bias = const()[name = tensor<string, []>("ts_asr_proj_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5715264)))];
+            tensor<fp32, [512, 192]> ts_asr_proj_0_weight = const()[name = tensor<string, []>("ts_asr_proj_0_weight"), val = tensor<fp32, [512, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5717376)))];
+            tensor<fp32, [512]> ts_asr_proj_2_bias = const()[name = tensor<string, []>("ts_asr_proj_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6110656)))];
+            tensor<fp32, [512, 512]> ts_asr_proj_2_weight = const()[name = tensor<string, []>("ts_asr_proj_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6112768)))];
+            tensor<fp32, []> var_258 = const()[name = tensor<string, []>("op_258"), val = tensor<fp32, []>(0x1.197998p-40)];
+            tensor<int32, []> var_260 = const()[name = tensor<string, []>("op_260"), val = tensor<int32, []>(-1)];
+            tensor<int32, [2]> var_266_shape = shape(x = input_ids)[name = tensor<string, []>("op_266_shape")];
+            tensor<int32, []> gather_0_batch_dims_0 = const()[name = tensor<string, []>("gather_0_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> gather_0_validate_indices_0 = const()[name = tensor<string, []>("gather_0_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<int32, []> select_1 = const()[name = tensor<string, []>("select_1"), val = tensor<int32, []>(1)];
+            tensor<int32, []> gather_0_axis_1 = const()[name = tensor<string, []>("gather_0_axis_1"), val = tensor<int32, []>(0)];
+            tensor<int32, []> gather_0 = gather(axis = gather_0_axis_1, batch_dims = gather_0_batch_dims_0, indices = select_1, validate_indices = gather_0_validate_indices_0, x = var_266_shape)[name = tensor<string, []>("gather_0")];
+            tensor<int32, []> const_0 = const()[name = tensor<string, []>("const_0"), val = tensor<int32, []>(0)];
+            tensor<int32, []> const_1 = const()[name = tensor<string, []>("const_1"), val = tensor<int32, []>(1)];
+            tensor<int32, [?]> var_267 = range_1d(end = gather_0, start = const_0, step = const_1)[name = tensor<string, []>("op_267")];
+            tensor<int32, [1]> input_1_axes_0 = const()[name = tensor<string, []>("input_1_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1, ?]> input_1 = expand_dims(axes = input_1_axes_0, x = var_267)[name = tensor<string, []>("input_1")];
+            tensor<int32, []> var_269_batch_dims_0 = const()[name = tensor<string, []>("op_269_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> var_269_validate_indices_0 = const()[name = tensor<string, []>("op_269_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<int32, []> greater_equal_1_y_0 = const()[name = tensor<string, []>("greater_equal_1_y_0"), val = tensor<int32, []>(0)];
+            tensor<bool, [1, ?]> greater_equal_1 = greater_equal(x = input_ids, y = greater_equal_1_y_0)[name = tensor<string, []>("greater_equal_1")];
+            tensor<int32, []> slice_by_index_1 = const()[name = tensor<string, []>("slice_by_index_1"), val = tensor<int32, []>(178)];
+            tensor<int32, [1, ?]> add_11 = add(x = input_ids, y = slice_by_index_1)[name = tensor<string, []>("add_11")];
+            tensor<int32, [1, ?]> select_2 = select(a = input_ids, b = add_11, cond = greater_equal_1)[name = tensor<string, []>("select_2")];
+            tensor<int32, []> var_269_axis_1 = const()[name = tensor<string, []>("op_269_axis_1"), val = tensor<int32, []>(0)];
+            tensor<fp32, [1, ?, 128]> var_269 = gather(axis = var_269_axis_1, batch_dims = var_269_batch_dims_0, indices = select_2, validate_indices = var_269_validate_indices_0, x = ts_bert_embeddings_word_embeddings_weight)[name = tensor<string, []>("op_269")];
+            tensor<int32, []> var_270_batch_dims_0 = const()[name = tensor<string, []>("op_270_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> var_270_validate_indices_0 = const()[name = tensor<string, []>("op_270_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<int32, []> greater_equal_2_y_0 = const()[name = tensor<string, []>("greater_equal_2_y_0"), val = tensor<int32, []>(0)];
+            tensor<bool, [1, ?]> greater_equal_2 = greater_equal(x = input_1, y = greater_equal_2_y_0)[name = tensor<string, []>("greater_equal_2")];
+            tensor<int32, []> slice_by_index_2 = const()[name = tensor<string, []>("slice_by_index_2"), val = tensor<int32, []>(512)];
+            tensor<int32, [1, ?]> add_12 = add(x = input_1, y = slice_by_index_2)[name = tensor<string, []>("add_12")];
+            tensor<int32, [1, ?]> select_3 = select(a = input_1, b = add_12, cond = greater_equal_2)[name = tensor<string, []>("select_3")];
+            tensor<int32, []> var_270_axis_1 = const()[name = tensor<string, []>("op_270_axis_1"), val = tensor<int32, []>(0)];
+            tensor<fp32, [1, ?, 128]> var_270 = gather(axis = var_270_axis_1, batch_dims = var_270_batch_dims_0, indices = select_3, validate_indices = var_270_validate_indices_0, x = ts_bert_embeddings_position_embeddings_weight)[name = tensor<string, []>("op_270")];
+            tensor<fp32, [1, ?, 128]> var_271 = add(x = var_269, y = var_270)[name = tensor<string, []>("op_271")];
+            tensor<int32, [1, ?]> input_3 = sub(x = input_ids, y = input_ids)[name = tensor<string, []>("sub_0")];
+            tensor<int32, []> var_273_batch_dims_0 = const()[name = tensor<string, []>("op_273_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> var_273_validate_indices_0 = const()[name = tensor<string, []>("op_273_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<int32, []> greater_equal_3_y_0 = const()[name = tensor<string, []>("greater_equal_3_y_0"), val = tensor<int32, []>(0)];
+            tensor<bool, [1, ?]> greater_equal_3 = greater_equal(x = input_3, y = greater_equal_3_y_0)[name = tensor<string, []>("greater_equal_3")];
+            tensor<int32, []> slice_by_index_3 = const()[name = tensor<string, []>("slice_by_index_3"), val = tensor<int32, []>(2)];
+            tensor<int32, [1, ?]> add_13 = add(x = input_3, y = slice_by_index_3)[name = tensor<string, []>("add_13")];
+            tensor<int32, [1, ?]> select_4 = select(a = input_3, b = add_13, cond = greater_equal_3)[name = tensor<string, []>("select_4")];
+            tensor<int32, []> var_273_axis_1 = const()[name = tensor<string, []>("op_273_axis_1"), val = tensor<int32, []>(0)];
+            tensor<fp32, [1, ?, 128]> var_273 = gather(axis = var_273_axis_1, batch_dims = var_273_batch_dims_0, indices = select_4, validate_indices = var_273_validate_indices_0, x = ts_bert_embeddings_token_type_embeddings_weight)[name = tensor<string, []>("op_273")];
+            tensor<fp32, [1, ?, 128]> input_5 = add(x = var_271, y = var_273)[name = tensor<string, []>("input_5")];
+            tensor<int32, [1]> input_7_axes_0 = const()[name = tensor<string, []>("input_7_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 128]> input_7 = layer_norm(axes = input_7_axes_0, beta = ts_bert_embeddings_LayerNorm_bias, epsilon = var_258, gamma = ts_bert_embeddings_LayerNorm_weight, x = input_5)[name = tensor<string, []>("input_7")];
+            tensor<fp32, [1, ?, 256]> input_9 = linear(bias = ts_bert_encoder_embedding_hidden_mapping_in_bias, weight = ts_bert_encoder_embedding_hidden_mapping_in_weight, x = input_7)[name = tensor<string, []>("linear_0")];
+            tensor<fp32, [1, ?, 256]> var_278 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_weight, x = input_9)[name = tensor<string, []>("linear_1")];
+            tensor<int32, [4]> concat_0x = const()[name = tensor<string, []>("concat_0x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_280 = reshape(shape = concat_0x, x = var_278)[name = tensor<string, []>("op_280")];
+            tensor<fp32, [1, ?, 256]> var_282 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_weight, x = input_9)[name = tensor<string, []>("linear_2")];
+            tensor<int32, [4]> concat_1x = const()[name = tensor<string, []>("concat_1x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_284 = reshape(shape = concat_1x, x = var_282)[name = tensor<string, []>("op_284")];
+            tensor<fp32, [1, ?, 256]> var_286 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_weight, x = input_9)[name = tensor<string, []>("linear_3")];
+            tensor<int32, [4]> concat_2x = const()[name = tensor<string, []>("concat_2x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_288 = reshape(shape = concat_2x, x = var_286)[name = tensor<string, []>("op_288")];
+            tensor<int32, [4]> v_1_perm_0 = const()[name = tensor<string, []>("v_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> var_291_transpose_x_0 = const()[name = tensor<string, []>("op_291_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_291_transpose_y_0 = const()[name = tensor<string, []>("op_291_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_28_perm_0 = const()[name = tensor<string, []>("transpose_28_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_29_perm_0 = const()[name = tensor<string, []>("transpose_29_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp32, [1, 4, 64, ?]> transpose_29 = transpose(perm = transpose_29_perm_0, x = var_284)[name = tensor<string, []>("transpose_82")];
+            tensor<fp32, [1, 4, ?, 64]> transpose_28 = transpose(perm = transpose_28_perm_0, x = var_280)[name = tensor<string, []>("transpose_83")];
+            tensor<fp32, [1, 4, ?, ?]> var_291 = matmul(transpose_x = var_291_transpose_x_0, transpose_y = var_291_transpose_y_0, x = transpose_28, y = transpose_29)[name = tensor<string, []>("op_291")];
+            tensor<fp32, []> _inversed_input_11_y_0 = const()[name = tensor<string, []>("_inversed_input_11_y_0"), val = tensor<fp32, []>(0x1p-3)];
+            tensor<fp32, [1, 4, ?, ?]> _inversed_input_11 = mul(x = var_291, y = _inversed_input_11_y_0)[name = tensor<string, []>("_inversed_input_11")];
+            tensor<fp32, [1, 4, ?, ?]> att_1 = softmax(axis = var_260, x = _inversed_input_11)[name = tensor<string, []>("att_1")];
+            tensor<bool, []> var_295_transpose_x_0 = const()[name = tensor<string, []>("op_295_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_295_transpose_y_0 = const()[name = tensor<string, []>("op_295_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 4, ?, 64]> v_1 = transpose(perm = v_1_perm_0, x = var_288)[name = tensor<string, []>("transpose_84")];
+            tensor<fp32, [1, 4, ?, 64]> var_295 = matmul(transpose_x = var_295_transpose_x_0, transpose_y = var_295_transpose_y_0, x = att_1, y = v_1)[name = tensor<string, []>("op_295")];
+            tensor<int32, [4]> var_296_perm_0 = const()[name = tensor<string, []>("op_296_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_3x = const()[name = tensor<string, []>("concat_3x"), val = tensor<int32, [3]>([1, -1, 256])];
+            tensor<fp32, [1, ?, 4, 64]> var_296 = transpose(perm = var_296_perm_0, x = var_295)[name = tensor<string, []>("transpose_81")];
+            tensor<fp32, [1, ?, 256]> input_13 = reshape(shape = concat_3x, x = var_296)[name = tensor<string, []>("input_13")];
+            tensor<fp32, [1, ?, 256]> var_299 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_weight, x = input_13)[name = tensor<string, []>("linear_4")];
+            tensor<fp32, [1, ?, 256]> input_15 = add(x = var_299, y = input_9)[name = tensor<string, []>("input_15")];
+            tensor<int32, [1]> input_17_axes_0 = const()[name = tensor<string, []>("input_17_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_17 = layer_norm(axes = input_17_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_weight, x = input_15)[name = tensor<string, []>("input_17")];
+            tensor<fp32, [1, ?, 768]> var_303 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_weight, x = input_17)[name = tensor<string, []>("linear_5")];
+            tensor<string, []> input_19_mode_0 = const()[name = tensor<string, []>("input_19_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
+            tensor<fp32, [1, ?, 768]> input_19 = gelu(mode = input_19_mode_0, x = var_303)[name = tensor<string, []>("input_19")];
+            tensor<fp32, [1, ?, 256]> f_1 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_weight, x = input_19)[name = tensor<string, []>("linear_6")];
+            tensor<fp32, [1, ?, 256]> input_21 = add(x = f_1, y = input_17)[name = tensor<string, []>("input_21")];
+            tensor<int32, [1]> input_23_axes_0 = const()[name = tensor<string, []>("input_23_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_23 = layer_norm(axes = input_23_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_weight, x = input_21)[name = tensor<string, []>("input_23")];
+            tensor<fp32, [1, ?, 256]> var_309 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_weight, x = input_23)[name = tensor<string, []>("linear_7")];
+            tensor<int32, [4]> concat_4x = const()[name = tensor<string, []>("concat_4x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_311 = reshape(shape = concat_4x, x = var_309)[name = tensor<string, []>("op_311")];
+            tensor<fp32, [1, ?, 256]> var_313 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_weight, x = input_23)[name = tensor<string, []>("linear_8")];
+            tensor<int32, [4]> concat_5x = const()[name = tensor<string, []>("concat_5x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_315 = reshape(shape = concat_5x, x = var_313)[name = tensor<string, []>("op_315")];
+            tensor<fp32, [1, ?, 256]> var_317 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_weight, x = input_23)[name = tensor<string, []>("linear_9")];
+            tensor<int32, [4]> concat_6x = const()[name = tensor<string, []>("concat_6x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_319 = reshape(shape = concat_6x, x = var_317)[name = tensor<string, []>("op_319")];
+            tensor<int32, [4]> v_3_perm_0 = const()[name = tensor<string, []>("v_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> var_322_transpose_x_0 = const()[name = tensor<string, []>("op_322_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_322_transpose_y_0 = const()[name = tensor<string, []>("op_322_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_30_perm_0 = const()[name = tensor<string, []>("transpose_30_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_31_perm_0 = const()[name = tensor<string, []>("transpose_31_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp32, [1, 4, 64, ?]> transpose_31 = transpose(perm = transpose_31_perm_0, x = var_315)[name = tensor<string, []>("transpose_78")];
+            tensor<fp32, [1, 4, ?, 64]> transpose_30 = transpose(perm = transpose_30_perm_0, x = var_311)[name = tensor<string, []>("transpose_79")];
+            tensor<fp32, [1, 4, ?, ?]> var_322 = matmul(transpose_x = var_322_transpose_x_0, transpose_y = var_322_transpose_y_0, x = transpose_30, y = transpose_31)[name = tensor<string, []>("op_322")];
+            tensor<fp32, []> _inversed_input_25_y_0 = const()[name = tensor<string, []>("_inversed_input_25_y_0"), val = tensor<fp32, []>(0x1p-3)];
+            tensor<fp32, [1, 4, ?, ?]> _inversed_input_25 = mul(x = var_322, y = _inversed_input_25_y_0)[name = tensor<string, []>("_inversed_input_25")];
+            tensor<fp32, [1, 4, ?, ?]> att_3 = softmax(axis = var_260, x = _inversed_input_25)[name = tensor<string, []>("att_3")];
+            tensor<bool, []> var_326_transpose_x_0 = const()[name = tensor<string, []>("op_326_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_326_transpose_y_0 = const()[name = tensor<string, []>("op_326_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 4, ?, 64]> v_3 = transpose(perm = v_3_perm_0, x = var_319)[name = tensor<string, []>("transpose_80")];
+            tensor<fp32, [1, 4, ?, 64]> var_326 = matmul(transpose_x = var_326_transpose_x_0, transpose_y = var_326_transpose_y_0, x = att_3, y = v_3)[name = tensor<string, []>("op_326")];
+            tensor<int32, [4]> var_327_perm_0 = const()[name = tensor<string, []>("op_327_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_7x = const()[name = tensor<string, []>("concat_7x"), val = tensor<int32, [3]>([1, -1, 256])];
+            tensor<fp32, [1, ?, 4, 64]> var_327 = transpose(perm = var_327_perm_0, x = var_326)[name = tensor<string, []>("transpose_77")];
+            tensor<fp32, [1, ?, 256]> input_27 = reshape(shape = concat_7x, x = var_327)[name = tensor<string, []>("input_27")];
+            tensor<fp32, [1, ?, 256]> var_330 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_weight, x = input_27)[name = tensor<string, []>("linear_10")];
+            tensor<fp32, [1, ?, 256]> input_29 = add(x = var_330, y = input_23)[name = tensor<string, []>("input_29")];
+            tensor<int32, [1]> input_31_axes_0 = const()[name = tensor<string, []>("input_31_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_31 = layer_norm(axes = input_31_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_weight, x = input_29)[name = tensor<string, []>("input_31")];
+            tensor<fp32, [1, ?, 768]> var_334 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_weight, x = input_31)[name = tensor<string, []>("linear_11")];
+            tensor<string, []> input_33_mode_0 = const()[name = tensor<string, []>("input_33_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
+            tensor<fp32, [1, ?, 768]> input_33 = gelu(mode = input_33_mode_0, x = var_334)[name = tensor<string, []>("input_33")];
+            tensor<fp32, [1, ?, 256]> f_3 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_weight, x = input_33)[name = tensor<string, []>("linear_12")];
+            tensor<fp32, [1, ?, 256]> input_35 = add(x = f_3, y = input_31)[name = tensor<string, []>("input_35")];
+            tensor<int32, [1]> input_37_axes_0 = const()[name = tensor<string, []>("input_37_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_37 = layer_norm(axes = input_37_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_weight, x = input_35)[name = tensor<string, []>("input_37")];
+            tensor<fp32, [1, ?, 256]> var_340 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_weight, x = input_37)[name = tensor<string, []>("linear_13")];
+            tensor<int32, [4]> concat_8x = const()[name = tensor<string, []>("concat_8x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_342 = reshape(shape = concat_8x, x = var_340)[name = tensor<string, []>("op_342")];
+            tensor<fp32, [1, ?, 256]> var_344 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_weight, x = input_37)[name = tensor<string, []>("linear_14")];
+            tensor<int32, [4]> concat_9x = const()[name = tensor<string, []>("concat_9x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_346 = reshape(shape = concat_9x, x = var_344)[name = tensor<string, []>("op_346")];
+            tensor<fp32, [1, ?, 256]> var_348 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_weight, x = input_37)[name = tensor<string, []>("linear_15")];
+            tensor<int32, [4]> concat_10x = const()[name = tensor<string, []>("concat_10x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_350 = reshape(shape = concat_10x, x = var_348)[name = tensor<string, []>("op_350")];
+            tensor<int32, [4]> v_5_perm_0 = const()[name = tensor<string, []>("v_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> var_353_transpose_x_0 = const()[name = tensor<string, []>("op_353_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_353_transpose_y_0 = const()[name = tensor<string, []>("op_353_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_32_perm_0 = const()[name = tensor<string, []>("transpose_32_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_33_perm_0 = const()[name = tensor<string, []>("transpose_33_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp32, [1, 4, 64, ?]> transpose_33 = transpose(perm = transpose_33_perm_0, x = var_346)[name = tensor<string, []>("transpose_74")];
+            tensor<fp32, [1, 4, ?, 64]> transpose_32 = transpose(perm = transpose_32_perm_0, x = var_342)[name = tensor<string, []>("transpose_75")];
+            tensor<fp32, [1, 4, ?, ?]> var_353 = matmul(transpose_x = var_353_transpose_x_0, transpose_y = var_353_transpose_y_0, x = transpose_32, y = transpose_33)[name = tensor<string, []>("op_353")];
+            tensor<fp32, []> _inversed_input_39_y_0 = const()[name = tensor<string, []>("_inversed_input_39_y_0"), val = tensor<fp32, []>(0x1p-3)];
+            tensor<fp32, [1, 4, ?, ?]> _inversed_input_39 = mul(x = var_353, y = _inversed_input_39_y_0)[name = tensor<string, []>("_inversed_input_39")];
+            tensor<fp32, [1, 4, ?, ?]> att_5 = softmax(axis = var_260, x = _inversed_input_39)[name = tensor<string, []>("att_5")];
+            tensor<bool, []> var_357_transpose_x_0 = const()[name = tensor<string, []>("op_357_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_357_transpose_y_0 = const()[name = tensor<string, []>("op_357_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 4, ?, 64]> v_5 = transpose(perm = v_5_perm_0, x = var_350)[name = tensor<string, []>("transpose_76")];
+            tensor<fp32, [1, 4, ?, 64]> var_357 = matmul(transpose_x = var_357_transpose_x_0, transpose_y = var_357_transpose_y_0, x = att_5, y = v_5)[name = tensor<string, []>("op_357")];
+            tensor<int32, [4]> var_358_perm_0 = const()[name = tensor<string, []>("op_358_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_11x = const()[name = tensor<string, []>("concat_11x"), val = tensor<int32, [3]>([1, -1, 256])];
+            tensor<fp32, [1, ?, 4, 64]> var_358 = transpose(perm = var_358_perm_0, x = var_357)[name = tensor<string, []>("transpose_73")];
+            tensor<fp32, [1, ?, 256]> input_41 = reshape(shape = concat_11x, x = var_358)[name = tensor<string, []>("input_41")];
+            tensor<fp32, [1, ?, 256]> var_361 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_weight, x = input_41)[name = tensor<string, []>("linear_16")];
+            tensor<fp32, [1, ?, 256]> input_43 = add(x = var_361, y = input_37)[name = tensor<string, []>("input_43")];
+            tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_45 = layer_norm(axes = input_45_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_weight, x = input_43)[name = tensor<string, []>("input_45")];
+            tensor<fp32, [1, ?, 768]> var_365 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_weight, x = input_45)[name = tensor<string, []>("linear_17")];
+            tensor<string, []> input_47_mode_0 = const()[name = tensor<string, []>("input_47_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
+            tensor<fp32, [1, ?, 768]> input_47 = gelu(mode = input_47_mode_0, x = var_365)[name = tensor<string, []>("input_47")];
+            tensor<fp32, [1, ?, 256]> f_5 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_weight, x = input_47)[name = tensor<string, []>("linear_18")];
+            tensor<fp32, [1, ?, 256]> input_49 = add(x = f_5, y = input_45)[name = tensor<string, []>("input_49")];
+            tensor<int32, [1]> input_51_axes_0 = const()[name = tensor<string, []>("input_51_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_51 = layer_norm(axes = input_51_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_weight, x = input_49)[name = tensor<string, []>("input_51")];
+            tensor<fp32, [1, ?, 256]> var_371 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_weight, x = input_51)[name = tensor<string, []>("linear_19")];
+            tensor<int32, [4]> concat_12x = const()[name = tensor<string, []>("concat_12x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_373 = reshape(shape = concat_12x, x = var_371)[name = tensor<string, []>("op_373")];
+            tensor<fp32, [1, ?, 256]> var_375 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_weight, x = input_51)[name = tensor<string, []>("linear_20")];
+            tensor<int32, [4]> concat_13x = const()[name = tensor<string, []>("concat_13x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_377 = reshape(shape = concat_13x, x = var_375)[name = tensor<string, []>("op_377")];
+            tensor<fp32, [1, ?, 256]> var_379 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_weight, x = input_51)[name = tensor<string, []>("linear_21")];
+            tensor<int32, [4]> concat_14x = const()[name = tensor<string, []>("concat_14x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_381 = reshape(shape = concat_14x, x = var_379)[name = tensor<string, []>("op_381")];
+            tensor<int32, [4]> v_7_perm_0 = const()[name = tensor<string, []>("v_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> var_384_transpose_x_0 = const()[name = tensor<string, []>("op_384_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_384_transpose_y_0 = const()[name = tensor<string, []>("op_384_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_34_perm_0 = const()[name = tensor<string, []>("transpose_34_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_35_perm_0 = const()[name = tensor<string, []>("transpose_35_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp32, [1, 4, 64, ?]> transpose_35 = transpose(perm = transpose_35_perm_0, x = var_377)[name = tensor<string, []>("transpose_70")];
+            tensor<fp32, [1, 4, ?, 64]> transpose_34 = transpose(perm = transpose_34_perm_0, x = var_373)[name = tensor<string, []>("transpose_71")];
+            tensor<fp32, [1, 4, ?, ?]> var_384 = matmul(transpose_x = var_384_transpose_x_0, transpose_y = var_384_transpose_y_0, x = transpose_34, y = transpose_35)[name = tensor<string, []>("op_384")];
+            tensor<fp32, []> _inversed_input_53_y_0 = const()[name = tensor<string, []>("_inversed_input_53_y_0"), val = tensor<fp32, []>(0x1p-3)];
+            tensor<fp32, [1, 4, ?, ?]> _inversed_input_53 = mul(x = var_384, y = _inversed_input_53_y_0)[name = tensor<string, []>("_inversed_input_53")];
+            tensor<fp32, [1, 4, ?, ?]> att_7 = softmax(axis = var_260, x = _inversed_input_53)[name = tensor<string, []>("att_7")];
+            tensor<bool, []> var_388_transpose_x_0 = const()[name = tensor<string, []>("op_388_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_388_transpose_y_0 = const()[name = tensor<string, []>("op_388_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 4, ?, 64]> v_7 = transpose(perm = v_7_perm_0, x = var_381)[name = tensor<string, []>("transpose_72")];
+            tensor<fp32, [1, 4, ?, 64]> var_388 = matmul(transpose_x = var_388_transpose_x_0, transpose_y = var_388_transpose_y_0, x = att_7, y = v_7)[name = tensor<string, []>("op_388")];
+            tensor<int32, [4]> var_389_perm_0 = const()[name = tensor<string, []>("op_389_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_15x = const()[name = tensor<string, []>("concat_15x"), val = tensor<int32, [3]>([1, -1, 256])];
+            tensor<fp32, [1, ?, 4, 64]> var_389 = transpose(perm = var_389_perm_0, x = var_388)[name = tensor<string, []>("transpose_69")];
+            tensor<fp32, [1, ?, 256]> input_55 = reshape(shape = concat_15x, x = var_389)[name = tensor<string, []>("input_55")];
+            tensor<fp32, [1, ?, 256]> var_392 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_weight, x = input_55)[name = tensor<string, []>("linear_22")];
+            tensor<fp32, [1, ?, 256]> input_57 = add(x = var_392, y = input_51)[name = tensor<string, []>("input_57")];
+            tensor<int32, [1]> input_59_axes_0 = const()[name = tensor<string, []>("input_59_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_59 = layer_norm(axes = input_59_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_weight, x = input_57)[name = tensor<string, []>("input_59")];
+            tensor<fp32, [1, ?, 768]> var_396 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_weight, x = input_59)[name = tensor<string, []>("linear_23")];
+            tensor<string, []> input_61_mode_0 = const()[name = tensor<string, []>("input_61_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
+            tensor<fp32, [1, ?, 768]> input_61 = gelu(mode = input_61_mode_0, x = var_396)[name = tensor<string, []>("input_61")];
+            tensor<fp32, [1, ?, 256]> f_7 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_weight, x = input_61)[name = tensor<string, []>("linear_24")];
+            tensor<fp32, [1, ?, 256]> input_63 = add(x = f_7, y = input_59)[name = tensor<string, []>("input_63")];
+            tensor<int32, [1]> input_65_axes_0 = const()[name = tensor<string, []>("input_65_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_65 = layer_norm(axes = input_65_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_weight, x = input_63)[name = tensor<string, []>("input_65")];
+            tensor<fp32, [1, ?, 256]> var_402 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_weight, x = input_65)[name = tensor<string, []>("linear_25")];
+            tensor<int32, [4]> concat_16x = const()[name = tensor<string, []>("concat_16x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_404 = reshape(shape = concat_16x, x = var_402)[name = tensor<string, []>("op_404")];
+            tensor<fp32, [1, ?, 256]> var_406 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_weight, x = input_65)[name = tensor<string, []>("linear_26")];
+            tensor<int32, [4]> concat_17x = const()[name = tensor<string, []>("concat_17x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_408 = reshape(shape = concat_17x, x = var_406)[name = tensor<string, []>("op_408")];
+            tensor<fp32, [1, ?, 256]> var_410 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_weight, x = input_65)[name = tensor<string, []>("linear_27")];
+            tensor<int32, [4]> concat_18x = const()[name = tensor<string, []>("concat_18x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_412 = reshape(shape = concat_18x, x = var_410)[name = tensor<string, []>("op_412")];
+            tensor<int32, [4]> v_9_perm_0 = const()[name = tensor<string, []>("v_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> var_415_transpose_x_0 = const()[name = tensor<string, []>("op_415_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_415_transpose_y_0 = const()[name = tensor<string, []>("op_415_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_36_perm_0 = const()[name = tensor<string, []>("transpose_36_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_37_perm_0 = const()[name = tensor<string, []>("transpose_37_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp32, [1, 4, 64, ?]> transpose_37 = transpose(perm = transpose_37_perm_0, x = var_408)[name = tensor<string, []>("transpose_66")];
+            tensor<fp32, [1, 4, ?, 64]> transpose_36 = transpose(perm = transpose_36_perm_0, x = var_404)[name = tensor<string, []>("transpose_67")];
+            tensor<fp32, [1, 4, ?, ?]> var_415 = matmul(transpose_x = var_415_transpose_x_0, transpose_y = var_415_transpose_y_0, x = transpose_36, y = transpose_37)[name = tensor<string, []>("op_415")];
+            tensor<fp32, []> _inversed_input_67_y_0 = const()[name = tensor<string, []>("_inversed_input_67_y_0"), val = tensor<fp32, []>(0x1p-3)];
+            tensor<fp32, [1, 4, ?, ?]> _inversed_input_67 = mul(x = var_415, y = _inversed_input_67_y_0)[name = tensor<string, []>("_inversed_input_67")];
+            tensor<fp32, [1, 4, ?, ?]> att_9 = softmax(axis = var_260, x = _inversed_input_67)[name = tensor<string, []>("att_9")];
+            tensor<bool, []> var_419_transpose_x_0 = const()[name = tensor<string, []>("op_419_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_419_transpose_y_0 = const()[name = tensor<string, []>("op_419_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 4, ?, 64]> v_9 = transpose(perm = v_9_perm_0, x = var_412)[name = tensor<string, []>("transpose_68")];
+            tensor<fp32, [1, 4, ?, 64]> var_419 = matmul(transpose_x = var_419_transpose_x_0, transpose_y = var_419_transpose_y_0, x = att_9, y = v_9)[name = tensor<string, []>("op_419")];
+            tensor<int32, [4]> var_420_perm_0 = const()[name = tensor<string, []>("op_420_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_19x = const()[name = tensor<string, []>("concat_19x"), val = tensor<int32, [3]>([1, -1, 256])];
+            tensor<fp32, [1, ?, 4, 64]> var_420 = transpose(perm = var_420_perm_0, x = var_419)[name = tensor<string, []>("transpose_65")];
+            tensor<fp32, [1, ?, 256]> input_69 = reshape(shape = concat_19x, x = var_420)[name = tensor<string, []>("input_69")];
+            tensor<fp32, [1, ?, 256]> var_423 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_weight, x = input_69)[name = tensor<string, []>("linear_28")];
+            tensor<fp32, [1, ?, 256]> input_71 = add(x = var_423, y = input_65)[name = tensor<string, []>("input_71")];
+            tensor<int32, [1]> input_73_axes_0 = const()[name = tensor<string, []>("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_73 = layer_norm(axes = input_73_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_weight, x = input_71)[name = tensor<string, []>("input_73")];
+            tensor<fp32, [1, ?, 768]> var_427 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_weight, x = input_73)[name = tensor<string, []>("linear_29")];
+            tensor<string, []> input_75_mode_0 = const()[name = tensor<string, []>("input_75_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
+            tensor<fp32, [1, ?, 768]> input_75 = gelu(mode = input_75_mode_0, x = var_427)[name = tensor<string, []>("input_75")];
+            tensor<fp32, [1, ?, 256]> f_9 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_weight, x = input_75)[name = tensor<string, []>("linear_30")];
+            tensor<fp32, [1, ?, 256]> input_77 = add(x = f_9, y = input_73)[name = tensor<string, []>("input_77")];
+            tensor<int32, [1]> input_79_axes_0 = const()[name = tensor<string, []>("input_79_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_79 = layer_norm(axes = input_79_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_weight, x = input_77)[name = tensor<string, []>("input_79")];
+            tensor<fp32, [1, ?, 256]> var_433 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_query_weight, x = input_79)[name = tensor<string, []>("linear_31")];
+            tensor<int32, [4]> concat_20x = const()[name = tensor<string, []>("concat_20x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_435 = reshape(shape = concat_20x, x = var_433)[name = tensor<string, []>("op_435")];
+            tensor<fp32, [1, ?, 256]> var_437 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_key_weight, x = input_79)[name = tensor<string, []>("linear_32")];
+            tensor<int32, [4]> concat_21x = const()[name = tensor<string, []>("concat_21x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_439 = reshape(shape = concat_21x, x = var_437)[name = tensor<string, []>("op_439")];
+            tensor<fp32, [1, ?, 256]> var_441 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_value_weight, x = input_79)[name = tensor<string, []>("linear_33")];
+            tensor<int32, [4]> concat_22x = const()[name = tensor<string, []>("concat_22x"), val = tensor<int32, [4]>([1, -1, 4, 64])];
+            tensor<fp32, [1, ?, 4, 64]> var_443 = reshape(shape = concat_22x, x = var_441)[name = tensor<string, []>("op_443")];
+            tensor<int32, [4]> v_11_perm_0 = const()[name = tensor<string, []>("v_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> var_446_transpose_x_0 = const()[name = tensor<string, []>("op_446_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_446_transpose_y_0 = const()[name = tensor<string, []>("op_446_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_38_perm_0 = const()[name = tensor<string, []>("transpose_38_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_39_perm_0 = const()[name = tensor<string, []>("transpose_39_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp32, [1, 4, 64, ?]> transpose_39 = transpose(perm = transpose_39_perm_0, x = var_439)[name = tensor<string, []>("transpose_62")];
+            tensor<fp32, [1, 4, ?, 64]> transpose_38 = transpose(perm = transpose_38_perm_0, x = var_435)[name = tensor<string, []>("transpose_63")];
+            tensor<fp32, [1, 4, ?, ?]> var_446 = matmul(transpose_x = var_446_transpose_x_0, transpose_y = var_446_transpose_y_0, x = transpose_38, y = transpose_39)[name = tensor<string, []>("op_446")];
+            tensor<fp32, []> _inversed_input_81_y_0 = const()[name = tensor<string, []>("_inversed_input_81_y_0"), val = tensor<fp32, []>(0x1p-3)];
+            tensor<fp32, [1, 4, ?, ?]> _inversed_input_81 = mul(x = var_446, y = _inversed_input_81_y_0)[name = tensor<string, []>("_inversed_input_81")];
+            tensor<fp32, [1, 4, ?, ?]> att = softmax(axis = var_260, x = _inversed_input_81)[name = tensor<string, []>("att")];
+            tensor<bool, []> var_450_transpose_x_0 = const()[name = tensor<string, []>("op_450_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_450_transpose_y_0 = const()[name = tensor<string, []>("op_450_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 4, ?, 64]> v_11 = transpose(perm = v_11_perm_0, x = var_443)[name = tensor<string, []>("transpose_64")];
+            tensor<fp32, [1, 4, ?, 64]> var_450 = matmul(transpose_x = var_450_transpose_x_0, transpose_y = var_450_transpose_y_0, x = att, y = v_11)[name = tensor<string, []>("op_450")];
+            tensor<int32, [4]> var_451_perm_0 = const()[name = tensor<string, []>("op_451_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_23x = const()[name = tensor<string, []>("concat_23x"), val = tensor<int32, [3]>([1, -1, 256])];
+            tensor<fp32, [1, ?, 4, 64]> var_451 = transpose(perm = var_451_perm_0, x = var_450)[name = tensor<string, []>("transpose_61")];
+            tensor<fp32, [1, ?, 256]> input_83 = reshape(shape = concat_23x, x = var_451)[name = tensor<string, []>("input_83")];
+            tensor<fp32, [1, ?, 256]> var_454 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_dense_weight, x = input_83)[name = tensor<string, []>("linear_34")];
+            tensor<fp32, [1, ?, 256]> input_85 = add(x = var_454, y = input_79)[name = tensor<string, []>("input_85")];
+            tensor<int32, [1]> input_87_axes_0 = const()[name = tensor<string, []>("input_87_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_87 = layer_norm(axes = input_87_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_attention_LayerNorm_weight, x = input_85)[name = tensor<string, []>("input_87")];
+            tensor<fp32, [1, ?, 768]> var_458 = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_weight, x = input_87)[name = tensor<string, []>("linear_35")];
+            tensor<string, []> input_89_mode_0 = const()[name = tensor<string, []>("input_89_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
+            tensor<fp32, [1, ?, 768]> input_89 = gelu(mode = input_89_mode_0, x = var_458)[name = tensor<string, []>("input_89")];
+            tensor<fp32, [1, ?, 256]> f = linear(bias = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_bias, weight = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_ffn_output_weight, x = input_89)[name = tensor<string, []>("linear_36")];
+            tensor<fp32, [1, ?, 256]> input_91 = add(x = f, y = input_87)[name = tensor<string, []>("input_91")];
+            tensor<int32, [1]> input_93_axes_0 = const()[name = tensor<string, []>("input_93_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 256]> input_93 = layer_norm(axes = input_93_axes_0, beta = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_bias, epsilon = var_258, gamma = ts_bert_encoder_albert_layer_groups_0_albert_layers_0_full_layer_layer_norm_weight, x = input_91)[name = tensor<string, []>("input_93")];
+            tensor<fp32, [1, ?, 192]> var_466 = linear(bias = ts_bert_encoder_bias, weight = ts_bert_encoder_weight, x = input_93)[name = tensor<string, []>("linear_37")];
+            tensor<int32, [3]> x_1_perm_0 = const()[name = tensor<string, []>("x_1_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<fp32, [1, 32, 1]> var_485 = const()[name = tensor<string, []>("op_485"), val = tensor<fp32, [1, 32, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7161408)))];
+            tensor<int32, []> var_486 = const()[name = tensor<string, []>("op_486"), val = tensor<int32, []>(-1)];
+            tensor<int32, []> var_487 = const()[name = tensor<string, []>("op_487"), val = tensor<int32, []>(-1)];
+            tensor<int32, []> concat_24_axis_0 = const()[name = tensor<string, []>("concat_24_axis_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> concat_24_interleave_0 = const()[name = tensor<string, []>("concat_24_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [3]> concat_24 = concat(axis = concat_24_axis_0, interleave = concat_24_interleave_0, values = (var_486, var_487, gather_0))[name = tensor<string, []>("concat_24")];
+            tensor<int32, [3]> shape_1 = const()[name = tensor<string, []>("shape_1"), val = tensor<int32, [3]>([1, 32, 1])];
+            tensor<int32, []> equal_0_y_0 = const()[name = tensor<string, []>("equal_0_y_0"), val = tensor<int32, []>(-1)];
+            tensor<bool, [3]> equal_0 = equal(x = concat_24, y = equal_0_y_0)[name = tensor<string, []>("equal_0")];
+            tensor<int32, [3]> select_0 = select(a = shape_1, b = concat_24, cond = equal_0)[name = tensor<string, []>("select_0")];
+            tensor<int32, [3]> real_div_0 = real_div(x = select_0, y = shape_1)[name = tensor<string, []>("real_div_0")];
+            tensor<fp32, [?, ?, ?]> sx = tile(reps = real_div_0, x = var_485)[name = tensor<string, []>("sx")];
+            tensor<int32, []> var_492 = const()[name = tensor<string, []>("op_492"), val = tensor<int32, []>(1)];
+            tensor<bool, []> x_3_interleave_0 = const()[name = tensor<string, []>("x_3_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 192, ?]> x_1 = transpose(perm = x_1_perm_0, x = var_466)[name = tensor<string, []>("transpose_60")];
+            tensor<fp32, [1, ?, ?]> x_3 = concat(axis = var_492, interleave = x_3_interleave_0, values = (x_1, sx))[name = tensor<string, []>("x_3")];
+            tensor<int32, [3]> transpose_6_perm_0 = const()[name = tensor<string, []>("transpose_6_perm_0"), val = tensor<int32, [3]>([-1, 0, -2])];
+            tensor<fp32, [384]> add_0 = const()[name = tensor<string, []>("add_0"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7161600)))];
+            tensor<fp32, [384]> add_1 = const()[name = tensor<string, []>("add_1"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7163200)))];
+            tensor<fp32, [384, 224]> concat_29 = const()[name = tensor<string, []>("concat_29"), val = tensor<fp32, [384, 224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7164800)))];
+            tensor<fp32, [384, 96]> concat_30 = const()[name = tensor<string, []>("concat_30"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7508928)))];
+            tensor<fp32, [384, 224]> concat_31 = const()[name = tensor<string, []>("concat_31"), val = tensor<fp32, [384, 224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7656448)))];
+            tensor<fp32, [384, 96]> concat_32 = const()[name = tensor<string, []>("concat_32"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8000576)))];
+            tensor<fp32, [1, 192]> var_522_batch_first_lstm_h0_reshaped = const()[name = tensor<string, []>("op_522_batch_first_lstm_h0_reshaped"), val = tensor<fp32, [1, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8148096)))];
+            tensor<string, []> var_522_batch_first_direction_0 = const()[name = tensor<string, []>("op_522_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> var_522_batch_first_output_sequence_0 = const()[name = tensor<string, []>("op_522_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> var_522_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("op_522_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> var_522_batch_first_cell_activation_0 = const()[name = tensor<string, []>("op_522_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> var_522_batch_first_activation_0 = const()[name = tensor<string, []>("op_522_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp32, [?, 1, ?]> transpose_6 = transpose(perm = transpose_6_perm_0, x = x_3)[name = tensor<string, []>("transpose_59")];
+            tensor<fp32, [?, 1, 192]> var_522_batch_first_0, tensor<fp32, [1, 192]> var_522_batch_first_1, tensor<fp32, [1, 192]> var_522_batch_first_2 = lstm(activation = var_522_batch_first_activation_0, bias = add_0, bias_back = add_1, cell_activation = var_522_batch_first_cell_activation_0, direction = var_522_batch_first_direction_0, initial_c = var_522_batch_first_lstm_h0_reshaped, initial_h = var_522_batch_first_lstm_h0_reshaped, output_sequence = var_522_batch_first_output_sequence_0, recurrent_activation = var_522_batch_first_recurrent_activation_0, weight_hh = concat_30, weight_hh_back = concat_32, weight_ih = concat_29, weight_ih_back = concat_31, x = transpose_6)[name = tensor<string, []>("op_522_batch_first")];
+            tensor<int32, [3]> transpose_13_perm_0 = const()[name = tensor<string, []>("transpose_13_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp32, []> var_532 = const()[name = tensor<string, []>("op_532"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
+            tensor<fp32, [1, 1, 192]> beta_3 = const()[name = tensor<string, []>("beta_3"), val = tensor<fp32, [1, 1, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8148928)))];
+            tensor<int32, [1]> x_11_axes_0 = const()[name = tensor<string, []>("x_11_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 192]> transpose_13 = transpose(perm = transpose_13_perm_0, x = var_522_batch_first_0)[name = tensor<string, []>("transpose_58")];
+            tensor<fp32, [1, ?, 192]> x_11 = layer_norm(axes = x_11_axes_0, epsilon = var_532, x = transpose_13)[name = tensor<string, []>("x_11")];
+            tensor<fp32, [1, 1, 192]> var_556 = const()[name = tensor<string, []>("op_556"), val = tensor<fp32, [1, 1, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8149760)))];
+            tensor<fp32, [1, ?, 192]> var_557 = mul(x = var_556, y = x_11)[name = tensor<string, []>("op_557")];
+            tensor<fp32, [1, ?, 192]> x_13 = add(x = var_557, y = beta_3)[name = tensor<string, []>("x_13")];
+            tensor<int32, [3]> var_563_perm_0 = const()[name = tensor<string, []>("op_563_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<int32, []> var_565 = const()[name = tensor<string, []>("op_565"), val = tensor<int32, []>(1)];
+            tensor<bool, []> x_15_interleave_0 = const()[name = tensor<string, []>("x_15_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 192, ?]> var_563 = transpose(perm = var_563_perm_0, x = x_13)[name = tensor<string, []>("transpose_57")];
+            tensor<fp32, [1, ?, ?]> x_15 = concat(axis = var_565, interleave = x_15_interleave_0, values = (var_563, sx))[name = tensor<string, []>("x_15")];
+            tensor<int32, [3]> transpose_8_perm_0 = const()[name = tensor<string, []>("transpose_8_perm_0"), val = tensor<int32, [3]>([-1, 0, -2])];
+            tensor<fp32, [384]> add_2 = const()[name = tensor<string, []>("add_2"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8150592)))];
+            tensor<fp32, [384]> add_3 = const()[name = tensor<string, []>("add_3"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8152192)))];
+            tensor<fp32, [384, 224]> concat_39 = const()[name = tensor<string, []>("concat_39"), val = tensor<fp32, [384, 224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8153792)))];
+            tensor<fp32, [384, 96]> concat_40 = const()[name = tensor<string, []>("concat_40"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8497920)))];
+            tensor<fp32, [384, 224]> concat_41 = const()[name = tensor<string, []>("concat_41"), val = tensor<fp32, [384, 224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8645440)))];
+            tensor<fp32, [384, 96]> concat_42 = const()[name = tensor<string, []>("concat_42"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8989568)))];
+            tensor<string, []> var_595_batch_first_direction_0 = const()[name = tensor<string, []>("op_595_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> var_595_batch_first_output_sequence_0 = const()[name = tensor<string, []>("op_595_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> var_595_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("op_595_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> var_595_batch_first_cell_activation_0 = const()[name = tensor<string, []>("op_595_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> var_595_batch_first_activation_0 = const()[name = tensor<string, []>("op_595_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp32, [?, 1, ?]> transpose_8 = transpose(perm = transpose_8_perm_0, x = x_15)[name = tensor<string, []>("transpose_56")];
+            tensor<fp32, [?, 1, 192]> var_595_batch_first_0, tensor<fp32, [1, 192]> var_595_batch_first_1, tensor<fp32, [1, 192]> var_595_batch_first_2 = lstm(activation = var_595_batch_first_activation_0, bias = add_2, bias_back = add_3, cell_activation = var_595_batch_first_cell_activation_0, direction = var_595_batch_first_direction_0, initial_c = var_522_batch_first_lstm_h0_reshaped, initial_h = var_522_batch_first_lstm_h0_reshaped, output_sequence = var_595_batch_first_output_sequence_0, recurrent_activation = var_595_batch_first_recurrent_activation_0, weight_hh = concat_40, weight_hh_back = concat_42, weight_ih = concat_39, weight_ih_back = concat_41, x = transpose_8)[name = tensor<string, []>("op_595_batch_first")];
+            tensor<int32, [3]> transpose_14_perm_0 = const()[name = tensor<string, []>("transpose_14_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp32, []> var_605 = const()[name = tensor<string, []>("op_605"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
+            tensor<fp32, [1, 1, 192]> beta_7 = const()[name = tensor<string, []>("beta_7"), val = tensor<fp32, [1, 1, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9137088)))];
+            tensor<int32, [1]> x_23_axes_0 = const()[name = tensor<string, []>("x_23_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 192]> transpose_14 = transpose(perm = transpose_14_perm_0, x = var_595_batch_first_0)[name = tensor<string, []>("transpose_55")];
+            tensor<fp32, [1, ?, 192]> x_23 = layer_norm(axes = x_23_axes_0, epsilon = var_605, x = transpose_14)[name = tensor<string, []>("x_23")];
+            tensor<fp32, [1, 1, 192]> var_629 = const()[name = tensor<string, []>("op_629"), val = tensor<fp32, [1, 1, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9137920)))];
+            tensor<fp32, [1, ?, 192]> var_630 = mul(x = var_629, y = x_23)[name = tensor<string, []>("op_630")];
+            tensor<fp32, [1, ?, 192]> x_25 = add(x = var_630, y = beta_7)[name = tensor<string, []>("x_25")];
+            tensor<int32, [3]> var_636_perm_0 = const()[name = tensor<string, []>("op_636_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<int32, []> var_638 = const()[name = tensor<string, []>("op_638"), val = tensor<int32, []>(1)];
+            tensor<bool, []> x_27_interleave_0 = const()[name = tensor<string, []>("x_27_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 192, ?]> var_636 = transpose(perm = var_636_perm_0, x = x_25)[name = tensor<string, []>("transpose_54")];
+            tensor<fp32, [1, ?, ?]> x_27 = concat(axis = var_638, interleave = x_27_interleave_0, values = (var_636, sx))[name = tensor<string, []>("x_27")];
+            tensor<int32, [3]> transpose_10_perm_0 = const()[name = tensor<string, []>("transpose_10_perm_0"), val = tensor<int32, [3]>([-1, 0, -2])];
+            tensor<fp32, [384]> add_4 = const()[name = tensor<string, []>("add_4"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9138752)))];
+            tensor<fp32, [384]> add_5 = const()[name = tensor<string, []>("add_5"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9140352)))];
+            tensor<fp32, [384, 224]> concat_49 = const()[name = tensor<string, []>("concat_49"), val = tensor<fp32, [384, 224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9141952)))];
+            tensor<fp32, [384, 96]> concat_50 = const()[name = tensor<string, []>("concat_50"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9486080)))];
+            tensor<fp32, [384, 224]> concat_51 = const()[name = tensor<string, []>("concat_51"), val = tensor<fp32, [384, 224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9633600)))];
+            tensor<fp32, [384, 96]> concat_52 = const()[name = tensor<string, []>("concat_52"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9977728)))];
+            tensor<string, []> var_668_batch_first_direction_0 = const()[name = tensor<string, []>("op_668_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> var_668_batch_first_output_sequence_0 = const()[name = tensor<string, []>("op_668_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> var_668_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("op_668_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> var_668_batch_first_cell_activation_0 = const()[name = tensor<string, []>("op_668_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> var_668_batch_first_activation_0 = const()[name = tensor<string, []>("op_668_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp32, [?, 1, ?]> transpose_10 = transpose(perm = transpose_10_perm_0, x = x_27)[name = tensor<string, []>("transpose_53")];
+            tensor<fp32, [?, 1, 192]> var_668_batch_first_0, tensor<fp32, [1, 192]> var_668_batch_first_1, tensor<fp32, [1, 192]> var_668_batch_first_2 = lstm(activation = var_668_batch_first_activation_0, bias = add_4, bias_back = add_5, cell_activation = var_668_batch_first_cell_activation_0, direction = var_668_batch_first_direction_0, initial_c = var_522_batch_first_lstm_h0_reshaped, initial_h = var_522_batch_first_lstm_h0_reshaped, output_sequence = var_668_batch_first_output_sequence_0, recurrent_activation = var_668_batch_first_recurrent_activation_0, weight_hh = concat_50, weight_hh_back = concat_52, weight_ih = concat_49, weight_ih_back = concat_51, x = transpose_10)[name = tensor<string, []>("op_668_batch_first")];
+            tensor<int32, [3]> transpose_15_perm_0 = const()[name = tensor<string, []>("transpose_15_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp32, []> var_678 = const()[name = tensor<string, []>("op_678"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
+            tensor<fp32, [1, 1, 192]> beta_11 = const()[name = tensor<string, []>("beta_11"), val = tensor<fp32, [1, 1, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10125248)))];
+            tensor<int32, [1]> x_35_axes_0 = const()[name = tensor<string, []>("x_35_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 192]> transpose_15 = transpose(perm = transpose_15_perm_0, x = var_668_batch_first_0)[name = tensor<string, []>("transpose_52")];
+            tensor<fp32, [1, ?, 192]> x_35 = layer_norm(axes = x_35_axes_0, epsilon = var_678, x = transpose_15)[name = tensor<string, []>("x_35")];
+            tensor<fp32, [1, 1, 192]> var_702 = const()[name = tensor<string, []>("op_702"), val = tensor<fp32, [1, 1, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10126080)))];
+            tensor<fp32, [1, ?, 192]> var_703 = mul(x = var_702, y = x_35)[name = tensor<string, []>("op_703")];
+            tensor<fp32, [1, ?, 192]> x_37 = add(x = var_703, y = beta_11)[name = tensor<string, []>("x_37")];
+            tensor<int32, [3]> var_709_perm_0 = const()[name = tensor<string, []>("op_709_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<int32, []> var_711 = const()[name = tensor<string, []>("op_711"), val = tensor<int32, []>(1)];
+            tensor<bool, []> x_39_interleave_0 = const()[name = tensor<string, []>("x_39_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 192, ?]> var_709 = transpose(perm = var_709_perm_0, x = x_37)[name = tensor<string, []>("transpose_51")];
+            tensor<fp32, [1, ?, ?]> d = concat(axis = var_711, interleave = x_39_interleave_0, values = (var_709, sx))[name = tensor<string, []>("x_39")];
+            tensor<int32, [3]> transpose_12_perm_0 = const()[name = tensor<string, []>("transpose_12_perm_0"), val = tensor<int32, [3]>([-1, 0, -2])];
+            tensor<fp32, [384]> add_6 = const()[name = tensor<string, []>("add_6"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10126912)))];
+            tensor<fp32, [384]> add_7 = const()[name = tensor<string, []>("add_7"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10128512)))];
+            tensor<fp32, [384, 224]> concat_59 = const()[name = tensor<string, []>("concat_59"), val = tensor<fp32, [384, 224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10130112)))];
+            tensor<fp32, [384, 96]> concat_60 = const()[name = tensor<string, []>("concat_60"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10474240)))];
+            tensor<fp32, [384, 224]> concat_61 = const()[name = tensor<string, []>("concat_61"), val = tensor<fp32, [384, 224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10621760)))];
+            tensor<fp32, [384, 96]> concat_62 = const()[name = tensor<string, []>("concat_62"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10965888)))];
+            tensor<string, []> input_109_batch_first_direction_0 = const()[name = tensor<string, []>("input_109_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> input_109_batch_first_output_sequence_0 = const()[name = tensor<string, []>("input_109_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_109_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("input_109_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_109_batch_first_cell_activation_0 = const()[name = tensor<string, []>("input_109_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_109_batch_first_activation_0 = const()[name = tensor<string, []>("input_109_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp32, [?, 1, ?]> transpose_12 = transpose(perm = transpose_12_perm_0, x = d)[name = tensor<string, []>("transpose_50")];
+            tensor<fp32, [?, 1, 192]> input_109_batch_first_0, tensor<fp32, [1, 192]> input_109_batch_first_1, tensor<fp32, [1, 192]> input_109_batch_first_2 = lstm(activation = input_109_batch_first_activation_0, bias = add_6, bias_back = add_7, cell_activation = input_109_batch_first_cell_activation_0, direction = input_109_batch_first_direction_0, initial_c = var_522_batch_first_lstm_h0_reshaped, initial_h = var_522_batch_first_lstm_h0_reshaped, output_sequence = input_109_batch_first_output_sequence_0, recurrent_activation = input_109_batch_first_recurrent_activation_0, weight_hh = concat_60, weight_hh_back = concat_62, weight_ih = concat_59, weight_ih_back = concat_61, x = transpose_12)[name = tensor<string, []>("input_109_batch_first")];
+            tensor<int32, [3]> input_109_perm_0 = const()[name = tensor<string, []>("input_109_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp32, [1, ?, 192]> input_109 = transpose(perm = input_109_perm_0, x = input_109_batch_first_0)[name = tensor<string, []>("transpose_49")];
+            tensor<fp32, [1, ?, 50]> var_747 = linear(bias = ts_predictor_duration_proj_linear_layer_bias, weight = ts_predictor_duration_proj_linear_layer_weight, x = input_109)[name = tensor<string, []>("linear_41")];
+            tensor<fp32, [1, ?, 50]> var_748 = sigmoid(x = var_747)[name = tensor<string, []>("op_748")];
+            tensor<int32, [1]> var_753_axes_0 = const()[name = tensor<string, []>("op_753_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<bool, []> var_753_keep_dims_0 = const()[name = tensor<string, []>("op_753_keep_dims_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, ?]> duration = reduce_sum(axes = var_753_axes_0, keep_dims = var_753_keep_dims_0, x = var_748)[name = tensor<string, []>("op_753")];
+            tensor<int32, []> var_757_batch_dims_0 = const()[name = tensor<string, []>("op_757_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> var_757_validate_indices_0 = const()[name = tensor<string, []>("op_757_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<int32, []> var_757_axis_1 = const()[name = tensor<string, []>("op_757_axis_1"), val = tensor<int32, []>(0)];
+            tensor<fp32, [1, ?, 192]> var_757 = gather(axis = var_757_axis_1, batch_dims = var_757_batch_dims_0, indices = select_2, validate_indices = var_757_validate_indices_0, x = ts_text_encoder_embedding_weight)[name = tensor<string, []>("op_757")];
+            tensor<int32, [3]> input_111_perm_0 = const()[name = tensor<string, []>("input_111_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp32, []> var_761 = const()[name = tensor<string, []>("op_761"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<fp32, []> var_764 = const()[name = tensor<string, []>("op_764"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
+            tensor<string, []> x_41_pad_type_0 = const()[name = tensor<string, []>("x_41_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_41_pad_0 = const()[name = tensor<string, []>("x_41_pad_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [1]> x_41_strides_0 = const()[name = tensor<string, []>("x_41_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> x_41_dilations_0 = const()[name = tensor<string, []>("x_41_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> x_41_groups_0 = const()[name = tensor<string, []>("x_41_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 192, ?]> input_111 = transpose(perm = input_111_perm_0, x = var_757)[name = tensor<string, []>("transpose_48")];
+            tensor<fp32, [1, 192, ?]> x_41 = conv(bias = ts_text_encoder_cnn_0_0_bias, dilations = x_41_dilations_0, groups = x_41_groups_0, pad = x_41_pad_0, pad_type = x_41_pad_type_0, strides = x_41_strides_0, weight = ts_text_encoder_cnn_0_0_weight, x = input_111)[name = tensor<string, []>("x_41")];
+            tensor<int32, [3]> input_113_perm_0 = const()[name = tensor<string, []>("input_113_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [1]> x_43_axes_0 = const()[name = tensor<string, []>("x_43_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 192]> input_113 = transpose(perm = input_113_perm_0, x = x_41)[name = tensor<string, []>("transpose_47")];
+            tensor<fp32, [1, ?, 192]> x_43 = layer_norm(axes = x_43_axes_0, beta = ts_text_encoder_cnn_0_1_beta, epsilon = var_764, gamma = ts_text_encoder_cnn_0_1_gamma, x = input_113)[name = tensor<string, []>("x_43")];
+            tensor<int32, [3]> input_115_perm_0 = const()[name = tensor<string, []>("input_115_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<fp32, [1, 192, ?]> input_115 = transpose(perm = input_115_perm_0, x = x_43)[name = tensor<string, []>("transpose_46")];
+            tensor<fp32, [1, 192, ?]> input_117 = leaky_relu(alpha = var_761, x = input_115)[name = tensor<string, []>("input_117")];
+            tensor<fp32, []> var_787 = const()[name = tensor<string, []>("op_787"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<fp32, []> var_790 = const()[name = tensor<string, []>("op_790"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
+            tensor<string, []> x_45_pad_type_0 = const()[name = tensor<string, []>("x_45_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_45_pad_0 = const()[name = tensor<string, []>("x_45_pad_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [1]> x_45_strides_0 = const()[name = tensor<string, []>("x_45_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> x_45_dilations_0 = const()[name = tensor<string, []>("x_45_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> x_45_groups_0 = const()[name = tensor<string, []>("x_45_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 192, ?]> x_45 = conv(bias = ts_text_encoder_cnn_1_0_bias, dilations = x_45_dilations_0, groups = x_45_groups_0, pad = x_45_pad_0, pad_type = x_45_pad_type_0, strides = x_45_strides_0, weight = ts_text_encoder_cnn_1_0_weight, x = input_117)[name = tensor<string, []>("x_45")];
+            tensor<int32, [3]> input_121_perm_0 = const()[name = tensor<string, []>("input_121_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [1]> x_47_axes_0 = const()[name = tensor<string, []>("x_47_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 192]> input_121 = transpose(perm = input_121_perm_0, x = x_45)[name = tensor<string, []>("transpose_45")];
+            tensor<fp32, [1, ?, 192]> x_47 = layer_norm(axes = x_47_axes_0, beta = ts_text_encoder_cnn_1_1_beta, epsilon = var_790, gamma = ts_text_encoder_cnn_1_1_gamma, x = input_121)[name = tensor<string, []>("x_47")];
+            tensor<int32, [3]> input_123_perm_0 = const()[name = tensor<string, []>("input_123_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<fp32, [1, 192, ?]> input_123 = transpose(perm = input_123_perm_0, x = x_47)[name = tensor<string, []>("transpose_44")];
+            tensor<fp32, [1, 192, ?]> input_125 = leaky_relu(alpha = var_787, x = input_123)[name = tensor<string, []>("input_125")];
+            tensor<fp32, []> var_813 = const()[name = tensor<string, []>("op_813"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<fp32, []> var_816 = const()[name = tensor<string, []>("op_816"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
+            tensor<string, []> x_49_pad_type_0 = const()[name = tensor<string, []>("x_49_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_49_pad_0 = const()[name = tensor<string, []>("x_49_pad_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [1]> x_49_strides_0 = const()[name = tensor<string, []>("x_49_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> x_49_dilations_0 = const()[name = tensor<string, []>("x_49_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> x_49_groups_0 = const()[name = tensor<string, []>("x_49_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 192, ?]> x_49 = conv(bias = ts_text_encoder_cnn_2_0_bias, dilations = x_49_dilations_0, groups = x_49_groups_0, pad = x_49_pad_0, pad_type = x_49_pad_type_0, strides = x_49_strides_0, weight = ts_text_encoder_cnn_2_0_weight, x = input_125)[name = tensor<string, []>("x_49")];
+            tensor<int32, [3]> input_129_perm_0 = const()[name = tensor<string, []>("input_129_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [1]> x_axes_0 = const()[name = tensor<string, []>("x_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp32, [1, ?, 192]> input_129 = transpose(perm = input_129_perm_0, x = x_49)[name = tensor<string, []>("transpose_43")];
+            tensor<fp32, [1, ?, 192]> x = layer_norm(axes = x_axes_0, beta = ts_text_encoder_cnn_2_1_beta, epsilon = var_816, gamma = ts_text_encoder_cnn_2_1_gamma, x = input_129)[name = tensor<string, []>("x")];
+            tensor<fp32, [1, ?, 192]> input_133 = leaky_relu(alpha = var_813, x = x)[name = tensor<string, []>("input_133")];
+            tensor<int32, [3]> input_135_batch_first_transpose_perm_0 = const()[name = tensor<string, []>("input_135_batch_first_transpose_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp32, [384]> add_8 = const()[name = tensor<string, []>("add_8"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11113408)))];
+            tensor<fp32, [384]> add_9 = const()[name = tensor<string, []>("add_9"), val = tensor<fp32, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11115008)))];
+            tensor<fp32, [384, 192]> concat_69 = const()[name = tensor<string, []>("concat_69"), val = tensor<fp32, [384, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11116608)))];
+            tensor<fp32, [384, 96]> concat_70 = const()[name = tensor<string, []>("concat_70"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11411584)))];
+            tensor<fp32, [384, 192]> concat_71 = const()[name = tensor<string, []>("concat_71"), val = tensor<fp32, [384, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11559104)))];
+            tensor<fp32, [384, 96]> concat_72 = const()[name = tensor<string, []>("concat_72"), val = tensor<fp32, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11854080)))];
+            tensor<string, []> input_137_batch_first_direction_0 = const()[name = tensor<string, []>("input_137_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> input_137_batch_first_output_sequence_0 = const()[name = tensor<string, []>("input_137_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_137_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("input_137_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_137_batch_first_cell_activation_0 = const()[name = tensor<string, []>("input_137_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_137_batch_first_activation_0 = const()[name = tensor<string, []>("input_137_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp32, [?, 1, 192]> input_135_batch_first_transpose = transpose(perm = input_135_batch_first_transpose_perm_0, x = input_133)[name = tensor<string, []>("transpose_42")];
+            tensor<fp32, [?, 1, 192]> input_137_batch_first_0, tensor<fp32, [1, 192]> input_137_batch_first_1, tensor<fp32, [1, 192]> input_137_batch_first_2 = lstm(activation = input_137_batch_first_activation_0, bias = add_8, bias_back = add_9, cell_activation = input_137_batch_first_cell_activation_0, direction = input_137_batch_first_direction_0, initial_c = var_522_batch_first_lstm_h0_reshaped, initial_h = var_522_batch_first_lstm_h0_reshaped, output_sequence = input_137_batch_first_output_sequence_0, recurrent_activation = input_137_batch_first_recurrent_activation_0, weight_hh = concat_70, weight_hh_back = concat_72, weight_ih = concat_69, weight_ih_back = concat_71, x = input_135_batch_first_transpose)[name = tensor<string, []>("input_137_batch_first")];
+            tensor<int32, [3]> input_137_perm_0 = const()[name = tensor<string, []>("input_137_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp32, [1, ?, 192]> input_137 = transpose(perm = input_137_perm_0, x = input_137_batch_first_0)[name = tensor<string, []>("transpose_41")];
+            tensor<fp32, [1, ?, 512]> input_139 = linear(bias = ts_asr_proj_0_bias, weight = ts_asr_proj_0_weight, x = input_137)[name = tensor<string, []>("linear_42")];
+            tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp32, [1, ?, 512]> input = gelu(mode = input_mode_0, x = input_139)[name = tensor<string, []>("input")];
+            tensor<fp32, [1, ?, 512]> var_879 = linear(bias = ts_asr_proj_2_bias, weight = ts_asr_proj_2_weight, x = input)[name = tensor<string, []>("linear_43")];
+            tensor<int32, [3]> var_882_perm_0 = const()[name = tensor<string, []>("op_882_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<fp32, [1, 512, ?]> asr_tok = transpose(perm = var_882_perm_0, x = var_879)[name = tensor<string, []>("transpose_40")];
+        } -> (duration, d, asr_tok);
+}

fp32/ParadeeText.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4eafab039e8eea472da30fdd4eddf7d0ebfeecfc1af29c83fead9573a2d46ae1
+size 12001600

fp32/vocab.json

@@ -0,0 +1,116 @@
+{
+ ";": 1,
+ ":": 2,
+ ",": 3,
+ ".": 4,
+ "!": 5,
+ "?": 6,
+ "—": 9,
+ "…": 10,
+ "\"": 11,
+ "(": 12,
+ ")": 13,
+ "“": 14,
+ "”": 15,
+ " ": 16,
+ "̃": 17,
+ "ʣ": 18,
+ "ʥ": 19,
+ "ʦ": 20,
+ "ʨ": 21,
+ "ᵝ": 22,
+ "ꭧ": 23,
+ "A": 24,
+ "I": 25,
+ "O": 31,
+ "Q": 33,
+ "S": 35,
+ "T": 36,
+ "W": 39,
+ "Y": 41,
+ "ᵊ": 42,
+ "a": 43,
+ "b": 44,
+ "c": 45,
+ "d": 46,
+ "e": 47,
+ "f": 48,
+ "h": 50,
+ "i": 51,
+ "j": 52,
+ "k": 53,
+ "l": 54,
+ "m": 55,
+ "n": 56,
+ "o": 57,
+ "p": 58,
+ "q": 59,
+ "r": 60,
+ "s": 61,
+ "t": 62,
+ "u": 63,
+ "v": 64,
+ "w": 65,
+ "x": 66,
+ "y": 67,
+ "z": 68,
+ "ɑ": 69,
+ "ɐ": 70,
+ "ɒ": 71,
+ "æ": 72,
+ "β": 75,
+ "ɔ": 76,
+ "ɕ": 77,
+ "ç": 78,
+ "ɖ": 80,
+ "ð": 81,
+ "ʤ": 82,
+ "ə": 83,
+ "ɚ": 85,
+ "ɛ": 86,
+ "ɜ": 87,
+ "ɟ": 90,
+ "ɡ": 92,
+ "ɥ": 99,
+ "ɨ": 101,
+ "ɪ": 102,
+ "ʝ": 103,
+ "ɯ": 110,
+ "ɰ": 111,
+ "ŋ": 112,
+ "ɳ": 113,
+ "ɲ": 114,
+ "ɴ": 115,
+ "ø": 116,
+ "ɸ": 118,
+ "θ": 119,
+ "œ": 120,
+ "ɹ": 123,
+ "ɾ": 125,
+ "ɻ": 126,
+ "ʁ": 128,
+ "ɽ": 129,
+ "ʂ": 130,
+ "ʃ": 131,
+ "ʈ": 132,
+ "ʧ": 133,
+ "ʊ": 135,
+ "ʋ": 136,
+ "ʌ": 138,
+ "ɣ": 139,
+ "ɤ": 140,
+ "χ": 142,
+ "ʎ": 143,
+ "ʒ": 147,
+ "ʔ": 148,
+ "ˈ": 156,
+ "ˌ": 157,
+ "ː": 158,
+ "ʰ": 162,
+ "ʲ": 164,
+ "↓": 169,
+ "→": 171,
+ "↗": 172,
+ "↘": 173,
+ "ᵻ": 177
+}

int8/ParadeeAcoustic.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75b72dd0cf2b79be6a26670bdb0ea606f1c18129dbccc8a399b46f3b5f25ad36
+size 243

int8/ParadeeAcoustic.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f721bdbe481f144947093a7839c6fdda91f30b05d7841873d129e1da9b0cd503
+size 560

int8/ParadeeAcoustic.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b12e686efd35e6922148bbd131011391d078637fc38d4164b90a7007979004d
+size 9122048

int8/ParadeeText.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49b416543979b7b6c66acbd6109519ba911de9aa601bc04182e00f29b628dccf
+size 243

int8/ParadeeText.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f31daa2542a0d519c9206e7973be437ad19810a4ffc417dc5d09fd58e523ce00
+size 503

int8/ParadeeText.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48ecd9d62f60cb016c4542c4e15be47ab14059d433ad6ca456f4160491bcd2fc
+size 3096704

int8/vocab.json

@@ -0,0 +1,116 @@
+{
+ ";": 1,
+ ":": 2,
+ ",": 3,
+ ".": 4,
+ "!": 5,
+ "?": 6,
+ "—": 9,
+ "…": 10,
+ "\"": 11,
+ "(": 12,
+ ")": 13,
+ "“": 14,
+ "”": 15,
+ " ": 16,
+ "̃": 17,
+ "ʣ": 18,
+ "ʥ": 19,
+ "ʦ": 20,
+ "ʨ": 21,
+ "ᵝ": 22,
+ "ꭧ": 23,
+ "A": 24,
+ "I": 25,
+ "O": 31,
+ "Q": 33,
+ "S": 35,
+ "T": 36,
+ "W": 39,
+ "Y": 41,
+ "ᵊ": 42,
+ "a": 43,
+ "b": 44,
+ "c": 45,
+ "d": 46,
+ "e": 47,
+ "f": 48,
+ "h": 50,
+ "i": 51,
+ "j": 52,
+ "k": 53,
+ "l": 54,
+ "m": 55,
+ "n": 56,
+ "o": 57,
+ "p": 58,
+ "q": 59,
+ "r": 60,
+ "s": 61,
+ "t": 62,
+ "u": 63,
+ "v": 64,
+ "w": 65,
+ "x": 66,
+ "y": 67,
+ "z": 68,
+ "ɑ": 69,
+ "ɐ": 70,
+ "ɒ": 71,
+ "æ": 72,
+ "β": 75,
+ "ɔ": 76,
+ "ɕ": 77,
+ "ç": 78,
+ "ɖ": 80,
+ "ð": 81,
+ "ʤ": 82,
+ "ə": 83,
+ "ɚ": 85,
+ "ɛ": 86,
+ "ɜ": 87,
+ "ɟ": 90,
+ "ɡ": 92,
+ "ɥ": 99,
+ "ɨ": 101,
+ "ɪ": 102,
+ "ʝ": 103,
+ "ɯ": 110,
+ "ɰ": 111,
+ "ŋ": 112,
+ "ɳ": 113,
+ "ɲ": 114,
+ "ɴ": 115,
+ "ø": 116,
+ "ɸ": 118,
+ "θ": 119,
+ "œ": 120,
+ "ɹ": 123,
+ "ɾ": 125,
+ "ɻ": 126,
+ "ʁ": 128,
+ "ɽ": 129,
+ "ʂ": 130,
+ "ʃ": 131,
+ "ʈ": 132,
+ "ʧ": 133,
+ "ʊ": 135,
+ "ʋ": 136,
+ "ʌ": 138,
+ "ɣ": 139,
+ "ɤ": 140,
+ "χ": 142,
+ "ʎ": 143,
+ "ʒ": 147,
+ "ʔ": 148,
+ "ˈ": 156,
+ "ˌ": 157,
+ "ː": 158,
+ "ʰ": 162,
+ "ʲ": 164,
+ "↓": 169,
+ "→": 171,
+ "↗": 172,
+ "↘": 173,
+ "ᵻ": 177
+}

mlpackage/fp32/ParadeeAcoustic.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b11577fdc8f8d5e23d1386ed947f131bb5e0d536c108e9691969c6809ea09577
+size 343044

mlpackage/fp32/ParadeeAcoustic.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5ddbd3cde24b2026c85c6cb8e7bdb53c80bb398f6b63e26f7cd16df7546b282
+size 23005056

mlpackage/fp32/ParadeeAcoustic.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "9E3499C1-6926-4AF4-B230-EDF4673FA785": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "EE6137E9-559B-48B1-B378-736A4BA743E7": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "EE6137E9-559B-48B1-B378-736A4BA743E7"
+}

mlpackage/fp32/ParadeeText.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5300414d775ba79242bcb4816ffc203776fdb341530bdcfa1fdbc81224d1bf54
+size 83201

mlpackage/fp32/ParadeeText.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4eafab039e8eea472da30fdd4eddf7d0ebfeecfc1af29c83fead9573a2d46ae1
+size 12001600

mlpackage/fp32/ParadeeText.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "2CD980B8-BCEB-429D-91EA-8BE0F732E685": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "73796F30-C6AF-4EB5-9EB0-19CEAD8BDD12": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        }
+    },
+    "rootModelIdentifier": "2CD980B8-BCEB-429D-91EA-8BE0F732E685"
+}

mlpackage/int8/ParadeeAcoustic.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f23123d0e324485c33f3a017b458f98a0aaeaf018ecaab73e2bc97cf3f83bc4
+size 363038

mlpackage/int8/ParadeeAcoustic.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b12e686efd35e6922148bbd131011391d078637fc38d4164b90a7007979004d
+size 9122048

mlpackage/int8/ParadeeAcoustic.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "684B48A1-4761-4252-A8BF-516F26588EFB": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "A94067AB-C927-4411-9F77-429B6A814B97": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "A94067AB-C927-4411-9F77-429B6A814B97"
+}

mlpackage/int8/ParadeeText.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc7817a3de669891238f4a83351432468b4032e2aa6fc34f385e81f59838714d
+size 93438

mlpackage/int8/ParadeeText.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48ecd9d62f60cb016c4542c4e15be47ab14059d433ad6ca456f4160491bcd2fc
+size 3096704

mlpackage/int8/ParadeeText.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "5D71F5CC-C1CE-485E-B612-E718B3192E91": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "64977AF2-E38D-4F1D-86DF-7EB0140C2A77": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        }
+    },
+    "rootModelIdentifier": "5D71F5CC-C1CE-485E-B612-E718B3192E91"
+}

samples/1_paradee_coreml_int8.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:173b5a9f6ec4bc25000744a7e2b22086f21dd32a98173826e1c6e4020b5b577f
+size 466844

samples/2_paradee_coreml_int8.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42c901f1cd5a50426108b4499d854691691d20c604d579ecd5415d4251177ed7
+size 283244

samples/3_paradee_coreml_int8.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc256d31aa704391dfe6cb43baaae489bddd73de2f6eb368008d5f09f1b4df42
+size 361244

samples/4_paradee_coreml_int8.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54aa0d2969d54561759607a5ac33b00d389e34ea43586a553360170dcad8625f
+size 460844

samples/5_paradee_coreml_int8.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1144add8fe6051063db88d48e93e3ddba9c1a96ef6dc55240399fb4d9f7abc9
+size 516044

samples/sentences.txt

@@ -0,0 +1,5 @@
+1. As of August 2015, there were 169 proposed targets for these goals and 304 proposed indicators to show compliance.
+2. White dwarf stars, if they have a near companion, may then become Type Ia supernovae.
+3. According to Herodotus, this was because they had decided in council that they could not beat the Allies in a naval battle.
+4. In 1997 Hurlbut's career focused on light as applied to photography and film, and he owned a lighting business in Pasadena, California.
+5. Harvey ended with a century on his Ashes debut, scoring 112 from 183 balls in an innings noted for powerful driving on both sides of the wicket.