clean reupload
Browse files- LICENSE +201 -0
- README.md +141 -5
- audio_vae_encoder.pnnx.bin +0 -3
- audio_vae_encoder.pnnx.param +0 -149
- audio_vae_encoder.pt +0 -3
- audio_vae_encoder_ncnn.py +0 -26
- audio_vae_encoder_pnnx.py +0 -378
- tokenization_voxcpm2.py +0 -72
LICENSE
ADDED
|
@@ -0,0 +1,201 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Apache License
|
| 2 |
+
Version 2.0, January 2004
|
| 3 |
+
http://www.apache.org/licenses/
|
| 4 |
+
|
| 5 |
+
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
|
| 6 |
+
|
| 7 |
+
1. Definitions.
|
| 8 |
+
|
| 9 |
+
"License" shall mean the terms and conditions for use, reproduction,
|
| 10 |
+
and distribution as defined by Sections 1 through 9 of this document.
|
| 11 |
+
|
| 12 |
+
"Licensor" shall mean the copyright owner or entity authorized by
|
| 13 |
+
the copyright owner that is granting the License.
|
| 14 |
+
|
| 15 |
+
"Legal Entity" shall mean the union of the acting entity and all
|
| 16 |
+
other entities that control, are controlled by, or are under common
|
| 17 |
+
control with that entity. For the purposes of this definition,
|
| 18 |
+
"control" means (i) the power, direct or indirect, to cause the
|
| 19 |
+
direction or management of such entity, whether by contract or
|
| 20 |
+
otherwise, or (ii) ownership of fifty percent (50%) or more of the
|
| 21 |
+
outstanding shares, or (iii) beneficial ownership of such entity.
|
| 22 |
+
|
| 23 |
+
"You" (or "Your") shall mean an individual or Legal Entity
|
| 24 |
+
exercising permissions granted by this License.
|
| 25 |
+
|
| 26 |
+
"Source" form shall mean the preferred form for making modifications,
|
| 27 |
+
including but not limited to software source code, documentation
|
| 28 |
+
source, and configuration files.
|
| 29 |
+
|
| 30 |
+
"Object" form shall mean any form resulting from mechanical
|
| 31 |
+
transformation or translation of a Source form, including but
|
| 32 |
+
not limited to compiled object code, generated documentation,
|
| 33 |
+
and conversions to other media types.
|
| 34 |
+
|
| 35 |
+
"Work" shall mean the work of authorship, whether in Source or
|
| 36 |
+
Object form, made available under the License, as indicated by a
|
| 37 |
+
copyright notice that is included in or attached to the work
|
| 38 |
+
(an example is provided in the Appendix below).
|
| 39 |
+
|
| 40 |
+
"Derivative Works" shall mean any work, whether in Source or Object
|
| 41 |
+
form, that is based on (or derived from) the Work and for which the
|
| 42 |
+
editorial revisions, annotations, elaborations, or other modifications
|
| 43 |
+
represent, as a whole, an original work of authorship. For the purposes
|
| 44 |
+
of this License, Derivative Works shall not include works that remain
|
| 45 |
+
separable from, or merely link (or bind by name) to the interfaces of,
|
| 46 |
+
the Work and Derivative Works thereof.
|
| 47 |
+
|
| 48 |
+
"Contribution" shall mean any work of authorship, including
|
| 49 |
+
the original version of the Work and any modifications or additions
|
| 50 |
+
to that Work or Derivative Works thereof, that is intentionally
|
| 51 |
+
submitted to Licensor for inclusion in the Work by the copyright owner
|
| 52 |
+
or by an individual or Legal Entity authorized to submit on behalf of
|
| 53 |
+
the copyright owner. For the purposes of this definition, "submitted"
|
| 54 |
+
means any form of electronic, verbal, or written communication sent
|
| 55 |
+
to the Licensor or its representatives, including but not limited to
|
| 56 |
+
communication on electronic mailing lists, source code control systems,
|
| 57 |
+
and issue tracking systems that are managed by, or on behalf of, the
|
| 58 |
+
Licensor for the purpose of discussing and improving the Work, but
|
| 59 |
+
excluding communication that is conspicuously marked or otherwise
|
| 60 |
+
designated in writing by the copyright owner as "Not a Contribution."
|
| 61 |
+
|
| 62 |
+
"Contributor" shall mean Licensor and any individual or Legal Entity
|
| 63 |
+
on behalf of whom a Contribution has been received by Licensor and
|
| 64 |
+
subsequently incorporated within the Work.
|
| 65 |
+
|
| 66 |
+
2. Grant of Copyright License. Subject to the terms and conditions of
|
| 67 |
+
this License, each Contributor hereby grants to You a perpetual,
|
| 68 |
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
| 69 |
+
copyright license to reproduce, prepare Derivative Works of,
|
| 70 |
+
publicly display, publicly perform, sublicense, and distribute the
|
| 71 |
+
Work and such Derivative Works in Source or Object form.
|
| 72 |
+
|
| 73 |
+
3. Grant of Patent License. Subject to the terms and conditions of
|
| 74 |
+
this License, each Contributor hereby grants to You a perpetual,
|
| 75 |
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
| 76 |
+
(except as stated in this section) patent license to make, have made,
|
| 77 |
+
use, offer to sell, sell, import, and otherwise transfer the Work,
|
| 78 |
+
where such license applies only to those patent claims licensable
|
| 79 |
+
by such Contributor that are necessarily infringed by their
|
| 80 |
+
Contribution(s) alone or by combination of their Contribution(s)
|
| 81 |
+
with the Work to which such Contribution(s) was submitted. If You
|
| 82 |
+
institute patent litigation against any entity (including a
|
| 83 |
+
cross-claim or counterclaim in a lawsuit) alleging that the Work
|
| 84 |
+
or a Contribution incorporated within the Work constitutes direct
|
| 85 |
+
or contributory patent infringement, then any patent licenses
|
| 86 |
+
granted to You under this License for that Work shall terminate
|
| 87 |
+
as of the date such litigation is filed.
|
| 88 |
+
|
| 89 |
+
4. Redistribution. You may reproduce and distribute copies of the
|
| 90 |
+
Work or Derivative Works thereof in any medium, with or without
|
| 91 |
+
modifications, and in Source or Object form, provided that You
|
| 92 |
+
meet the following conditions:
|
| 93 |
+
|
| 94 |
+
(a) You must give any other recipients of the Work or
|
| 95 |
+
Derivative Works a copy of this License; and
|
| 96 |
+
|
| 97 |
+
(b) You must cause any modified files to carry prominent notices
|
| 98 |
+
stating that You changed the files; and
|
| 99 |
+
|
| 100 |
+
(c) You must retain, in the Source form of any Derivative Works
|
| 101 |
+
that You distribute, all copyright, patent, trademark, and
|
| 102 |
+
attribution notices from the Source form of the Work,
|
| 103 |
+
excluding those notices that do not pertain to any part of
|
| 104 |
+
the Derivative Works; and
|
| 105 |
+
|
| 106 |
+
(d) If the Work includes a "NOTICE" text file as part of its
|
| 107 |
+
distribution, then any Derivative Works that You distribute must
|
| 108 |
+
include a readable copy of the attribution notices contained
|
| 109 |
+
within such NOTICE file, excluding those notices that do not
|
| 110 |
+
pertain to any part of the Derivative Works, in at least one
|
| 111 |
+
of the following places: within a NOTICE text file distributed
|
| 112 |
+
as part of the Derivative Works; within the Source form or
|
| 113 |
+
documentation, if provided along with the Derivative Works; or,
|
| 114 |
+
within a display generated by the Derivative Works, if and
|
| 115 |
+
wherever such third-party notices normally appear. The contents
|
| 116 |
+
of the NOTICE file are for informational purposes only and
|
| 117 |
+
do not modify the License. You may add Your own attribution
|
| 118 |
+
notices within Derivative Works that You distribute, alongside
|
| 119 |
+
or as an addendum to the NOTICE text from the Work, provided
|
| 120 |
+
that such additional attribution notices cannot be construed
|
| 121 |
+
as modifying the License.
|
| 122 |
+
|
| 123 |
+
You may add Your own copyright statement to Your modifications and
|
| 124 |
+
may provide additional or different license terms and conditions
|
| 125 |
+
for use, reproduction, or distribution of Your modifications, or
|
| 126 |
+
for any such Derivative Works as a whole, provided Your use,
|
| 127 |
+
reproduction, and distribution of the Work otherwise complies with
|
| 128 |
+
the conditions stated in this License.
|
| 129 |
+
|
| 130 |
+
5. Submission of Contributions. Unless You explicitly state otherwise,
|
| 131 |
+
any Contribution intentionally submitted for inclusion in the Work
|
| 132 |
+
by You to the Licensor shall be under the terms and conditions of
|
| 133 |
+
this License, without any additional terms or conditions.
|
| 134 |
+
Notwithstanding the above, nothing herein shall supersede or modify
|
| 135 |
+
the terms of any separate license agreement you may have executed
|
| 136 |
+
with Licensor regarding such Contributions.
|
| 137 |
+
|
| 138 |
+
6. Trademarks. This License does not grant permission to use the trade
|
| 139 |
+
names, trademarks, service marks, or product names of the Licensor,
|
| 140 |
+
except as required for reasonable and customary use in describing the
|
| 141 |
+
origin of the Work and reproducing the content of the NOTICE file.
|
| 142 |
+
|
| 143 |
+
7. Disclaimer of Warranty. Unless required by applicable law or
|
| 144 |
+
agreed to in writing, Licensor provides the Work (and each
|
| 145 |
+
Contributor provides its Contributions) on an "AS IS" BASIS,
|
| 146 |
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
|
| 147 |
+
implied, including, without limitation, any warranties or conditions
|
| 148 |
+
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
|
| 149 |
+
PARTICULAR PURPOSE. You are solely responsible for determining the
|
| 150 |
+
appropriateness of using or redistributing the Work and assume any
|
| 151 |
+
risks associated with Your exercise of permissions under this License.
|
| 152 |
+
|
| 153 |
+
8. Limitation of Liability. In no event and under no legal theory,
|
| 154 |
+
whether in tort (including negligence), contract, or otherwise,
|
| 155 |
+
unless required by applicable law (such as deliberate and grossly
|
| 156 |
+
negligent acts) or agreed to in writing, shall any Contributor be
|
| 157 |
+
liable to You for damages, including any direct, indirect, special,
|
| 158 |
+
incidental, or consequential damages of any character arising as a
|
| 159 |
+
result of this License or out of the use or inability to use the
|
| 160 |
+
Work (including but not limited to damages for loss of goodwill,
|
| 161 |
+
work stoppage, computer failure or malfunction, or any and all
|
| 162 |
+
other commercial damages or losses), even if such Contributor
|
| 163 |
+
has been advised of the possibility of such damages.
|
| 164 |
+
|
| 165 |
+
9. Accepting Warranty or Additional Liability. While redistributing
|
| 166 |
+
the Work or Derivative Works thereof, You may choose to offer,
|
| 167 |
+
and charge a fee for, acceptance of support, warranty, indemnity,
|
| 168 |
+
or other liability obligations and/or rights consistent with this
|
| 169 |
+
License. However, in accepting such obligations, You may act only
|
| 170 |
+
on Your own behalf and on Your sole responsibility, not on behalf
|
| 171 |
+
of any other Contributor, and only if You agree to indemnify,
|
| 172 |
+
defend, and hold each Contributor harmless for any liability
|
| 173 |
+
incurred by, or claims asserted against, such Contributor by reason
|
| 174 |
+
of your accepting any such warranty or additional liability.
|
| 175 |
+
|
| 176 |
+
END OF TERMS AND CONDITIONS
|
| 177 |
+
|
| 178 |
+
APPENDIX: How to apply the Apache License to your work.
|
| 179 |
+
|
| 180 |
+
To apply the Apache License to your work, attach the following
|
| 181 |
+
boilerplate notice, with the fields enclosed by brackets "[]"
|
| 182 |
+
replaced with your own identifying information. (Don't include
|
| 183 |
+
the brackets!) The text should be enclosed in the appropriate
|
| 184 |
+
comment syntax for the file format. We also recommend that a
|
| 185 |
+
file or class name and description of purpose be included on the
|
| 186 |
+
same "printed page" as the copyright notice for easier
|
| 187 |
+
identification within third-party archives.
|
| 188 |
+
|
| 189 |
+
Copyright [yyyy] [name of copyright owner]
|
| 190 |
+
|
| 191 |
+
Licensed under the Apache License, Version 2.0 (the "License");
|
| 192 |
+
you may not use this file except in compliance with the License.
|
| 193 |
+
You may obtain a copy of the License at
|
| 194 |
+
|
| 195 |
+
http://www.apache.org/licenses/LICENSE-2.0
|
| 196 |
+
|
| 197 |
+
Unless required by applicable law or agreed to in writing, software
|
| 198 |
+
distributed under the License is distributed on an "AS IS" BASIS,
|
| 199 |
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
| 200 |
+
See the License for the specific language governing permissions and
|
| 201 |
+
limitations under the License.
|
README.md
CHANGED
|
@@ -1,8 +1,144 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
language:
|
| 4 |
-
-
|
| 5 |
-
-
|
| 6 |
-
|
| 7 |
-
-
|
| 8 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
+
base_model: openbmb/VoxCPM2
|
| 4 |
+
pipeline_tag: text-to-speech
|
| 5 |
+
library_name: ncnn
|
| 6 |
+
tags:
|
| 7 |
+
- voxcpm2
|
| 8 |
+
- ncnn
|
| 9 |
+
- text-to-speech
|
| 10 |
+
- speech-synthesis
|
| 11 |
+
- voice-cloning
|
| 12 |
+
- edge-ai
|
| 13 |
language:
|
| 14 |
+
- ar
|
| 15 |
+
- my
|
| 16 |
+
- zh
|
| 17 |
+
- da
|
| 18 |
+
- nl
|
| 19 |
+
- en
|
| 20 |
+
- fi
|
| 21 |
+
- fr
|
| 22 |
+
- de
|
| 23 |
+
- el
|
| 24 |
+
- he
|
| 25 |
+
- hi
|
| 26 |
+
- id
|
| 27 |
+
- it
|
| 28 |
+
- ja
|
| 29 |
+
- km
|
| 30 |
+
- ko
|
| 31 |
+
- lo
|
| 32 |
+
- ms
|
| 33 |
+
- "no"
|
| 34 |
+
- pl
|
| 35 |
+
- pt
|
| 36 |
+
- ru
|
| 37 |
+
- es
|
| 38 |
+
- sw
|
| 39 |
+
- sv
|
| 40 |
+
- tl
|
| 41 |
+
- th
|
| 42 |
+
- tr
|
| 43 |
+
- vi
|
| 44 |
+
---
|
| 45 |
+
|
| 46 |
+
# VoxCPM2 NCNN
|
| 47 |
+
|
| 48 |
+
This repository contains an NCNN export of [openbmb/VoxCPM2](https://huggingface.co/openbmb/VoxCPM2) for use with the `voxcpm2-ncnn` C++ runtime.
|
| 49 |
+
|
| 50 |
+
It is a converted runtime asset package, not a newly trained or fine-tuned model. The model weights keep the same Apache-2.0 license as the upstream VoxCPM2 release.
|
| 51 |
+
|
| 52 |
+
## Model Details
|
| 53 |
+
|
| 54 |
+
- Base model: `openbmb/VoxCPM2`
|
| 55 |
+
- Format: NCNN `.param` / `.bin` component graphs
|
| 56 |
+
- Task: multilingual text-to-speech
|
| 57 |
+
- Output audio: 48 kHz mono PCM, written by the runtime through FFmpeg
|
| 58 |
+
- Runtime target: `voxcpm2-ncnn`
|
| 59 |
+
- License: Apache-2.0 for the model assets
|
| 60 |
+
|
| 61 |
+
VoxCPM2 is a multilingual controllable speech generation model. The upstream release describes support for 30 languages, 9 Chinese dialects, voice design, style-controllable voice cloning, and high-fidelity continuation cloning. This NCNN package targets the modes currently exposed by the `voxcpm2-ncnn` runtime.
|
| 62 |
+
|
| 63 |
+
## Files
|
| 64 |
+
|
| 65 |
+
The exported model directory contains the runtime assets and a local `LICENSE` copy. The repository root keeps an additional `LICENSE` copy for hosting tools that expect the license at the top level.
|
| 66 |
+
|
| 67 |
+
- `model.json`: NCNN component manifest and runtime settings
|
| 68 |
+
- `*.ncnn.param`, `*.ncnn.bin`: exported NCNN component graphs and weights
|
| 69 |
+
- `tokenizer.json`, `tokenizer_config.json`, `special_tokens_map.json`: tokenizer assets
|
| 70 |
+
- `config.json`, `tokenization_voxcpm2.py`: upstream configuration/tokenizer reference files
|
| 71 |
+
- `LICENSE`: Apache-2.0 license text for the model assets
|
| 72 |
+
|
| 73 |
+
Export-time intermediate files such as TorchScript, PNNX graphs, and generated Python wrappers are intentionally not included.
|
| 74 |
+
|
| 75 |
+
## Usage
|
| 76 |
+
|
| 77 |
+
Download the model assets into the runtime repository:
|
| 78 |
+
|
| 79 |
+
```bash
|
| 80 |
+
huggingface-cli download lyrin/voxpm2-ncnn --local-dir assets
|
| 81 |
+
```
|
| 82 |
+
|
| 83 |
+
Smoke-test the NCNN components:
|
| 84 |
+
|
| 85 |
+
```bash
|
| 86 |
+
xmake run voxcpm2 -m assets/voxcpm2 --smoke-components
|
| 87 |
+
```
|
| 88 |
+
|
| 89 |
+
Generate speech from text:
|
| 90 |
+
|
| 91 |
+
```bash
|
| 92 |
+
xmake run voxcpm2 -m assets/voxcpm2 \
|
| 93 |
+
-t "你好，欢迎使用 VoxCPM2 NCNN。" \
|
| 94 |
+
-o out.wav
|
| 95 |
+
```
|
| 96 |
+
|
| 97 |
+
Use prompt continuation with prompt audio:
|
| 98 |
+
|
| 99 |
+
```bash
|
| 100 |
+
xmake run voxcpm2 -m assets/voxcpm2 \
|
| 101 |
+
-t "这是续写测试。" \
|
| 102 |
+
--prompt "你好。" \
|
| 103 |
+
--prompt-audio prompt.wav \
|
| 104 |
+
-o out.wav
|
| 105 |
+
```
|
| 106 |
+
|
| 107 |
+
Use reference audio:
|
| 108 |
+
|
| 109 |
+
```bash
|
| 110 |
+
xmake run voxcpm2 -m assets/voxcpm2 \
|
| 111 |
+
-t "这是参考音频测试。" \
|
| 112 |
+
--reference-audio reference.wav \
|
| 113 |
+
-o out.flac
|
| 114 |
+
```
|
| 115 |
+
|
| 116 |
+
The output format is inferred from the `-o` extension.
|
| 117 |
+
|
| 118 |
+
## Conversion Notes
|
| 119 |
+
|
| 120 |
+
This package splits VoxCPM2 into NCNN component graphs:
|
| 121 |
+
|
| 122 |
+
- text embedding
|
| 123 |
+
- base and residual decoder steps
|
| 124 |
+
- FSQ and projection layers
|
| 125 |
+
- DiT estimator
|
| 126 |
+
- stop-token head
|
| 127 |
+
- AudioVAE encoder and decoder
|
| 128 |
+
|
| 129 |
+
The runtime uses a page-style KV cache internally while adapting to the current exported decoder-step NCNN graphs.
|
| 130 |
+
|
| 131 |
+
## Limitations
|
| 132 |
+
|
| 133 |
+
- This is a conversion package; numerical behavior and performance can differ from the upstream PyTorch runtime.
|
| 134 |
+
- Not all upstream inference modes are necessarily exposed by the C++ runtime.
|
| 135 |
+
- Quality, speaker similarity, latency, and memory use depend on the NCNN build, device, Vulkan driver, and input audio quality.
|
| 136 |
+
- Generated speech and voice cloning should be used only with appropriate rights, consent, and safety review.
|
| 137 |
+
|
| 138 |
+
## Attribution
|
| 139 |
+
|
| 140 |
+
The original VoxCPM2 model is by OpenBMB / ModelBest. Please refer to the upstream [VoxCPM2 model card](https://huggingface.co/openbmb/VoxCPM2), [project repository](https://github.com/OpenBMB/VoxCPM), and [technical report](https://arxiv.org/abs/2606.06928) for model architecture, training, evaluation, and intended-use details.
|
| 141 |
+
|
| 142 |
+
## License
|
| 143 |
+
|
| 144 |
+
The model assets in this repository are released under Apache-2.0, matching the upstream VoxCPM2 release. The license text is included in `LICENSE`.
|
audio_vae_encoder.pnnx.bin
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:86bf64bcb7354bc480553ecd70af11da689240e32ccdc852429f1696fd235840
|
| 3 |
-
size 192180216
|
|
|
|
|
|
|
|
|
|
|
|
audio_vae_encoder.pnnx.param
DELETED
|
@@ -1,149 +0,0 @@
|
|
| 1 |
-
7767517
|
| 2 |
-
147 146
|
| 3 |
-
pnnx.Input pnnx_input_0 0 1 0 #0=(1,5120)f32
|
| 4 |
-
torch.unsqueeze torch.unsqueeze_57 1 1 0 1 dim=1 $input=0 #0=(1,5120)f32 #1=(1,1,5120)f32
|
| 5 |
-
F.pad F.pad_58 1 1 1 2 mode=constant pad=(6,0) value=None $input=1 #1=(1,1,5120)f32 #2=(1,1,5126)f32
|
| 6 |
-
nn.Conv1d conv1d_0 1 1 2 3 bias=True dilation=(1) groups=1 in_channels=1 kernel_size=(7) out_channels=128 padding=(0) padding_mode=zeros stride=(1) @bias=(128)f32 @weight=(128,1,7)f32 $input=2 #2=(1,1,5126)f32 #3=(1,128,5120)f32
|
| 7 |
-
pnnx.Attribute audio_vae.encoder.block.1.block.0.block.0 0 1 4 @data=(1,128,1)f32 #4=(1,128,1)f32
|
| 8 |
-
pnnx.Attribute pnnx_fold_114 0 1 5 @data=(1,128,1)f32 #5=(1,128,1)f32
|
| 9 |
-
pnnx.Expression pnnx_expr_481 3 1 3 5 4 6 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #3=(1,128,5120)f32 #5=(1,128,1)f32 #4=(1,128,1)f32 #6=(1,128,5120)f32
|
| 10 |
-
F.pad F.pad_59 1 1 6 7 mode=constant pad=(6,0) value=None $input=6 #6=(1,128,5120)f32 #7=(1,128,5126)f32
|
| 11 |
-
nn.Conv1d conv1d_1 1 1 7 8 bias=True dilation=(1) groups=128 in_channels=128 kernel_size=(7) out_channels=128 padding=(0) padding_mode=zeros stride=(1) @bias=(128)f32 @weight=(128,1,7)f32 $input=7 #7=(1,128,5126)f32 #8=(1,128,5120)f32
|
| 12 |
-
pnnx.Attribute audio_vae.encoder.block.1.block.0.block.2 0 1 9 @data=(1,128,1)f32 #9=(1,128,1)f32
|
| 13 |
-
pnnx.Attribute pnnx_fold_151 0 1 10 @data=(1,128,1)f32 #10=(1,128,1)f32
|
| 14 |
-
pnnx.Expression pnnx_expr_465 3 1 8 10 9 11 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #8=(1,128,5120)f32 #10=(1,128,1)f32 #9=(1,128,1)f32 #11=(1,128,5120)f32
|
| 15 |
-
nn.Conv1d padconv1d_0 1 1 11 12 bias=True dilation=(1) groups=1 in_channels=128 kernel_size=(1) out_channels=128 padding=(0) padding_mode=zeros stride=(1) @bias=(128)f32 @weight=(128,128,1)f32 $input=11 #11=(1,128,5120)f32 #12=(1,128,5120)f32
|
| 16 |
-
pnnx.Expression pnnx_expr_463 2 1 3 12 13 expr=add(@0,@1) #3=(1,128,5120)f32 #12=(1,128,5120)f32 #13=(1,128,5120)f32
|
| 17 |
-
pnnx.Attribute audio_vae.encoder.block.1.block.1.block.0 0 1 14 @data=(1,128,1)f32 #14=(1,128,1)f32
|
| 18 |
-
pnnx.Attribute pnnx_fold_202 0 1 15 @data=(1,128,1)f32 #15=(1,128,1)f32
|
| 19 |
-
pnnx.Expression pnnx_expr_445 3 1 13 15 14 16 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #13=(1,128,5120)f32 #15=(1,128,1)f32 #14=(1,128,1)f32 #16=(1,128,5120)f32
|
| 20 |
-
F.pad F.pad_61 1 1 16 17 mode=constant pad=(18,0) value=None $input=16 #16=(1,128,5120)f32 #17=(1,128,5138)f32
|
| 21 |
-
nn.Conv1d conv1d_3 1 1 17 18 bias=True dilation=(3) groups=128 in_channels=128 kernel_size=(7) out_channels=128 padding=(0) padding_mode=zeros stride=(1) @bias=(128)f32 @weight=(128,1,7)f32 $input=17 #17=(1,128,5138)f32 #18=(1,128,5120)f32
|
| 22 |
-
pnnx.Attribute audio_vae.encoder.block.1.block.1.block.2 0 1 19 @data=(1,128,1)f32 #19=(1,128,1)f32
|
| 23 |
-
pnnx.Attribute pnnx_fold_240 0 1 20 @data=(1,128,1)f32 #20=(1,128,1)f32
|
| 24 |
-
pnnx.Expression pnnx_expr_429 3 1 18 20 19 21 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #18=(1,128,5120)f32 #20=(1,128,1)f32 #19=(1,128,1)f32 #21=(1,128,5120)f32
|
| 25 |
-
nn.Conv1d padconv1d_1 1 1 21 22 bias=True dilation=(1) groups=1 in_channels=128 kernel_size=(1) out_channels=128 padding=(0) padding_mode=zeros stride=(1) @bias=(128)f32 @weight=(128,128,1)f32 $input=21 #21=(1,128,5120)f32 #22=(1,128,5120)f32
|
| 26 |
-
pnnx.Expression pnnx_expr_427 2 1 13 22 23 expr=add(@0,@1) #13=(1,128,5120)f32 #22=(1,128,5120)f32 #23=(1,128,5120)f32
|
| 27 |
-
pnnx.Attribute audio_vae.encoder.block.1.block.2.block.0 0 1 24 @data=(1,128,1)f32 #24=(1,128,1)f32
|
| 28 |
-
pnnx.Attribute pnnx_fold_291 0 1 25 @data=(1,128,1)f32 #25=(1,128,1)f32
|
| 29 |
-
pnnx.Expression pnnx_expr_409 3 1 23 25 24 26 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #23=(1,128,5120)f32 #25=(1,128,1)f32 #24=(1,128,1)f32 #26=(1,128,5120)f32
|
| 30 |
-
F.pad F.pad_63 1 1 26 27 mode=constant pad=(54,0) value=None $input=26 #26=(1,128,5120)f32 #27=(1,128,5174)f32
|
| 31 |
-
nn.Conv1d conv1d_5 1 1 27 28 bias=True dilation=(9) groups=128 in_channels=128 kernel_size=(7) out_channels=128 padding=(0) padding_mode=zeros stride=(1) @bias=(128)f32 @weight=(128,1,7)f32 $input=27 #27=(1,128,5174)f32 #28=(1,128,5120)f32
|
| 32 |
-
pnnx.Attribute audio_vae.encoder.block.1.block.2.block.2 0 1 29 @data=(1,128,1)f32 #29=(1,128,1)f32
|
| 33 |
-
pnnx.Attribute pnnx_fold_329 0 1 30 @data=(1,128,1)f32 #30=(1,128,1)f32
|
| 34 |
-
pnnx.Expression pnnx_expr_393 3 1 28 30 29 31 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #28=(1,128,5120)f32 #30=(1,128,1)f32 #29=(1,128,1)f32 #31=(1,128,5120)f32
|
| 35 |
-
nn.Conv1d padconv1d_2 1 1 31 32 bias=True dilation=(1) groups=1 in_channels=128 kernel_size=(1) out_channels=128 padding=(0) padding_mode=zeros stride=(1) @bias=(128)f32 @weight=(128,128,1)f32 $input=31 #31=(1,128,5120)f32 #32=(1,128,5120)f32
|
| 36 |
-
pnnx.Expression pnnx_expr_391 2 1 23 32 33 expr=add(@0,@1) #23=(1,128,5120)f32 #32=(1,128,5120)f32 #33=(1,128,5120)f32
|
| 37 |
-
pnnx.Attribute audio_vae.encoder.block.1.block.3 0 1 34 @data=(1,128,1)f32 #34=(1,128,1)f32
|
| 38 |
-
pnnx.Attribute pnnx_fold_365 0 1 35 @data=(1,128,1)f32 #35=(1,128,1)f32
|
| 39 |
-
pnnx.Expression pnnx_expr_375 3 1 33 35 34 36 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #33=(1,128,5120)f32 #35=(1,128,1)f32 #34=(1,128,1)f32 #36=(1,128,5120)f32
|
| 40 |
-
F.pad F.pad_65 1 1 36 37 mode=constant pad=(2,0) value=None $input=36 #36=(1,128,5120)f32 #37=(1,128,5122)f32
|
| 41 |
-
nn.Conv1d conv1d_7 1 1 37 38 bias=True dilation=(1) groups=1 in_channels=128 kernel_size=(4) out_channels=256 padding=(0) padding_mode=zeros stride=(2) @bias=(256)f32 @weight=(256,128,4)f32 $input=37 #37=(1,128,5122)f32 #38=(1,256,2560)f32
|
| 42 |
-
pnnx.Attribute audio_vae.encoder.block.2.block.0.block.0 0 1 39 @data=(1,256,1)f32 #39=(1,256,1)f32
|
| 43 |
-
pnnx.Attribute pnnx_fold_427 0 1 40 @data=(1,256,1)f32 #40=(1,256,1)f32
|
| 44 |
-
pnnx.Expression pnnx_expr_356 3 1 38 40 39 41 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #38=(1,256,2560)f32 #40=(1,256,1)f32 #39=(1,256,1)f32 #41=(1,256,2560)f32
|
| 45 |
-
F.pad F.pad_66 1 1 41 42 mode=constant pad=(6,0) value=None $input=41 #41=(1,256,2560)f32 #42=(1,256,2566)f32
|
| 46 |
-
nn.Conv1d conv1d_8 1 1 42 43 bias=True dilation=(1) groups=256 in_channels=256 kernel_size=(7) out_channels=256 padding=(0) padding_mode=zeros stride=(1) @bias=(256)f32 @weight=(256,1,7)f32 $input=42 #42=(1,256,2566)f32 #43=(1,256,2560)f32
|
| 47 |
-
pnnx.Attribute audio_vae.encoder.block.2.block.0.block.2 0 1 44 @data=(1,256,1)f32 #44=(1,256,1)f32
|
| 48 |
-
pnnx.Attribute pnnx_fold_464 0 1 45 @data=(1,256,1)f32 #45=(1,256,1)f32
|
| 49 |
-
pnnx.Expression pnnx_expr_340 3 1 43 45 44 46 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #43=(1,256,2560)f32 #45=(1,256,1)f32 #44=(1,256,1)f32 #46=(1,256,2560)f32
|
| 50 |
-
nn.Conv1d padconv1d_3 1 1 46 47 bias=True dilation=(1) groups=1 in_channels=256 kernel_size=(1) out_channels=256 padding=(0) padding_mode=zeros stride=(1) @bias=(256)f32 @weight=(256,256,1)f32 $input=46 #46=(1,256,2560)f32 #47=(1,256,2560)f32
|
| 51 |
-
pnnx.Expression pnnx_expr_338 2 1 38 47 48 expr=add(@0,@1) #38=(1,256,2560)f32 #47=(1,256,2560)f32 #48=(1,256,2560)f32
|
| 52 |
-
pnnx.Attribute audio_vae.encoder.block.2.block.1.block.0 0 1 49 @data=(1,256,1)f32 #49=(1,256,1)f32
|
| 53 |
-
pnnx.Attribute pnnx_fold_515 0 1 50 @data=(1,256,1)f32 #50=(1,256,1)f32
|
| 54 |
-
pnnx.Expression pnnx_expr_320 3 1 48 50 49 51 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #48=(1,256,2560)f32 #50=(1,256,1)f32 #49=(1,256,1)f32 #51=(1,256,2560)f32
|
| 55 |
-
F.pad F.pad_68 1 1 51 52 mode=constant pad=(18,0) value=None $input=51 #51=(1,256,2560)f32 #52=(1,256,2578)f32
|
| 56 |
-
nn.Conv1d conv1d_10 1 1 52 53 bias=True dilation=(3) groups=256 in_channels=256 kernel_size=(7) out_channels=256 padding=(0) padding_mode=zeros stride=(1) @bias=(256)f32 @weight=(256,1,7)f32 $input=52 #52=(1,256,2578)f32 #53=(1,256,2560)f32
|
| 57 |
-
pnnx.Attribute audio_vae.encoder.block.2.block.1.block.2 0 1 54 @data=(1,256,1)f32 #54=(1,256,1)f32
|
| 58 |
-
pnnx.Attribute pnnx_fold_553 0 1 55 @data=(1,256,1)f32 #55=(1,256,1)f32
|
| 59 |
-
pnnx.Expression pnnx_expr_304 3 1 53 55 54 56 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #53=(1,256,2560)f32 #55=(1,256,1)f32 #54=(1,256,1)f32 #56=(1,256,2560)f32
|
| 60 |
-
nn.Conv1d padconv1d_4 1 1 56 57 bias=True dilation=(1) groups=1 in_channels=256 kernel_size=(1) out_channels=256 padding=(0) padding_mode=zeros stride=(1) @bias=(256)f32 @weight=(256,256,1)f32 $input=56 #56=(1,256,2560)f32 #57=(1,256,2560)f32
|
| 61 |
-
pnnx.Expression pnnx_expr_302 2 1 48 57 58 expr=add(@0,@1) #48=(1,256,2560)f32 #57=(1,256,2560)f32 #58=(1,256,2560)f32
|
| 62 |
-
pnnx.Attribute audio_vae.encoder.block.2.block.2.block.0 0 1 59 @data=(1,256,1)f32 #59=(1,256,1)f32
|
| 63 |
-
pnnx.Attribute pnnx_fold_604 0 1 60 @data=(1,256,1)f32 #60=(1,256,1)f32
|
| 64 |
-
pnnx.Expression pnnx_expr_284 3 1 58 60 59 61 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #58=(1,256,2560)f32 #60=(1,256,1)f32 #59=(1,256,1)f32 #61=(1,256,2560)f32
|
| 65 |
-
F.pad F.pad_70 1 1 61 62 mode=constant pad=(54,0) value=None $input=61 #61=(1,256,2560)f32 #62=(1,256,2614)f32
|
| 66 |
-
nn.Conv1d conv1d_12 1 1 62 63 bias=True dilation=(9) groups=256 in_channels=256 kernel_size=(7) out_channels=256 padding=(0) padding_mode=zeros stride=(1) @bias=(256)f32 @weight=(256,1,7)f32 $input=62 #62=(1,256,2614)f32 #63=(1,256,2560)f32
|
| 67 |
-
pnnx.Attribute audio_vae.encoder.block.2.block.2.block.2 0 1 64 @data=(1,256,1)f32 #64=(1,256,1)f32
|
| 68 |
-
pnnx.Attribute pnnx_fold_642 0 1 65 @data=(1,256,1)f32 #65=(1,256,1)f32
|
| 69 |
-
pnnx.Expression pnnx_expr_268 3 1 63 65 64 66 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #63=(1,256,2560)f32 #65=(1,256,1)f32 #64=(1,256,1)f32 #66=(1,256,2560)f32
|
| 70 |
-
nn.Conv1d padconv1d_5 1 1 66 67 bias=True dilation=(1) groups=1 in_channels=256 kernel_size=(1) out_channels=256 padding=(0) padding_mode=zeros stride=(1) @bias=(256)f32 @weight=(256,256,1)f32 $input=66 #66=(1,256,2560)f32 #67=(1,256,2560)f32
|
| 71 |
-
pnnx.Expression pnnx_expr_266 2 1 58 67 68 expr=add(@0,@1) #58=(1,256,2560)f32 #67=(1,256,2560)f32 #68=(1,256,2560)f32
|
| 72 |
-
pnnx.Attribute audio_vae.encoder.block.2.block.3 0 1 69 @data=(1,256,1)f32 #69=(1,256,1)f32
|
| 73 |
-
pnnx.Attribute pnnx_fold_678 0 1 70 @data=(1,256,1)f32 #70=(1,256,1)f32
|
| 74 |
-
pnnx.Expression pnnx_expr_250 3 1 68 70 69 71 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #68=(1,256,2560)f32 #70=(1,256,1)f32 #69=(1,256,1)f32 #71=(1,256,2560)f32
|
| 75 |
-
F.pad F.pad_72 1 1 71 72 mode=constant pad=(5,0) value=None $input=71 #71=(1,256,2560)f32 #72=(1,256,2565)f32
|
| 76 |
-
nn.Conv1d conv1d_14 1 1 72 73 bias=True dilation=(1) groups=1 in_channels=256 kernel_size=(10) out_channels=512 padding=(0) padding_mode=zeros stride=(5) @bias=(512)f32 @weight=(512,256,10)f32 $input=72 #72=(1,256,2565)f32 #73=(1,512,512)f32
|
| 77 |
-
pnnx.Attribute audio_vae.encoder.block.3.block.0.block.0 0 1 74 @data=(1,512,1)f32 #74=(1,512,1)f32
|
| 78 |
-
pnnx.Attribute pnnx_fold_740 0 1 75 @data=(1,512,1)f32 #75=(1,512,1)f32
|
| 79 |
-
pnnx.Expression pnnx_expr_231 3 1 73 75 74 76 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #73=(1,512,512)f32 #75=(1,512,1)f32 #74=(1,512,1)f32 #76=(1,512,512)f32
|
| 80 |
-
F.pad F.pad_73 1 1 76 77 mode=constant pad=(6,0) value=None $input=76 #76=(1,512,512)f32 #77=(1,512,518)f32
|
| 81 |
-
nn.Conv1d conv1d_15 1 1 77 78 bias=True dilation=(1) groups=512 in_channels=512 kernel_size=(7) out_channels=512 padding=(0) padding_mode=zeros stride=(1) @bias=(512)f32 @weight=(512,1,7)f32 $input=77 #77=(1,512,518)f32 #78=(1,512,512)f32
|
| 82 |
-
pnnx.Attribute audio_vae.encoder.block.3.block.0.block.2 0 1 79 @data=(1,512,1)f32 #79=(1,512,1)f32
|
| 83 |
-
pnnx.Attribute pnnx_fold_777 0 1 80 @data=(1,512,1)f32 #80=(1,512,1)f32
|
| 84 |
-
pnnx.Expression pnnx_expr_215 3 1 78 80 79 81 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #78=(1,512,512)f32 #80=(1,512,1)f32 #79=(1,512,1)f32 #81=(1,512,512)f32
|
| 85 |
-
nn.Conv1d padconv1d_6 1 1 81 82 bias=True dilation=(1) groups=1 in_channels=512 kernel_size=(1) out_channels=512 padding=(0) padding_mode=zeros stride=(1) @bias=(512)f32 @weight=(512,512,1)f32 $input=81 #81=(1,512,512)f32 #82=(1,512,512)f32
|
| 86 |
-
pnnx.Expression pnnx_expr_213 2 1 73 82 83 expr=add(@0,@1) #73=(1,512,512)f32 #82=(1,512,512)f32 #83=(1,512,512)f32
|
| 87 |
-
pnnx.Attribute audio_vae.encoder.block.3.block.1.block.0 0 1 84 @data=(1,512,1)f32 #84=(1,512,1)f32
|
| 88 |
-
pnnx.Attribute pnnx_fold_828 0 1 85 @data=(1,512,1)f32 #85=(1,512,1)f32
|
| 89 |
-
pnnx.Expression pnnx_expr_195 3 1 83 85 84 86 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #83=(1,512,512)f32 #85=(1,512,1)f32 #84=(1,512,1)f32 #86=(1,512,512)f32
|
| 90 |
-
F.pad F.pad_75 1 1 86 87 mode=constant pad=(18,0) value=None $input=86 #86=(1,512,512)f32 #87=(1,512,530)f32
|
| 91 |
-
nn.Conv1d conv1d_17 1 1 87 88 bias=True dilation=(3) groups=512 in_channels=512 kernel_size=(7) out_channels=512 padding=(0) padding_mode=zeros stride=(1) @bias=(512)f32 @weight=(512,1,7)f32 $input=87 #87=(1,512,530)f32 #88=(1,512,512)f32
|
| 92 |
-
pnnx.Attribute audio_vae.encoder.block.3.block.1.block.2 0 1 89 @data=(1,512,1)f32 #89=(1,512,1)f32
|
| 93 |
-
pnnx.Attribute pnnx_fold_866 0 1 90 @data=(1,512,1)f32 #90=(1,512,1)f32
|
| 94 |
-
pnnx.Expression pnnx_expr_179 3 1 88 90 89 91 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #88=(1,512,512)f32 #90=(1,512,1)f32 #89=(1,512,1)f32 #91=(1,512,512)f32
|
| 95 |
-
nn.Conv1d padconv1d_7 1 1 91 92 bias=True dilation=(1) groups=1 in_channels=512 kernel_size=(1) out_channels=512 padding=(0) padding_mode=zeros stride=(1) @bias=(512)f32 @weight=(512,512,1)f32 $input=91 #91=(1,512,512)f32 #92=(1,512,512)f32
|
| 96 |
-
pnnx.Expression pnnx_expr_177 2 1 83 92 93 expr=add(@0,@1) #83=(1,512,512)f32 #92=(1,512,512)f32 #93=(1,512,512)f32
|
| 97 |
-
pnnx.Attribute audio_vae.encoder.block.3.block.2.block.0 0 1 94 @data=(1,512,1)f32 #94=(1,512,1)f32
|
| 98 |
-
pnnx.Attribute pnnx_fold_917 0 1 95 @data=(1,512,1)f32 #95=(1,512,1)f32
|
| 99 |
-
pnnx.Expression pnnx_expr_159 3 1 93 95 94 96 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #93=(1,512,512)f32 #95=(1,512,1)f32 #94=(1,512,1)f32 #96=(1,512,512)f32
|
| 100 |
-
F.pad F.pad_77 1 1 96 97 mode=constant pad=(54,0) value=None $input=96 #96=(1,512,512)f32 #97=(1,512,566)f32
|
| 101 |
-
nn.Conv1d conv1d_19 1 1 97 98 bias=True dilation=(9) groups=512 in_channels=512 kernel_size=(7) out_channels=512 padding=(0) padding_mode=zeros stride=(1) @bias=(512)f32 @weight=(512,1,7)f32 $input=97 #97=(1,512,566)f32 #98=(1,512,512)f32
|
| 102 |
-
pnnx.Attribute audio_vae.encoder.block.3.block.2.block.2 0 1 99 @data=(1,512,1)f32 #99=(1,512,1)f32
|
| 103 |
-
pnnx.Attribute pnnx_fold_955 0 1 100 @data=(1,512,1)f32 #100=(1,512,1)f32
|
| 104 |
-
pnnx.Expression pnnx_expr_143 3 1 98 100 99 101 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #98=(1,512,512)f32 #100=(1,512,1)f32 #99=(1,512,1)f32 #101=(1,512,512)f32
|
| 105 |
-
nn.Conv1d padconv1d_8 1 1 101 102 bias=True dilation=(1) groups=1 in_channels=512 kernel_size=(1) out_channels=512 padding=(0) padding_mode=zeros stride=(1) @bias=(512)f32 @weight=(512,512,1)f32 $input=101 #101=(1,512,512)f32 #102=(1,512,512)f32
|
| 106 |
-
pnnx.Expression pnnx_expr_141 2 1 93 102 103 expr=add(@0,@1) #93=(1,512,512)f32 #102=(1,512,512)f32 #103=(1,512,512)f32
|
| 107 |
-
pnnx.Attribute audio_vae.encoder.block.3.block.3 0 1 104 @data=(1,512,1)f32 #104=(1,512,1)f32
|
| 108 |
-
pnnx.Attribute pnnx_fold_991 0 1 105 @data=(1,512,1)f32 #105=(1,512,1)f32
|
| 109 |
-
pnnx.Expression pnnx_expr_125 3 1 103 105 104 106 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #103=(1,512,512)f32 #105=(1,512,1)f32 #104=(1,512,1)f32 #106=(1,512,512)f32
|
| 110 |
-
F.pad F.pad_79 1 1 106 107 mode=constant pad=(8,0) value=None $input=106 #106=(1,512,512)f32 #107=(1,512,520)f32
|
| 111 |
-
nn.Conv1d conv1d_21 1 1 107 108 bias=True dilation=(1) groups=1 in_channels=512 kernel_size=(16) out_channels=1024 padding=(0) padding_mode=zeros stride=(8) @bias=(1024)f32 @weight=(1024,512,16)f32 $input=107 #107=(1,512,520)f32 #108=(1,1024,64)f32
|
| 112 |
-
pnnx.Attribute audio_vae.encoder.block.4.block.0.block.0 0 1 109 @data=(1,1024,1)f32 #109=(1,1024,1)f32
|
| 113 |
-
pnnx.Attribute pnnx_fold_1053 0 1 110 @data=(1,1024,1)f32 #110=(1,1024,1)f32
|
| 114 |
-
pnnx.Expression pnnx_expr_106 3 1 108 110 109 111 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #108=(1,1024,64)f32 #110=(1,1024,1)f32 #109=(1,1024,1)f32 #111=(1,1024,64)f32
|
| 115 |
-
F.pad F.pad_80 1 1 111 112 mode=constant pad=(6,0) value=None $input=111 #111=(1,1024,64)f32 #112=(1,1024,70)f32
|
| 116 |
-
nn.Conv1d conv1d_22 1 1 112 113 bias=True dilation=(1) groups=1024 in_channels=1024 kernel_size=(7) out_channels=1024 padding=(0) padding_mode=zeros stride=(1) @bias=(1024)f32 @weight=(1024,1,7)f32 $input=112 #112=(1,1024,70)f32 #113=(1,1024,64)f32
|
| 117 |
-
pnnx.Attribute audio_vae.encoder.block.4.block.0.block.2 0 1 114 @data=(1,1024,1)f32 #114=(1,1024,1)f32
|
| 118 |
-
pnnx.Attribute pnnx_fold_1090 0 1 115 @data=(1,1024,1)f32 #115=(1,1024,1)f32
|
| 119 |
-
pnnx.Expression pnnx_expr_90 3 1 113 115 114 116 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #113=(1,1024,64)f32 #115=(1,1024,1)f32 #114=(1,1024,1)f32 #116=(1,1024,64)f32
|
| 120 |
-
nn.Conv1d padconv1d_9 1 1 116 117 bias=True dilation=(1) groups=1 in_channels=1024 kernel_size=(1) out_channels=1024 padding=(0) padding_mode=zeros stride=(1) @bias=(1024)f32 @weight=(1024,1024,1)f32 $input=116 #116=(1,1024,64)f32 #117=(1,1024,64)f32
|
| 121 |
-
pnnx.Expression pnnx_expr_88 2 1 108 117 118 expr=add(@0,@1) #108=(1,1024,64)f32 #117=(1,1024,64)f32 #118=(1,1024,64)f32
|
| 122 |
-
pnnx.Attribute audio_vae.encoder.block.4.block.1.block.0 0 1 119 @data=(1,1024,1)f32 #119=(1,1024,1)f32
|
| 123 |
-
pnnx.Attribute pnnx_fold_1141 0 1 120 @data=(1,1024,1)f32 #120=(1,1024,1)f32
|
| 124 |
-
pnnx.Expression pnnx_expr_70 3 1 118 120 119 121 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #118=(1,1024,64)f32 #120=(1,1024,1)f32 #119=(1,1024,1)f32 #121=(1,1024,64)f32
|
| 125 |
-
F.pad F.pad_82 1 1 121 122 mode=constant pad=(18,0) value=None $input=121 #121=(1,1024,64)f32 #122=(1,1024,82)f32
|
| 126 |
-
nn.Conv1d conv1d_24 1 1 122 123 bias=True dilation=(3) groups=1024 in_channels=1024 kernel_size=(7) out_channels=1024 padding=(0) padding_mode=zeros stride=(1) @bias=(1024)f32 @weight=(1024,1,7)f32 $input=122 #122=(1,1024,82)f32 #123=(1,1024,64)f32
|
| 127 |
-
pnnx.Attribute audio_vae.encoder.block.4.block.1.block.2 0 1 124 @data=(1,1024,1)f32 #124=(1,1024,1)f32
|
| 128 |
-
pnnx.Attribute pnnx_fold_1179 0 1 125 @data=(1,1024,1)f32 #125=(1,1024,1)f32
|
| 129 |
-
pnnx.Expression pnnx_expr_54 3 1 123 125 124 126 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #123=(1,1024,64)f32 #125=(1,1024,1)f32 #124=(1,1024,1)f32 #126=(1,1024,64)f32
|
| 130 |
-
nn.Conv1d padconv1d_10 1 1 126 127 bias=True dilation=(1) groups=1 in_channels=1024 kernel_size=(1) out_channels=1024 padding=(0) padding_mode=zeros stride=(1) @bias=(1024)f32 @weight=(1024,1024,1)f32 $input=126 #126=(1,1024,64)f32 #127=(1,1024,64)f32
|
| 131 |
-
pnnx.Expression pnnx_expr_52 2 1 118 127 128 expr=add(@0,@1) #118=(1,1024,64)f32 #127=(1,1024,64)f32 #128=(1,1024,64)f32
|
| 132 |
-
pnnx.Attribute audio_vae.encoder.block.4.block.2.block.0 0 1 129 @data=(1,1024,1)f32 #129=(1,1024,1)f32
|
| 133 |
-
pnnx.Attribute pnnx_fold_1230 0 1 130 @data=(1,1024,1)f32 #130=(1,1024,1)f32
|
| 134 |
-
pnnx.Expression pnnx_expr_34 3 1 128 130 129 131 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #128=(1,1024,64)f32 #130=(1,1024,1)f32 #129=(1,1024,1)f32 #131=(1,1024,64)f32
|
| 135 |
-
F.pad F.pad_84 1 1 131 132 mode=constant pad=(54,0) value=None $input=131 #131=(1,1024,64)f32 #132=(1,1024,118)f32
|
| 136 |
-
nn.Conv1d conv1d_26 1 1 132 133 bias=True dilation=(9) groups=1024 in_channels=1024 kernel_size=(7) out_channels=1024 padding=(0) padding_mode=zeros stride=(1) @bias=(1024)f32 @weight=(1024,1,7)f32 $input=132 #132=(1,1024,118)f32 #133=(1,1024,64)f32
|
| 137 |
-
pnnx.Attribute audio_vae.encoder.block.4.block.2.block.2 0 1 134 @data=(1,1024,1)f32 #134=(1,1024,1)f32
|
| 138 |
-
pnnx.Attribute pnnx_fold_1268 0 1 135 @data=(1,1024,1)f32 #135=(1,1024,1)f32
|
| 139 |
-
pnnx.Expression pnnx_expr_18 3 1 133 135 134 136 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #133=(1,1024,64)f32 #135=(1,1024,1)f32 #134=(1,1024,1)f32 #136=(1,1024,64)f32
|
| 140 |
-
nn.Conv1d padconv1d_11 1 1 136 137 bias=True dilation=(1) groups=1 in_channels=1024 kernel_size=(1) out_channels=1024 padding=(0) padding_mode=zeros stride=(1) @bias=(1024)f32 @weight=(1024,1024,1)f32 $input=136 #136=(1,1024,64)f32 #137=(1,1024,64)f32
|
| 141 |
-
pnnx.Expression pnnx_expr_16 2 1 128 137 138 expr=add(@0,@1) #128=(1,1024,64)f32 #137=(1,1024,64)f32 #138=(1,1024,64)f32
|
| 142 |
-
pnnx.Attribute audio_vae.encoder.block.4.block.3 0 1 139 @data=(1,1024,1)f32 #139=(1,1024,1)f32
|
| 143 |
-
pnnx.Attribute pnnx_fold_1304 0 1 140 @data=(1,1024,1)f32 #140=(1,1024,1)f32
|
| 144 |
-
pnnx.Expression pnnx_expr_0 3 1 138 140 139 141 expr=add(@0,mul(@1,pow(sin(mul(@2,@0)),2))) #138=(1,1024,64)f32 #140=(1,1024,1)f32 #139=(1,1024,1)f32 #141=(1,1024,64)f32
|
| 145 |
-
F.pad F.pad_86 1 1 141 142 mode=constant pad=(8,0) value=None $input=141 #141=(1,1024,64)f32 #142=(1,1024,72)f32
|
| 146 |
-
nn.Conv1d conv1d_28 1 1 142 143 bias=True dilation=(1) groups=1 in_channels=1024 kernel_size=(16) out_channels=2048 padding=(0) padding_mode=zeros stride=(8) @bias=(2048)f32 @weight=(2048,1024,16)f32 $input=142 #142=(1,1024,72)f32 #143=(1,2048,8)f32
|
| 147 |
-
F.pad F.pad_87 1 1 143 144 mode=constant pad=(2,0) value=None $input=143 #143=(1,2048,8)f32 #144=(1,2048,10)f32
|
| 148 |
-
nn.Conv1d conv1d_29 1 1 144 145 bias=True dilation=(1) groups=1 in_channels=2048 kernel_size=(3) out_channels=64 padding=(0) padding_mode=zeros stride=(1) @bias=(64)f32 @weight=(64,2048,3)f32 $input=144 #144=(1,2048,10)f32 #145=(1,64,8)f32
|
| 149 |
-
pnnx.Output pnnx_output_0 1 0 145 #145=(1,64,8)f32
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
audio_vae_encoder.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:27af108b21c56faa3d22fc8f140adfe47cb6e8d54735b2ce1d1c5408cf38d618
|
| 3 |
-
size 377232113
|
|
|
|
|
|
|
|
|
|
|
|
audio_vae_encoder_ncnn.py
DELETED
|
@@ -1,26 +0,0 @@
|
|
| 1 |
-
import numpy as np
|
| 2 |
-
import ncnn
|
| 3 |
-
import torch
|
| 4 |
-
|
| 5 |
-
def test_inference():
|
| 6 |
-
torch.manual_seed(0)
|
| 7 |
-
in0 = torch.rand(1, 5120, dtype=torch.float)
|
| 8 |
-
out = []
|
| 9 |
-
|
| 10 |
-
with ncnn.Net() as net:
|
| 11 |
-
net.load_param("/home/liyulin/Tools/voxcpm-ncnn/assets/voxcpm2/audio_vae_encoder.ncnn.param")
|
| 12 |
-
net.load_model("/home/liyulin/Tools/voxcpm-ncnn/assets/voxcpm2/audio_vae_encoder.ncnn.bin")
|
| 13 |
-
|
| 14 |
-
with net.create_extractor() as ex:
|
| 15 |
-
ex.input("in0", ncnn.Mat(in0.squeeze(0).numpy()).clone())
|
| 16 |
-
|
| 17 |
-
_, out0 = ex.extract("out0")
|
| 18 |
-
out.append(torch.from_numpy(np.array(out0)).unsqueeze(0))
|
| 19 |
-
|
| 20 |
-
if len(out) == 1:
|
| 21 |
-
return out[0]
|
| 22 |
-
else:
|
| 23 |
-
return tuple(out)
|
| 24 |
-
|
| 25 |
-
if __name__ == "__main__":
|
| 26 |
-
print(test_inference())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
audio_vae_encoder_pnnx.py
DELETED
|
@@ -1,378 +0,0 @@
|
|
| 1 |
-
# pnnx model stat
|
| 2 |
-
# model inputshape = [1,5120]f32
|
| 3 |
-
# FLOPS = 6.569G
|
| 4 |
-
# memory OPS = 122.184M
|
| 5 |
-
|
| 6 |
-
import os
|
| 7 |
-
import numpy as np
|
| 8 |
-
import tempfile, zipfile
|
| 9 |
-
import torch
|
| 10 |
-
import torch.nn as nn
|
| 11 |
-
import torch.nn.functional as F
|
| 12 |
-
try:
|
| 13 |
-
import torchvision
|
| 14 |
-
import torchaudio
|
| 15 |
-
except:
|
| 16 |
-
pass
|
| 17 |
-
|
| 18 |
-
class Model(nn.Module):
|
| 19 |
-
def __init__(self):
|
| 20 |
-
super(Model, self).__init__()
|
| 21 |
-
|
| 22 |
-
self.conv1d_0 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=1, kernel_size=(7,), out_channels=128, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 23 |
-
self.conv1d_1 = nn.Conv1d(bias=True, dilation=(1,), groups=128, in_channels=128, kernel_size=(7,), out_channels=128, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 24 |
-
self.padconv1d_0 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=128, kernel_size=(1,), out_channels=128, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 25 |
-
self.conv1d_3 = nn.Conv1d(bias=True, dilation=(3,), groups=128, in_channels=128, kernel_size=(7,), out_channels=128, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 26 |
-
self.padconv1d_1 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=128, kernel_size=(1,), out_channels=128, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 27 |
-
self.conv1d_5 = nn.Conv1d(bias=True, dilation=(9,), groups=128, in_channels=128, kernel_size=(7,), out_channels=128, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 28 |
-
self.padconv1d_2 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=128, kernel_size=(1,), out_channels=128, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 29 |
-
self.conv1d_7 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=128, kernel_size=(4,), out_channels=256, padding=(0,), padding_mode='zeros', stride=(2,))
|
| 30 |
-
self.conv1d_8 = nn.Conv1d(bias=True, dilation=(1,), groups=256, in_channels=256, kernel_size=(7,), out_channels=256, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 31 |
-
self.padconv1d_3 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=256, kernel_size=(1,), out_channels=256, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 32 |
-
self.conv1d_10 = nn.Conv1d(bias=True, dilation=(3,), groups=256, in_channels=256, kernel_size=(7,), out_channels=256, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 33 |
-
self.padconv1d_4 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=256, kernel_size=(1,), out_channels=256, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 34 |
-
self.conv1d_12 = nn.Conv1d(bias=True, dilation=(9,), groups=256, in_channels=256, kernel_size=(7,), out_channels=256, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 35 |
-
self.padconv1d_5 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=256, kernel_size=(1,), out_channels=256, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 36 |
-
self.conv1d_14 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=256, kernel_size=(10,), out_channels=512, padding=(0,), padding_mode='zeros', stride=(5,))
|
| 37 |
-
self.conv1d_15 = nn.Conv1d(bias=True, dilation=(1,), groups=512, in_channels=512, kernel_size=(7,), out_channels=512, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 38 |
-
self.padconv1d_6 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=512, kernel_size=(1,), out_channels=512, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 39 |
-
self.conv1d_17 = nn.Conv1d(bias=True, dilation=(3,), groups=512, in_channels=512, kernel_size=(7,), out_channels=512, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 40 |
-
self.padconv1d_7 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=512, kernel_size=(1,), out_channels=512, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 41 |
-
self.conv1d_19 = nn.Conv1d(bias=True, dilation=(9,), groups=512, in_channels=512, kernel_size=(7,), out_channels=512, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 42 |
-
self.padconv1d_8 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=512, kernel_size=(1,), out_channels=512, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 43 |
-
self.conv1d_21 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=512, kernel_size=(16,), out_channels=1024, padding=(0,), padding_mode='zeros', stride=(8,))
|
| 44 |
-
self.conv1d_22 = nn.Conv1d(bias=True, dilation=(1,), groups=1024, in_channels=1024, kernel_size=(7,), out_channels=1024, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 45 |
-
self.padconv1d_9 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=1024, kernel_size=(1,), out_channels=1024, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 46 |
-
self.conv1d_24 = nn.Conv1d(bias=True, dilation=(3,), groups=1024, in_channels=1024, kernel_size=(7,), out_channels=1024, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 47 |
-
self.padconv1d_10 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=1024, kernel_size=(1,), out_channels=1024, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 48 |
-
self.conv1d_26 = nn.Conv1d(bias=True, dilation=(9,), groups=1024, in_channels=1024, kernel_size=(7,), out_channels=1024, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 49 |
-
self.padconv1d_11 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=1024, kernel_size=(1,), out_channels=1024, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 50 |
-
self.conv1d_28 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=1024, kernel_size=(16,), out_channels=2048, padding=(0,), padding_mode='zeros', stride=(8,))
|
| 51 |
-
self.conv1d_29 = nn.Conv1d(bias=True, dilation=(1,), groups=1, in_channels=2048, kernel_size=(3,), out_channels=64, padding=(0,), padding_mode='zeros', stride=(1,))
|
| 52 |
-
|
| 53 |
-
archive = zipfile.ZipFile('/home/liyulin/Tools/voxcpm-ncnn/assets/voxcpm2/audio_vae_encoder.pnnx.bin', 'r')
|
| 54 |
-
self.conv1d_0.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_0.bias', (128), 'float32')
|
| 55 |
-
self.conv1d_0.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_0.weight', (128,1,7), 'float32')
|
| 56 |
-
self.conv1d_1.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_1.bias', (128), 'float32')
|
| 57 |
-
self.conv1d_1.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_1.weight', (128,1,7), 'float32')
|
| 58 |
-
self.padconv1d_0.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_0.bias', (128), 'float32')
|
| 59 |
-
self.padconv1d_0.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_0.weight', (128,128,1), 'float32')
|
| 60 |
-
self.conv1d_3.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_3.bias', (128), 'float32')
|
| 61 |
-
self.conv1d_3.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_3.weight', (128,1,7), 'float32')
|
| 62 |
-
self.padconv1d_1.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_1.bias', (128), 'float32')
|
| 63 |
-
self.padconv1d_1.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_1.weight', (128,128,1), 'float32')
|
| 64 |
-
self.conv1d_5.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_5.bias', (128), 'float32')
|
| 65 |
-
self.conv1d_5.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_5.weight', (128,1,7), 'float32')
|
| 66 |
-
self.padconv1d_2.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_2.bias', (128), 'float32')
|
| 67 |
-
self.padconv1d_2.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_2.weight', (128,128,1), 'float32')
|
| 68 |
-
self.conv1d_7.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_7.bias', (256), 'float32')
|
| 69 |
-
self.conv1d_7.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_7.weight', (256,128,4), 'float32')
|
| 70 |
-
self.conv1d_8.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_8.bias', (256), 'float32')
|
| 71 |
-
self.conv1d_8.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_8.weight', (256,1,7), 'float32')
|
| 72 |
-
self.padconv1d_3.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_3.bias', (256), 'float32')
|
| 73 |
-
self.padconv1d_3.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_3.weight', (256,256,1), 'float32')
|
| 74 |
-
self.conv1d_10.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_10.bias', (256), 'float32')
|
| 75 |
-
self.conv1d_10.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_10.weight', (256,1,7), 'float32')
|
| 76 |
-
self.padconv1d_4.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_4.bias', (256), 'float32')
|
| 77 |
-
self.padconv1d_4.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_4.weight', (256,256,1), 'float32')
|
| 78 |
-
self.conv1d_12.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_12.bias', (256), 'float32')
|
| 79 |
-
self.conv1d_12.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_12.weight', (256,1,7), 'float32')
|
| 80 |
-
self.padconv1d_5.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_5.bias', (256), 'float32')
|
| 81 |
-
self.padconv1d_5.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_5.weight', (256,256,1), 'float32')
|
| 82 |
-
self.conv1d_14.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_14.bias', (512), 'float32')
|
| 83 |
-
self.conv1d_14.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_14.weight', (512,256,10), 'float32')
|
| 84 |
-
self.conv1d_15.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_15.bias', (512), 'float32')
|
| 85 |
-
self.conv1d_15.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_15.weight', (512,1,7), 'float32')
|
| 86 |
-
self.padconv1d_6.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_6.bias', (512), 'float32')
|
| 87 |
-
self.padconv1d_6.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_6.weight', (512,512,1), 'float32')
|
| 88 |
-
self.conv1d_17.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_17.bias', (512), 'float32')
|
| 89 |
-
self.conv1d_17.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_17.weight', (512,1,7), 'float32')
|
| 90 |
-
self.padconv1d_7.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_7.bias', (512), 'float32')
|
| 91 |
-
self.padconv1d_7.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_7.weight', (512,512,1), 'float32')
|
| 92 |
-
self.conv1d_19.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_19.bias', (512), 'float32')
|
| 93 |
-
self.conv1d_19.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_19.weight', (512,1,7), 'float32')
|
| 94 |
-
self.padconv1d_8.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_8.bias', (512), 'float32')
|
| 95 |
-
self.padconv1d_8.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_8.weight', (512,512,1), 'float32')
|
| 96 |
-
self.conv1d_21.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_21.bias', (1024), 'float32')
|
| 97 |
-
self.conv1d_21.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_21.weight', (1024,512,16), 'float32')
|
| 98 |
-
self.conv1d_22.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_22.bias', (1024), 'float32')
|
| 99 |
-
self.conv1d_22.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_22.weight', (1024,1,7), 'float32')
|
| 100 |
-
self.padconv1d_9.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_9.bias', (1024), 'float32')
|
| 101 |
-
self.padconv1d_9.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_9.weight', (1024,1024,1), 'float32')
|
| 102 |
-
self.conv1d_24.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_24.bias', (1024), 'float32')
|
| 103 |
-
self.conv1d_24.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_24.weight', (1024,1,7), 'float32')
|
| 104 |
-
self.padconv1d_10.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_10.bias', (1024), 'float32')
|
| 105 |
-
self.padconv1d_10.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_10.weight', (1024,1024,1), 'float32')
|
| 106 |
-
self.conv1d_26.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_26.bias', (1024), 'float32')
|
| 107 |
-
self.conv1d_26.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_26.weight', (1024,1,7), 'float32')
|
| 108 |
-
self.padconv1d_11.bias = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_11.bias', (1024), 'float32')
|
| 109 |
-
self.padconv1d_11.weight = self.load_pnnx_bin_as_parameter(archive, 'padconv1d_11.weight', (1024,1024,1), 'float32')
|
| 110 |
-
self.conv1d_28.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_28.bias', (2048), 'float32')
|
| 111 |
-
self.conv1d_28.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_28.weight', (2048,1024,16), 'float32')
|
| 112 |
-
self.conv1d_29.bias = self.load_pnnx_bin_as_parameter(archive, 'conv1d_29.bias', (64), 'float32')
|
| 113 |
-
self.conv1d_29.weight = self.load_pnnx_bin_as_parameter(archive, 'conv1d_29.weight', (64,2048,3), 'float32')
|
| 114 |
-
self.audio_vae_encoder_block_1_block_0_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.1.block.0.block.0.data', (1,128,1,), 'float32')
|
| 115 |
-
self.pnnx_fold_114_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_114.data', (1,128,1,), 'float32')
|
| 116 |
-
self.audio_vae_encoder_block_1_block_0_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.1.block.0.block.2.data', (1,128,1,), 'float32')
|
| 117 |
-
self.pnnx_fold_151_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_151.data', (1,128,1,), 'float32')
|
| 118 |
-
self.audio_vae_encoder_block_1_block_1_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.1.block.1.block.0.data', (1,128,1,), 'float32')
|
| 119 |
-
self.pnnx_fold_202_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_202.data', (1,128,1,), 'float32')
|
| 120 |
-
self.audio_vae_encoder_block_1_block_1_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.1.block.1.block.2.data', (1,128,1,), 'float32')
|
| 121 |
-
self.pnnx_fold_240_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_240.data', (1,128,1,), 'float32')
|
| 122 |
-
self.audio_vae_encoder_block_1_block_2_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.1.block.2.block.0.data', (1,128,1,), 'float32')
|
| 123 |
-
self.pnnx_fold_291_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_291.data', (1,128,1,), 'float32')
|
| 124 |
-
self.audio_vae_encoder_block_1_block_2_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.1.block.2.block.2.data', (1,128,1,), 'float32')
|
| 125 |
-
self.pnnx_fold_329_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_329.data', (1,128,1,), 'float32')
|
| 126 |
-
self.audio_vae_encoder_block_1_block_3_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.1.block.3.data', (1,128,1,), 'float32')
|
| 127 |
-
self.pnnx_fold_365_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_365.data', (1,128,1,), 'float32')
|
| 128 |
-
self.audio_vae_encoder_block_2_block_0_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.2.block.0.block.0.data', (1,256,1,), 'float32')
|
| 129 |
-
self.pnnx_fold_427_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_427.data', (1,256,1,), 'float32')
|
| 130 |
-
self.audio_vae_encoder_block_2_block_0_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.2.block.0.block.2.data', (1,256,1,), 'float32')
|
| 131 |
-
self.pnnx_fold_464_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_464.data', (1,256,1,), 'float32')
|
| 132 |
-
self.audio_vae_encoder_block_2_block_1_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.2.block.1.block.0.data', (1,256,1,), 'float32')
|
| 133 |
-
self.pnnx_fold_515_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_515.data', (1,256,1,), 'float32')
|
| 134 |
-
self.audio_vae_encoder_block_2_block_1_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.2.block.1.block.2.data', (1,256,1,), 'float32')
|
| 135 |
-
self.pnnx_fold_553_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_553.data', (1,256,1,), 'float32')
|
| 136 |
-
self.audio_vae_encoder_block_2_block_2_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.2.block.2.block.0.data', (1,256,1,), 'float32')
|
| 137 |
-
self.pnnx_fold_604_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_604.data', (1,256,1,), 'float32')
|
| 138 |
-
self.audio_vae_encoder_block_2_block_2_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.2.block.2.block.2.data', (1,256,1,), 'float32')
|
| 139 |
-
self.pnnx_fold_642_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_642.data', (1,256,1,), 'float32')
|
| 140 |
-
self.audio_vae_encoder_block_2_block_3_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.2.block.3.data', (1,256,1,), 'float32')
|
| 141 |
-
self.pnnx_fold_678_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_678.data', (1,256,1,), 'float32')
|
| 142 |
-
self.audio_vae_encoder_block_3_block_0_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.3.block.0.block.0.data', (1,512,1,), 'float32')
|
| 143 |
-
self.pnnx_fold_740_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_740.data', (1,512,1,), 'float32')
|
| 144 |
-
self.audio_vae_encoder_block_3_block_0_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.3.block.0.block.2.data', (1,512,1,), 'float32')
|
| 145 |
-
self.pnnx_fold_777_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_777.data', (1,512,1,), 'float32')
|
| 146 |
-
self.audio_vae_encoder_block_3_block_1_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.3.block.1.block.0.data', (1,512,1,), 'float32')
|
| 147 |
-
self.pnnx_fold_828_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_828.data', (1,512,1,), 'float32')
|
| 148 |
-
self.audio_vae_encoder_block_3_block_1_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.3.block.1.block.2.data', (1,512,1,), 'float32')
|
| 149 |
-
self.pnnx_fold_866_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_866.data', (1,512,1,), 'float32')
|
| 150 |
-
self.audio_vae_encoder_block_3_block_2_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.3.block.2.block.0.data', (1,512,1,), 'float32')
|
| 151 |
-
self.pnnx_fold_917_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_917.data', (1,512,1,), 'float32')
|
| 152 |
-
self.audio_vae_encoder_block_3_block_2_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.3.block.2.block.2.data', (1,512,1,), 'float32')
|
| 153 |
-
self.pnnx_fold_955_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_955.data', (1,512,1,), 'float32')
|
| 154 |
-
self.audio_vae_encoder_block_3_block_3_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.3.block.3.data', (1,512,1,), 'float32')
|
| 155 |
-
self.pnnx_fold_991_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_991.data', (1,512,1,), 'float32')
|
| 156 |
-
self.audio_vae_encoder_block_4_block_0_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.4.block.0.block.0.data', (1,1024,1,), 'float32')
|
| 157 |
-
self.pnnx_fold_1053_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_1053.data', (1,1024,1,), 'float32')
|
| 158 |
-
self.audio_vae_encoder_block_4_block_0_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.4.block.0.block.2.data', (1,1024,1,), 'float32')
|
| 159 |
-
self.pnnx_fold_1090_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_1090.data', (1,1024,1,), 'float32')
|
| 160 |
-
self.audio_vae_encoder_block_4_block_1_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.4.block.1.block.0.data', (1,1024,1,), 'float32')
|
| 161 |
-
self.pnnx_fold_1141_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_1141.data', (1,1024,1,), 'float32')
|
| 162 |
-
self.audio_vae_encoder_block_4_block_1_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.4.block.1.block.2.data', (1,1024,1,), 'float32')
|
| 163 |
-
self.pnnx_fold_1179_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_1179.data', (1,1024,1,), 'float32')
|
| 164 |
-
self.audio_vae_encoder_block_4_block_2_block_0_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.4.block.2.block.0.data', (1,1024,1,), 'float32')
|
| 165 |
-
self.pnnx_fold_1230_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_1230.data', (1,1024,1,), 'float32')
|
| 166 |
-
self.audio_vae_encoder_block_4_block_2_block_2_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.4.block.2.block.2.data', (1,1024,1,), 'float32')
|
| 167 |
-
self.pnnx_fold_1268_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_1268.data', (1,1024,1,), 'float32')
|
| 168 |
-
self.audio_vae_encoder_block_4_block_3_data = self.load_pnnx_bin_as_parameter(archive, 'audio_vae.encoder.block.4.block.3.data', (1,1024,1,), 'float32')
|
| 169 |
-
self.pnnx_fold_1304_data = self.load_pnnx_bin_as_parameter(archive, 'pnnx_fold_1304.data', (1,1024,1,), 'float32')
|
| 170 |
-
archive.close()
|
| 171 |
-
|
| 172 |
-
def load_pnnx_bin_as_parameter(self, archive, key, shape, dtype, requires_grad=True):
|
| 173 |
-
return nn.Parameter(self.load_pnnx_bin_as_tensor(archive, key, shape, dtype), requires_grad)
|
| 174 |
-
|
| 175 |
-
def load_pnnx_bin_as_tensor(self, archive, key, shape, dtype):
|
| 176 |
-
fd, tmppath = tempfile.mkstemp()
|
| 177 |
-
with os.fdopen(fd, 'wb') as tmpf, archive.open(key) as keyfile:
|
| 178 |
-
tmpf.write(keyfile.read())
|
| 179 |
-
m = np.memmap(tmppath, dtype=dtype, mode='r', shape=shape).copy()
|
| 180 |
-
os.remove(tmppath)
|
| 181 |
-
return torch.from_numpy(m)
|
| 182 |
-
|
| 183 |
-
def forward(self, v_0):
|
| 184 |
-
v_1 = v_0.unsqueeze(1)
|
| 185 |
-
v_2 = F.pad(v_1, mode='constant', pad=(6,0), value=None)
|
| 186 |
-
v_3 = self.conv1d_0(v_2)
|
| 187 |
-
v_4 = self.audio_vae_encoder_block_1_block_0_block_0_data
|
| 188 |
-
v_5 = self.pnnx_fold_114_data
|
| 189 |
-
v_6 = (v_3 + (v_5 * torch.pow(torch.sin((v_4 * v_3)), 2)))
|
| 190 |
-
v_7 = F.pad(v_6, mode='constant', pad=(6,0), value=None)
|
| 191 |
-
v_8 = self.conv1d_1(v_7)
|
| 192 |
-
v_9 = self.audio_vae_encoder_block_1_block_0_block_2_data
|
| 193 |
-
v_10 = self.pnnx_fold_151_data
|
| 194 |
-
v_11 = (v_8 + (v_10 * torch.pow(torch.sin((v_9 * v_8)), 2)))
|
| 195 |
-
v_12 = self.padconv1d_0(v_11)
|
| 196 |
-
v_13 = (v_3 + v_12)
|
| 197 |
-
v_14 = self.audio_vae_encoder_block_1_block_1_block_0_data
|
| 198 |
-
v_15 = self.pnnx_fold_202_data
|
| 199 |
-
v_16 = (v_13 + (v_15 * torch.pow(torch.sin((v_14 * v_13)), 2)))
|
| 200 |
-
v_17 = F.pad(v_16, mode='constant', pad=(18,0), value=None)
|
| 201 |
-
v_18 = self.conv1d_3(v_17)
|
| 202 |
-
v_19 = self.audio_vae_encoder_block_1_block_1_block_2_data
|
| 203 |
-
v_20 = self.pnnx_fold_240_data
|
| 204 |
-
v_21 = (v_18 + (v_20 * torch.pow(torch.sin((v_19 * v_18)), 2)))
|
| 205 |
-
v_22 = self.padconv1d_1(v_21)
|
| 206 |
-
v_23 = (v_13 + v_22)
|
| 207 |
-
v_24 = self.audio_vae_encoder_block_1_block_2_block_0_data
|
| 208 |
-
v_25 = self.pnnx_fold_291_data
|
| 209 |
-
v_26 = (v_23 + (v_25 * torch.pow(torch.sin((v_24 * v_23)), 2)))
|
| 210 |
-
v_27 = F.pad(v_26, mode='constant', pad=(54,0), value=None)
|
| 211 |
-
v_28 = self.conv1d_5(v_27)
|
| 212 |
-
v_29 = self.audio_vae_encoder_block_1_block_2_block_2_data
|
| 213 |
-
v_30 = self.pnnx_fold_329_data
|
| 214 |
-
v_31 = (v_28 + (v_30 * torch.pow(torch.sin((v_29 * v_28)), 2)))
|
| 215 |
-
v_32 = self.padconv1d_2(v_31)
|
| 216 |
-
v_33 = (v_23 + v_32)
|
| 217 |
-
v_34 = self.audio_vae_encoder_block_1_block_3_data
|
| 218 |
-
v_35 = self.pnnx_fold_365_data
|
| 219 |
-
v_36 = (v_33 + (v_35 * torch.pow(torch.sin((v_34 * v_33)), 2)))
|
| 220 |
-
v_37 = F.pad(v_36, mode='constant', pad=(2,0), value=None)
|
| 221 |
-
v_38 = self.conv1d_7(v_37)
|
| 222 |
-
v_39 = self.audio_vae_encoder_block_2_block_0_block_0_data
|
| 223 |
-
v_40 = self.pnnx_fold_427_data
|
| 224 |
-
v_41 = (v_38 + (v_40 * torch.pow(torch.sin((v_39 * v_38)), 2)))
|
| 225 |
-
v_42 = F.pad(v_41, mode='constant', pad=(6,0), value=None)
|
| 226 |
-
v_43 = self.conv1d_8(v_42)
|
| 227 |
-
v_44 = self.audio_vae_encoder_block_2_block_0_block_2_data
|
| 228 |
-
v_45 = self.pnnx_fold_464_data
|
| 229 |
-
v_46 = (v_43 + (v_45 * torch.pow(torch.sin((v_44 * v_43)), 2)))
|
| 230 |
-
v_47 = self.padconv1d_3(v_46)
|
| 231 |
-
v_48 = (v_38 + v_47)
|
| 232 |
-
v_49 = self.audio_vae_encoder_block_2_block_1_block_0_data
|
| 233 |
-
v_50 = self.pnnx_fold_515_data
|
| 234 |
-
v_51 = (v_48 + (v_50 * torch.pow(torch.sin((v_49 * v_48)), 2)))
|
| 235 |
-
v_52 = F.pad(v_51, mode='constant', pad=(18,0), value=None)
|
| 236 |
-
v_53 = self.conv1d_10(v_52)
|
| 237 |
-
v_54 = self.audio_vae_encoder_block_2_block_1_block_2_data
|
| 238 |
-
v_55 = self.pnnx_fold_553_data
|
| 239 |
-
v_56 = (v_53 + (v_55 * torch.pow(torch.sin((v_54 * v_53)), 2)))
|
| 240 |
-
v_57 = self.padconv1d_4(v_56)
|
| 241 |
-
v_58 = (v_48 + v_57)
|
| 242 |
-
v_59 = self.audio_vae_encoder_block_2_block_2_block_0_data
|
| 243 |
-
v_60 = self.pnnx_fold_604_data
|
| 244 |
-
v_61 = (v_58 + (v_60 * torch.pow(torch.sin((v_59 * v_58)), 2)))
|
| 245 |
-
v_62 = F.pad(v_61, mode='constant', pad=(54,0), value=None)
|
| 246 |
-
v_63 = self.conv1d_12(v_62)
|
| 247 |
-
v_64 = self.audio_vae_encoder_block_2_block_2_block_2_data
|
| 248 |
-
v_65 = self.pnnx_fold_642_data
|
| 249 |
-
v_66 = (v_63 + (v_65 * torch.pow(torch.sin((v_64 * v_63)), 2)))
|
| 250 |
-
v_67 = self.padconv1d_5(v_66)
|
| 251 |
-
v_68 = (v_58 + v_67)
|
| 252 |
-
v_69 = self.audio_vae_encoder_block_2_block_3_data
|
| 253 |
-
v_70 = self.pnnx_fold_678_data
|
| 254 |
-
v_71 = (v_68 + (v_70 * torch.pow(torch.sin((v_69 * v_68)), 2)))
|
| 255 |
-
v_72 = F.pad(v_71, mode='constant', pad=(5,0), value=None)
|
| 256 |
-
v_73 = self.conv1d_14(v_72)
|
| 257 |
-
v_74 = self.audio_vae_encoder_block_3_block_0_block_0_data
|
| 258 |
-
v_75 = self.pnnx_fold_740_data
|
| 259 |
-
v_76 = (v_73 + (v_75 * torch.pow(torch.sin((v_74 * v_73)), 2)))
|
| 260 |
-
v_77 = F.pad(v_76, mode='constant', pad=(6,0), value=None)
|
| 261 |
-
v_78 = self.conv1d_15(v_77)
|
| 262 |
-
v_79 = self.audio_vae_encoder_block_3_block_0_block_2_data
|
| 263 |
-
v_80 = self.pnnx_fold_777_data
|
| 264 |
-
v_81 = (v_78 + (v_80 * torch.pow(torch.sin((v_79 * v_78)), 2)))
|
| 265 |
-
v_82 = self.padconv1d_6(v_81)
|
| 266 |
-
v_83 = (v_73 + v_82)
|
| 267 |
-
v_84 = self.audio_vae_encoder_block_3_block_1_block_0_data
|
| 268 |
-
v_85 = self.pnnx_fold_828_data
|
| 269 |
-
v_86 = (v_83 + (v_85 * torch.pow(torch.sin((v_84 * v_83)), 2)))
|
| 270 |
-
v_87 = F.pad(v_86, mode='constant', pad=(18,0), value=None)
|
| 271 |
-
v_88 = self.conv1d_17(v_87)
|
| 272 |
-
v_89 = self.audio_vae_encoder_block_3_block_1_block_2_data
|
| 273 |
-
v_90 = self.pnnx_fold_866_data
|
| 274 |
-
v_91 = (v_88 + (v_90 * torch.pow(torch.sin((v_89 * v_88)), 2)))
|
| 275 |
-
v_92 = self.padconv1d_7(v_91)
|
| 276 |
-
v_93 = (v_83 + v_92)
|
| 277 |
-
v_94 = self.audio_vae_encoder_block_3_block_2_block_0_data
|
| 278 |
-
v_95 = self.pnnx_fold_917_data
|
| 279 |
-
v_96 = (v_93 + (v_95 * torch.pow(torch.sin((v_94 * v_93)), 2)))
|
| 280 |
-
v_97 = F.pad(v_96, mode='constant', pad=(54,0), value=None)
|
| 281 |
-
v_98 = self.conv1d_19(v_97)
|
| 282 |
-
v_99 = self.audio_vae_encoder_block_3_block_2_block_2_data
|
| 283 |
-
v_100 = self.pnnx_fold_955_data
|
| 284 |
-
v_101 = (v_98 + (v_100 * torch.pow(torch.sin((v_99 * v_98)), 2)))
|
| 285 |
-
v_102 = self.padconv1d_8(v_101)
|
| 286 |
-
v_103 = (v_93 + v_102)
|
| 287 |
-
v_104 = self.audio_vae_encoder_block_3_block_3_data
|
| 288 |
-
v_105 = self.pnnx_fold_991_data
|
| 289 |
-
v_106 = (v_103 + (v_105 * torch.pow(torch.sin((v_104 * v_103)), 2)))
|
| 290 |
-
v_107 = F.pad(v_106, mode='constant', pad=(8,0), value=None)
|
| 291 |
-
v_108 = self.conv1d_21(v_107)
|
| 292 |
-
v_109 = self.audio_vae_encoder_block_4_block_0_block_0_data
|
| 293 |
-
v_110 = self.pnnx_fold_1053_data
|
| 294 |
-
v_111 = (v_108 + (v_110 * torch.pow(torch.sin((v_109 * v_108)), 2)))
|
| 295 |
-
v_112 = F.pad(v_111, mode='constant', pad=(6,0), value=None)
|
| 296 |
-
v_113 = self.conv1d_22(v_112)
|
| 297 |
-
v_114 = self.audio_vae_encoder_block_4_block_0_block_2_data
|
| 298 |
-
v_115 = self.pnnx_fold_1090_data
|
| 299 |
-
v_116 = (v_113 + (v_115 * torch.pow(torch.sin((v_114 * v_113)), 2)))
|
| 300 |
-
v_117 = self.padconv1d_9(v_116)
|
| 301 |
-
v_118 = (v_108 + v_117)
|
| 302 |
-
v_119 = self.audio_vae_encoder_block_4_block_1_block_0_data
|
| 303 |
-
v_120 = self.pnnx_fold_1141_data
|
| 304 |
-
v_121 = (v_118 + (v_120 * torch.pow(torch.sin((v_119 * v_118)), 2)))
|
| 305 |
-
v_122 = F.pad(v_121, mode='constant', pad=(18,0), value=None)
|
| 306 |
-
v_123 = self.conv1d_24(v_122)
|
| 307 |
-
v_124 = self.audio_vae_encoder_block_4_block_1_block_2_data
|
| 308 |
-
v_125 = self.pnnx_fold_1179_data
|
| 309 |
-
v_126 = (v_123 + (v_125 * torch.pow(torch.sin((v_124 * v_123)), 2)))
|
| 310 |
-
v_127 = self.padconv1d_10(v_126)
|
| 311 |
-
v_128 = (v_118 + v_127)
|
| 312 |
-
v_129 = self.audio_vae_encoder_block_4_block_2_block_0_data
|
| 313 |
-
v_130 = self.pnnx_fold_1230_data
|
| 314 |
-
v_131 = (v_128 + (v_130 * torch.pow(torch.sin((v_129 * v_128)), 2)))
|
| 315 |
-
v_132 = F.pad(v_131, mode='constant', pad=(54,0), value=None)
|
| 316 |
-
v_133 = self.conv1d_26(v_132)
|
| 317 |
-
v_134 = self.audio_vae_encoder_block_4_block_2_block_2_data
|
| 318 |
-
v_135 = self.pnnx_fold_1268_data
|
| 319 |
-
v_136 = (v_133 + (v_135 * torch.pow(torch.sin((v_134 * v_133)), 2)))
|
| 320 |
-
v_137 = self.padconv1d_11(v_136)
|
| 321 |
-
v_138 = (v_128 + v_137)
|
| 322 |
-
v_139 = self.audio_vae_encoder_block_4_block_3_data
|
| 323 |
-
v_140 = self.pnnx_fold_1304_data
|
| 324 |
-
v_141 = (v_138 + (v_140 * torch.pow(torch.sin((v_139 * v_138)), 2)))
|
| 325 |
-
v_142 = F.pad(v_141, mode='constant', pad=(8,0), value=None)
|
| 326 |
-
v_143 = self.conv1d_28(v_142)
|
| 327 |
-
v_144 = F.pad(v_143, mode='constant', pad=(2,0), value=None)
|
| 328 |
-
v_145 = self.conv1d_29(v_144)
|
| 329 |
-
return v_145
|
| 330 |
-
|
| 331 |
-
def export_torchscript():
|
| 332 |
-
net = Model()
|
| 333 |
-
net.float()
|
| 334 |
-
net.eval()
|
| 335 |
-
|
| 336 |
-
torch.manual_seed(0)
|
| 337 |
-
v_0 = torch.rand(1, 5120, dtype=torch.float)
|
| 338 |
-
|
| 339 |
-
mod = torch.jit.trace(net, v_0)
|
| 340 |
-
mod.save("/home/liyulin/Tools/voxcpm-ncnn/assets/voxcpm2/audio_vae_encoder_pnnx.py.pt")
|
| 341 |
-
|
| 342 |
-
def export_onnx():
|
| 343 |
-
net = Model()
|
| 344 |
-
net.float()
|
| 345 |
-
net.eval()
|
| 346 |
-
|
| 347 |
-
torch.manual_seed(0)
|
| 348 |
-
v_0 = torch.rand(1, 5120, dtype=torch.float)
|
| 349 |
-
|
| 350 |
-
torch.onnx.export(net, v_0, "/home/liyulin/Tools/voxcpm-ncnn/assets/voxcpm2/audio_vae_encoder_pnnx.py.onnx", export_params=True, operator_export_type=torch.onnx.OperatorExportTypes.ONNX_ATEN_FALLBACK, opset_version=13, input_names=['in0'], output_names=['out0'])
|
| 351 |
-
|
| 352 |
-
def export_pnnx():
|
| 353 |
-
net = Model()
|
| 354 |
-
net.float()
|
| 355 |
-
net.eval()
|
| 356 |
-
|
| 357 |
-
torch.manual_seed(0)
|
| 358 |
-
v_0 = torch.rand(1, 5120, dtype=torch.float)
|
| 359 |
-
|
| 360 |
-
import pnnx
|
| 361 |
-
pnnx.export(net, "/home/liyulin/Tools/voxcpm-ncnn/assets/voxcpm2/audio_vae_encoder_pnnx.py.pt", v_0)
|
| 362 |
-
|
| 363 |
-
def export_ncnn():
|
| 364 |
-
export_pnnx()
|
| 365 |
-
|
| 366 |
-
@torch.no_grad()
|
| 367 |
-
def test_inference():
|
| 368 |
-
net = Model()
|
| 369 |
-
net.float()
|
| 370 |
-
net.eval()
|
| 371 |
-
|
| 372 |
-
torch.manual_seed(0)
|
| 373 |
-
v_0 = torch.rand(1, 5120, dtype=torch.float)
|
| 374 |
-
|
| 375 |
-
return net(v_0)
|
| 376 |
-
|
| 377 |
-
if __name__ == "__main__":
|
| 378 |
-
print(test_inference())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tokenization_voxcpm2.py
DELETED
|
@@ -1,72 +0,0 @@
|
|
| 1 |
-
"""Custom tokenizer for VoxCPM2 that splits multi-character Chinese tokens.
|
| 2 |
-
|
| 3 |
-
VoxCPM2 was trained with ``mask_multichar_chinese_tokens`` which splits
|
| 4 |
-
multi-character Chinese tokens (e.g. "你好" -> ["你", "好"]) into individual
|
| 5 |
-
character IDs before embedding. The base LlamaTokenizerFast produces
|
| 6 |
-
multi-character Chinese tokens that the model has never seen during training,
|
| 7 |
-
yielding garbled Chinese audio output in downstream inference frameworks.
|
| 8 |
-
|
| 9 |
-
This module provides ``VoxCPM2Tokenizer`` which transparently applies the
|
| 10 |
-
character splitting inside ``encode()`` and ``__call__()``, so any downstream
|
| 11 |
-
consumer (vLLM, vLLM-Omni, Nano-vLLM, etc.) gets correct single-character
|
| 12 |
-
IDs without code changes.
|
| 13 |
-
"""
|
| 14 |
-
|
| 15 |
-
from transformers import LlamaTokenizerFast
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
class VoxCPM2Tokenizer(LlamaTokenizerFast):
|
| 19 |
-
|
| 20 |
-
def __init__(self, *args, **kwargs):
|
| 21 |
-
super().__init__(*args, **kwargs)
|
| 22 |
-
self._split_map = self._build_split_map()
|
| 23 |
-
|
| 24 |
-
def _build_split_map(self) -> dict[int, list[int]]:
|
| 25 |
-
vocab = self.get_vocab()
|
| 26 |
-
split_map: dict[int, list[int]] = {}
|
| 27 |
-
for token, tid in vocab.items():
|
| 28 |
-
clean = token.replace("\u2581", "")
|
| 29 |
-
if len(clean) >= 2 and all(self._is_cjk(c) for c in clean):
|
| 30 |
-
char_ids = self.convert_tokens_to_ids(list(clean))
|
| 31 |
-
if all(c != self.unk_token_id for c in char_ids):
|
| 32 |
-
split_map[tid] = char_ids
|
| 33 |
-
return split_map
|
| 34 |
-
|
| 35 |
-
@staticmethod
|
| 36 |
-
def _is_cjk(c: str) -> bool:
|
| 37 |
-
return (
|
| 38 |
-
"\u4e00" <= c <= "\u9fff"
|
| 39 |
-
or "\u3400" <= c <= "\u4dbf"
|
| 40 |
-
or "\uf900" <= c <= "\ufaff"
|
| 41 |
-
or "\U00020000" <= c <= "\U0002a6df"
|
| 42 |
-
)
|
| 43 |
-
|
| 44 |
-
def _expand_ids(self, ids: list[int]) -> list[int]:
|
| 45 |
-
result: list[int] = []
|
| 46 |
-
for tid in ids:
|
| 47 |
-
expansion = self._split_map.get(tid)
|
| 48 |
-
if expansion is not None:
|
| 49 |
-
result.extend(expansion)
|
| 50 |
-
else:
|
| 51 |
-
result.append(tid)
|
| 52 |
-
return result
|
| 53 |
-
|
| 54 |
-
def encode(self, text, *args, **kwargs):
|
| 55 |
-
ids = super().encode(text, *args, **kwargs)
|
| 56 |
-
return self._expand_ids(ids)
|
| 57 |
-
|
| 58 |
-
def __call__(self, text, *args, **kwargs):
|
| 59 |
-
result = super().__call__(text, *args, **kwargs)
|
| 60 |
-
if hasattr(result, "input_ids"):
|
| 61 |
-
ids = result["input_ids"]
|
| 62 |
-
if isinstance(ids, list) and ids and isinstance(ids[0], list):
|
| 63 |
-
result["input_ids"] = [self._expand_ids(x) for x in ids]
|
| 64 |
-
if "attention_mask" in result:
|
| 65 |
-
result["attention_mask"] = [
|
| 66 |
-
[1] * len(x) for x in result["input_ids"]
|
| 67 |
-
]
|
| 68 |
-
elif isinstance(ids, list):
|
| 69 |
-
result["input_ids"] = self._expand_ids(ids)
|
| 70 |
-
if "attention_mask" in result:
|
| 71 |
-
result["attention_mask"] = [1] * len(result["input_ids"])
|
| 72 |
-
return result
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|