initial commit

.gitattributes

@@ -33,3 +33,4 @@ 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
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,457 @@
+Tencent is pleased to support the open source community by making Unified_Audio_Schema available.
+
+Copyright (C) 2026 Tencent.  All rights reserved. 
+
+The open-source software and/or models included in this distribution may have been modified by Tencent (“Tencent Modifications”). All Tencent Modifications are Copyright (C) Tencent.
+
+Unified_Audio_Schema is licensed under the License Terms of Unified_Audio_Schema, except for the third-party components listed below, which remain licensed under their respective original terms. Unified_Audio_Schema does not impose any additional restrictions beyond those specified in the original licenses of these third-party components. Users are required to comply with all applicable terms and conditions of the original licenses and to ensure that the use of these third-party components conforms to all relevant laws and regulations.
+
+For the avoidance of doubt, Unified_Audio_Schema refers solely to code, parameters, and weights made publicly available by Tencent in accordance with the License Terms of Unified_Audio_Schema. 
+
+Terms of the Unified_Audio_Schema:
+--------------------------------------------------------------------
+Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+
+1. Unified_Audio_Schema IS NOT INTENDED FOR USE WITHIN THE EUROPEAN UNION.
+
+2. The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+
+
+Dependencies and Licenses:
+
+This open-source project, Unified_Audio_Schema, builds upon the following open-source models and/or software components, each of which remains licensed under its original license. Certain models or software may include modifications made by Tencent (“Tencent Modifications”), which are Copyright (C) Tencent.
+
+In case you believe there have been errors in the attribution below, you may submit the concerns to us for review and correction.
+
+
+Open Source Model Licensed under the Apache-2.0:
+--------------------------------------------------------------------
+1. Qwen2.5
+Copyright 2024 Alibaba Cloud
+Please note this model has been modified by Tencent in this distribution.
+
+2. Qwen3-Omni
+Copyright Alibaba Cloud
+Please note this model has been modified by Tencent in this distribution.
+
+ 
+
+Terms of the Apache-2.0：
+--------------------------------------------------------------------
+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.
+
+
+Open Source Software Licensed under the Apache-2.0
+--------------------------------------------------------------------
+1. MiMo-Audio-Eval
+Copyright 2025 Xiaomi Corporation.
+Please note this software has been modified by Tencent in this distribution.
+
+
+
+Terms of the Apache-2.0
+--------------------------------------------------------------------
+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 +1,175 @@
+---
+license: other
+license_name: license-term-of-unified-audio-schema
+language:
+- en
+- zh
+tags:
+- audio
+- speech
+- sound
+- music
+- audio-understanding
+- ASR
+- audio-captioning
+- TTS
+- audio-language-model
+- audio-llm
+- speech-to-text
+- text-to-speech
+- multimodal
+base_model:
+- Qwen/Qwen2.5-7B
+pipeline_tag: audio-text-to-text
+---
+
+# Beyond Transcription: Unified Audio Schema for Perception-Aware AudioLLMs
+
+**Unified Audio Schema** is a novel holistic framework for audio supervision that disentangles and restructures supervision across **transcription**, **paralinguistics**, and **non-linguistic events**.
+
+📄 [Paper](https://arxiv.org/abs/2604.12506) | 💻 [GitHub](https://github.com/Tencent/Unified_Audio_Schema)
+
+This repository provides our model checkpoints trained using **Unified Audio Schema**. For the complete codebase, please refer to the corresponding [GitHub repository](https://github.com/Tencent/Unified_Audio_Schema).
+
+## Model Details
+
+| Attribute | Value |
+|:----------|:------|
+| Input Modality | Text and audio |
+| Output Modality | Text and audio |
+| Base LLM | [Qwen2.5-7B](https://huggingface.co/Qwen/Qwen2.5-7B) |
+| Audio Encoder | AuT encoder |
+| Input Audio Representation Frame Rate | 12.5 Hz |
+| Output Audio Token Codebook Size | 8,192 |
+| Output Audio Token Frame Rate | 25 Hz |
+
+Notes:
+- The model supports interleaved text and audio input/output, enabling flexible multimodal interactions.
+- Speech waveform reconstruction for generated audio tokens relies on the [StableToken](https://huggingface.co/tencent/StableToken) decoder.
+
+## Quick Start
+
+### Installation
+
+```bash
+git clone --recursive https://github.com/Tencent/Unified_Audio_Schema.git
+cd Unified_Audio_Schema && pip install -r requirements.txt
+```
+
+### Download Checkpoints
+
+```bash
+# Model weights
+huggingface-cli download tencent/Unified_Audio_Schema --local-dir checkpoints/Unified_Audio_Schema
+
+# StableToken decoder (required for speech waveform reconstruction)
+huggingface-cli download tencent/StableToken --local-dir checkpoints/StableToken
+```
+
+## Inference
+
+```python
+import torch
+import torchaudio
+from src.model import UASAudio
+
+model = UASAudio(
+	model_path="checkpoints/Unified_Audio_Schema",
+	audio_decoder_path="checkpoints/StableToken/decoder",
+	device="cuda" if torch.cuda.is_available() else "cpu",
+)
+
+dialogue_system_prompt = (
+	"User will provide you with a speech instruction. Do it step by step. "
+	"First, think about the instruction and respond in a interleaved manner, "
+	"with 13 text token followed by 52 audio tokens."
+)
+
+messages = [
+	{"role": "system", "content": dialogue_system_prompt},
+	{
+		"role": "user",
+		"content": [
+			{"type": "audio", "audio": "assets/give_me_a_brief_introduction_to_the_great_wall.wav"},
+		],
+	},
+	{"role": "assistant", "content": None},
+]
+
+generation_config = {
+    "max_new_tokens": 4096,
+    "temperature": 0.7,
+    "repetition_penalty": 1.05,
+    "top_p": 0.9,
+    "do_sample": True
+}
+
+_, text, audio_tokens = model(messages, **generation_config)
+print(text)
+
+if len(audio_tokens) > 0:
+	audio_array, sampling_rate = model.tokens_to_audio(audio_tokens)
+	torchaudio.save("response.wav", audio_array, sampling_rate)
+```
+
+## Supported Scenarios
+
+Our model can be applied to a wide range of audio understanding and generation tasks, including:
+
+- Text-input conversation
+- Speech-input conversation
+- Automatic Speech Recognition (ASR)
+- Audio captioning
+- Text-to-Speech (TTS)
+
+For more runnable examples, please refer to [`example_usage.ipynb`](https://github.com/Tencent/Unified_Audio_Schema/blob/main/example_usage.ipynb) in the GitHub repository.
+
+## Evaluation Highlights
+
+UAS-Audio demonstrates strong performance on audio understanding, ASR, and TTS benchmarks.
+
+### Audio Understanding
+
+| **Model** | MMSU<br>(Percep.) | MMSU<br>(Reason.) | **MMSU<br>(Overall)** | MMAR<br>(Speech) | MMAR<br>(Sound) | MMAR<br>(Music) | **MMAR<br>(Overall)** | MMAU<br>(Speech) | MMAU<br>(Sound) | MMAU<br>(Music) | **MMAU<br>(Overall)** | **Avg.** |
+| :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
+| [Kimi-Audio](https://github.com/MoonshotAI/Kimi-Audio) | <u>44.8</u> | 75.7 | <u>59.8</u> | 58.5 | 49.7 | 33.0 | 48.0 | 62.2 | 75.7 | 66.8 | 68.2 | 58.7 |
+| [Qwen2.5-Omni](https://github.com/QwenLM/Qwen2.5-Omni) | 42.7 | **77.6** | 58.1 | 59.9 | **58.8** | 40.8 | 56.7 | **70.6** | <u>78.1</u> | 65.9 | <u>71.5</u> | <u>62.1</u> |
+| [Step-Audio2](https://github.com/stepfun-ai/Step-Audio2) | 42.9 | 73.2 | 57.6 | <u>61.2</u> | 54.6 | <u>42.2</u> | <u>56.8</u> | <u>68.2</u> | **79.3** | <u>68.4</u> | **72.7** | 61.9 |
+| **Ours** | **55.7** | <u>77.4</u> | **66.2** | **66.0** | **58.8** | **45.2** | **60.1** | 67.0 | 70.0 | **71.3** | 69.4 | **65.2** |
+
+### ASR & TTS
+
+| Model | ASR<br>(LS-clean) | ASR<br>(AISHELL-1) | TTS<br>(SeedTTS-en) | TTS<br>(SeedTTS-zh) |
+| :--- | :---: | :---: | :---: | :---: |
+| [Qwen2.5-Omni](https://github.com/QwenLM/Qwen2.5-Omni) | - | - | 2.3 | 1.4 |
+| [Step-Audio2](https://github.com/stepfun-ai/Step-Audio2) | 1.9 | 1.0 | 2.1 | 3.2 |
+| [MiMo-Audio](https://github.com/XiaomiMiMo/MiMo-Audio) | 3.8 | 1.8 | 5.4 | 2.0 |
+| **Ours** | 2.2 | 2.3 | 1.7 | 1.4 |
+
+## Citation
+
+If you find Unified Audio Schema or our model useful for your research, please cite:
+
+```bibtex
+@misc{zhang2026transcriptionunifiedaudioschema,
+	title={Beyond Transcription: Unified Audio Schema for Perception-Aware AudioLLMs}, 
+	author={Linhao Zhang and Yuhan Song and Aiwei Liu and Chuhan Wu and Sijun Zhang and Wei Jia and Yuan Liu and Houfeng Wang and Xiao Zhou},
+	year={2026},
+	eprint={2604.12506},
+	archivePrefix={arXiv},
+	primaryClass={cs.CL},
+	url={https://arxiv.org/abs/2604.12506},
+}
+
+@inproceedings{song2026stabletoken,
+	title={StableToken: A Noise-Robust Semantic Speech Tokenizer for Resilient Speech{LLM}s},
+	author={Yuhan Song and Linhao Zhang and Chuhan Wu and Aiwei Liu and Wei Jia and Houfeng Wang and Zhou Xiao},
+	booktitle={The Fourteenth International Conference on Learning Representations},
+	year={2026},
+	url={https://openreview.net/forum?id=17DNmdQ9aU}
+}
+```
+
+## License
+
+This project is licensed under the [License Term of Unified_Audio_Schema](LICENSE).

chat_template.jinja

@@ -0,0 +1,7 @@
+{% set audio_count = namespace(value=0) %}{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system
+You are a helpful assistant.<|im_end|>
+{% endif %}<|im_start|>{{ message['role'] }}
+{% if message['content'] is string %}{{ message['content'] }}<|im_end|>
+{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_bos|><|IMAGE|><|vision_eos|>{% elif content['type'] == 'audio' or 'audio' in content or 'audio_url' in content %}{% set audio_count.value = audio_count.value + 1 %}{% if add_audio_id %}Audio {{ audio_count.value }}: {% endif %}<|audio_bos|><|AUDIO|><|audio_eos|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_bos|><|VIDEO|><|vision_eos|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>
+{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant
+{% endif %}

config.json

@@ -0,0 +1,87 @@
+{
+  "architectures": [
+    "UASAudioForCausalLM"
+  ],
+  "audio_encoder_config": {
+    "activation_dropout": 0,
+    "activation_function": "gelu",
+    "attention_dropout": 0,
+    "conv_chunksize": 500,
+    "d_model": 1280,
+    "downsample_hidden_size": 480,
+    "dropout": 0,
+    "encoder_attention_heads": 20,
+    "encoder_ffn_dim": 5120,
+    "encoder_layers": 32,
+    "initializer_range": 0.02,
+    "max_source_positions": 1500,
+    "model_type": "uas_audio_encoder",
+    "n_window": 50,
+    "n_window_infer": 800,
+    "num_hidden_layers": 32,
+    "num_mel_bins": 128
+  },
+  "audio_token": 151646,
+  "auto_map": {
+    "AutoConfig": "configuration_uas_audio.UASAudioConfig",
+    "AutoModelForCausalLM": "modeling_uas_audio.UASAudioForCausalLM"
+  },
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "model_type": "uas_audio",
+  "pad_token_id": 151643,
+  "text_config": {
+    "architectures": [
+      "Qwen2ForCausalLM"
+    ],
+    "attention_dropout": 0.0,
+    "hidden_act": "silu",
+    "hidden_size": 3584,
+    "initializer_range": 0.02,
+    "intermediate_size": 18944,
+    "layer_types": [
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention"
+    ],
+    "max_position_embeddings": 16384,
+    "max_window_layers": 28,
+    "model_type": "qwen2",
+    "num_attention_heads": 28,
+    "num_hidden_layers": 28,
+    "num_key_value_heads": 4,
+    "rms_norm_eps": 1e-06,
+    "rope_scaling": null,
+    "rope_theta": 1000000.0,
+    "sliding_window": null,
+    "use_cache": true,
+    "use_sliding_window": false,
+    "vocab_size": 159864
+  },
+  "transformers_version": "4.57.3"
+}

configuration_uas_audio.py

@@ -0,0 +1,148 @@
+from typing import Optional, Union
+from transformers import Qwen2Config
+from transformers.configuration_utils import PretrainedConfig
+
+
+class UASAudioEncoderConfig(PretrainedConfig):
+    model_type = "uas_audio_encoder"
+    def __init__(
+        self,
+        num_mel_bins: Optional[int] = 128,
+        encoder_layers: Optional[int] = 32,
+        encoder_attention_heads: Optional[int] = 20,
+        encoder_ffn_dim: Optional[int] = 5120,
+        d_model: Optional[int] = 1280,
+        dropout: Optional[int] = 0,
+        attention_dropout: Optional[int] = 0,
+        activation_function: Optional[int] = "gelu",
+        activation_dropout: Optional[int] = 0,
+        initializer_range: Optional[int] = 0.02,
+        max_source_positions: Optional[int] = 1500,
+        n_window: Optional[int] = 50,
+        n_window_infer: Optional[int] = 800,
+        conv_chunksize: Optional[int] = 500,
+        downsample_hidden_size: Optional[int] = 480,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.num_mel_bins = num_mel_bins
+        self.d_model = d_model
+        self.encoder_layers = encoder_layers
+        self.encoder_attention_heads = encoder_attention_heads
+        self.encoder_ffn_dim = encoder_ffn_dim
+        self.dropout = dropout
+        self.attention_dropout = attention_dropout
+        self.activation_function = activation_function
+        self.activation_dropout = activation_dropout
+        self.num_hidden_layers = encoder_layers
+        self.initializer_range = initializer_range
+        self.max_source_positions = max_source_positions
+        self.n_window = n_window
+        self.n_window_infer = n_window_infer
+        self.conv_chunksize = conv_chunksize
+        self.downsample_hidden_size = downsample_hidden_size
+
+
+class UASAudioTextConfig(PretrainedConfig):
+    model_type = "uas_audio_text"
+
+    def __init__(
+        self,
+        vocab_size=64012,
+        hidden_size=4096,
+        intermediate_size=11008,
+        num_hidden_layers=48,
+        num_attention_heads=32,
+        num_attention_groups=4,
+        num_key_value_heads=4,
+        hidden_act="silu",
+        max_position_embeddings=8192,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        rope_theta=1000000.0,
+        rope_scaling=None,
+        eos_token_id=None,
+        **kwargs
+    ):
+
+        super().__init__(
+            **kwargs)
+
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_attention_groups = num_attention_groups
+        self.num_key_value_heads = num_key_value_heads
+        assert self.num_attention_groups == self.num_key_value_heads, \
+            "num_attention_groups must be equal to num_key_value_heads"
+        self.hidden_act = hidden_act
+        self.max_position_embeddings = max_position_embeddings
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self.eos_token_id = eos_token_id
+
+        self.text_config = Qwen2Config(
+            vocab_size=vocab_size,
+            hidden_size=hidden_size,
+            intermediate_size=intermediate_size,
+            num_hidden_layers=num_hidden_layers,
+            num_attention_heads=num_attention_heads,
+            num_key_value_heads=num_key_value_heads,
+            hidden_act=hidden_act,
+            max_position_embeddings=max_position_embeddings,
+            initializer_range=initializer_range,
+            rms_norm_eps=rms_norm_eps,
+            rope_theta=rope_theta,
+            rope_scaling=rope_scaling,
+            architectures=["Qwen2ForCausalLM"],
+            dtype=getattr(self, "dtype", "bfloat16"),
+        )
+
+
+class UASAudioConfig(PretrainedConfig):
+    model_type = "uas_audio"
+    architectures = ["UASAudioForCausalLM"]
+
+    def __init__(
+        self,
+        audio_encoder_config: Optional[Union[dict, UASAudioEncoderConfig]] = None,
+        text_config: Optional[Union[dict, UASAudioTextConfig]] = None,
+        **kwargs
+    ):
+        super().__init__(**kwargs)
+        if text_config is None:
+            text_config = UASAudioTextConfig().text_config
+        elif isinstance(text_config, dict):
+            text_config = UASAudioTextConfig(**text_config).text_config
+        self.text_config = text_config
+
+        if audio_encoder_config is None:
+            self.audio_encoder_config = UASAudioEncoderConfig()
+        elif isinstance(audio_encoder_config, dict):
+            self.audio_encoder_config = UASAudioEncoderConfig(**audio_encoder_config)
+        elif isinstance(audio_encoder_config, UASAudioEncoderConfig):
+            self.audio_encoder_config = audio_encoder_config
+
+
+class UASAudioEncoderOnlyConfig(PretrainedConfig):
+    model_type = "uas_audio_encoder_only"
+    architectures = ["UASAudioEncoderOnly"]
+
+    def __init__(
+        self,
+        audio_encoder_config: Optional[Union[dict, UASAudioEncoderConfig]] = None,
+        hidden_size: Optional[int] = 4096,  # LLM hidden size for adapter output
+        **kwargs
+    ):
+        super().__init__(**kwargs)
+        if audio_encoder_config is None:
+            self.audio_encoder_config = UASAudioEncoderConfig()
+        elif isinstance(audio_encoder_config, dict):
+            self.audio_encoder_config = UASAudioEncoderConfig(**audio_encoder_config)
+        elif isinstance(audio_encoder_config, UASAudioEncoderConfig):
+            self.audio_encoder_config = audio_encoder_config
+        self.hidden_size = hidden_size  # Output dimension of adapter

generation_config.json

@@ -0,0 +1,8 @@
+{
+  "_from_model_config": true,
+  "eos_token_id": [
+    151645
+  ],
+  "pad_token_id": 151643,
+  "transformers_version": "4.57.3"
+}

model-00001-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0347fb972948c92745717fe31d60e294a2160340903b4a8d01fa767b511ebbd
+size 4933571592

model-00002-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb3e561c05f13fac34d64d0ab62b1b4a2792ee99fc732ff6624de882d14476c3
+size 4932751496

model-00003-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a748a462fcc581ba2ccb54d7b801cf9f5b7f34ce35bc53a5572f7fc1d537a81
+size 4330865648

model-00004-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ca5e3b3619fd76d451760e2443ca13a70f13da273f099764e446d6b86190cbc
+size 2468174256

model.safetensors.index.json

@@ -0,0 +1,872 @@
+{
+  "metadata": {
+    "total_parameters": 8332630656,
+    "total_size": 16665261312
+  },
+  "weight_map": {
+    "adapter.audio_projector.0.bias": "model-00004-of-00004.safetensors",
+    "adapter.audio_projector.0.weight": "model-00004-of-00004.safetensors",
+    "adapter.audio_projector.2.bias": "model-00004-of-00004.safetensors",
+    "adapter.audio_projector.2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.conv2d1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.conv2d1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.conv2d2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.conv2d2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.conv2d3.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.conv2d3.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.conv_out.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.fc1.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.fc1.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.fc2.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.fc2.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.final_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.final_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.out_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.out_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn_layer_norm.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn_layer_norm.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.ln_post.bias": "model-00004-of-00004.safetensors",
+    "audio_encoder.ln_post.weight": "model-00004-of-00004.safetensors",
+    "llm.lm_head.weight": "model-00004-of-00004.safetensors",
+    "llm.model.embed_tokens.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.0.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.1.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.10.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.10.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.11.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.12.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.13.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.14.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.15.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.16.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.17.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.18.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.18.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.18.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.18.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.18.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.18.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.18.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.18.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.18.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.18.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.18.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.18.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.19.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.19.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.2.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.2.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.20.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.20.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.21.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.22.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.23.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.24.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.25.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.26.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "llm.model.layers.27.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "llm.model.layers.3.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.3.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.4.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.5.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.6.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.7.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.8.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.8.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.8.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.8.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.8.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.8.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.8.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.8.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.8.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.8.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.8.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "llm.model.layers.8.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "llm.model.layers.9.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "llm.model.layers.9.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "llm.model.norm.weight": "model-00003-of-00004.safetensors"
+  }
+}

modeling_uas_audio.py

@@ -0,0 +1,869 @@
+from collections.abc import Callable
+from typing import Optional
+import numpy as np
+import torch
+import torch.nn.functional as F
+from torch import Tensor, nn
+from transformers import PreTrainedModel, Qwen2ForCausalLM
+from transformers.activations import ACT2FN
+from transformers.generation import GenerationMixin
+from transformers.modeling_layers import GradientCheckpointingLayer
+from transformers.modeling_outputs import BaseModelOutput
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.utils import auto_docstring
+from .configuration_uas_audio import UASAudioConfig, UASAudioEncoderConfig, UASAudioEncoderOnlyConfig
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+def _get_feat_extract_output_lengths(input_lengths):
+    """
+    Computes the output length of the convolutional layers and the output length of the audio encoder
+    """
+
+    input_lengths_leave = input_lengths % 100
+    feat_lengths = (input_lengths_leave - 1) // 2 + 1
+    output_lengths = ((feat_lengths - 1) // 2 + 1 - 1) // 2 + 1 + (input_lengths // 100) * 13
+    return output_lengths
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs,
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+class SinusoidsPositionEmbedding(nn.Module):
+    def __init__(self, length, channels, max_timescale=10000):
+        super().__init__()
+        if channels % 2 != 0:
+            raise ValueError("SinusoidsPositionEmbedding needs even channels input")
+        log_timescale_increment = np.log(max_timescale) / (channels // 2 - 1)
+        inv_timescales = torch.exp(-log_timescale_increment * torch.arange(channels // 2).float())
+        scaled_time = torch.arange(length)[:, np.newaxis] * inv_timescales[np.newaxis, :]
+        self.register_buffer(
+            "positional_embedding",
+            torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], dim=1),
+            persistent=False,
+        )
+
+    def forward(self, seqlen: int):
+        return self.positional_embedding[:seqlen, :]
+
+
+class UASAudioAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config):
+        super().__init__()
+        self.embed_dim = config.d_model
+        self.num_heads = config.encoder_attention_heads
+        self.dropout = config.attention_dropout
+        self.head_dim = self.embed_dim // self.num_heads
+        self.num_key_value_groups = 1  # needed for eager attention
+        self.config = config
+
+        if (self.head_dim * self.num_heads) != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim}"
+                f" and `num_heads`: {self.num_heads})."
+            )
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = 0.0
+        self.is_decoder = False
+        self.is_causal = False
+        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=True)
+        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=True)
+        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=True)
+        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=True)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        cu_seqlens: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        **kwargs,
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
+        """Input shape: Batch x Time x Channel"""
+
+        seq_length, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states).reshape(seq_length, self.num_heads, -1)
+        key_states = self.k_proj(hidden_states).reshape(seq_length, self.num_heads, -1)
+        value_states = self.v_proj(hidden_states).reshape(seq_length, self.num_heads, -1)
+
+        query_states = query_states.transpose(0, 1).unsqueeze(0)
+        key_states = key_states.transpose(0, 1).unsqueeze(0)
+        value_states = value_states.transpose(0, 1).unsqueeze(0)
+        max_seqlen = (cu_seqlens[1:] - cu_seqlens[:-1]).max()
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, _ = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask=attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            cu_seq_lens_q=cu_seqlens,  # pass cu seq lens for FA2
+            cu_seq_lens_k=cu_seqlens,
+            max_length_q=max_seqlen,
+            max_length_k=max_seqlen,
+            is_causal=False,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(seq_length, -1).contiguous()
+        attn_output = self.out_proj(attn_output)
+
+        return attn_output
+
+
+class UASAudioEncoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: UASAudioEncoderConfig):
+        super().__init__()
+        self.embed_dim = config.d_model
+        self.self_attn = UASAudioAttention(config)
+        self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
+        self.dropout = config.dropout
+        self.activation_fn = ACT2FN[config.activation_function]
+        self.activation_dropout = config.activation_dropout
+        self.fc1 = nn.Linear(self.embed_dim, config.encoder_ffn_dim)
+        self.fc2 = nn.Linear(config.encoder_ffn_dim, self.embed_dim)
+        self.final_layer_norm = nn.LayerNorm(self.embed_dim)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        cu_seqlens: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        **kwargs,
+    ) -> torch.Tensor:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+        """
+        residual = hidden_states
+        hidden_states = self.self_attn_layer_norm(hidden_states)
+        hidden_states = self.self_attn(
+            hidden_states=hidden_states,
+            cu_seqlens=cu_seqlens,
+            attention_mask=attention_mask,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+        residual = hidden_states
+        hidden_states = self.final_layer_norm(hidden_states)
+        hidden_states = self.fc1(hidden_states)
+        hidden_states = self.activation_fn(hidden_states)
+        hidden_states = self.fc2(hidden_states)
+        hidden_states = residual + hidden_states
+
+        if hidden_states.dtype == torch.float16:
+            clamp_value = torch.finfo(hidden_states.dtype).max - 1000
+            hidden_states = torch.clamp(hidden_states, min=-clamp_value, max=clamp_value)
+
+        outputs = (hidden_states,)
+
+        return outputs
+
+
+class UASAudioEncoder(PreTrainedModel):
+    config: UASAudioEncoderConfig
+    main_input_name = "input_features"
+    input_modalities = "audio"
+    _no_split_modules = ["UASAudioEncoderLayer"]
+    _supports_sdpa = True
+
+    def __init__(self, config: UASAudioEncoderConfig):
+        super().__init__(config)
+        self.dropout = config.dropout
+
+        embed_dim = config.d_model
+        self.num_mel_bins = config.num_mel_bins
+        self.max_source_positions = config.max_source_positions
+        self.n_window = config.n_window
+        self.positional_embedding = SinusoidsPositionEmbedding(self.max_source_positions, embed_dim)
+        self.layers = nn.ModuleList([UASAudioEncoderLayer(config) for _ in range(config.encoder_layers)])
+        self.ln_post = nn.LayerNorm(config.d_model)
+        self.gradient_checkpointing = False
+        self.conv2d1 = nn.Conv2d(1, config.downsample_hidden_size, 3, 2, padding=1)
+        self.conv2d2 = nn.Conv2d(config.downsample_hidden_size, config.downsample_hidden_size, 3, 2, padding=1)
+        self.conv2d3 = nn.Conv2d(config.downsample_hidden_size, config.downsample_hidden_size, 3, 2, padding=1)
+        self.conv_out = nn.Linear(
+            config.downsample_hidden_size * ((((config.num_mel_bins + 1) // 2 + 1) // 2 + 1) // 2),
+            config.d_model,
+            bias=False,
+        )
+        self.n_window_infer = self.config.n_window_infer
+        self.conv_chunksize = self.config.conv_chunksize
+        self.post_init()
+
+    def _freeze_parameters(self):
+        for param in self.parameters():
+            param.requires_grad = False
+        self._requires_grad = False
+
+    def get_input_embeddings(self) -> nn.Module:
+        return self.conv1
+
+    def set_input_embeddings(self, value: nn.Module):
+        self.conv1 = value
+
+    def _prepare_attention_mask(self, inputs_tensor: torch.Tensor, cu_seqlens: torch.Tensor) -> torch.Tensor:
+        # Flash Attention 2 doesn't need a 4D mask and relies on `cu_seqlens/max_seqlen`
+        # NOTE: the created attention masl only approximates the ragged FA2 attention by
+        # allowing bidirectional attention within `cu_seqlens` blocks, and not attending between
+        # blocks. Though it will not be a 100% match for FA2's `varlen` path
+        if self.config._attn_implementation == "flash_attention_2":
+            return None
+
+        seq_length = inputs_tensor.shape[0]
+        attention_mask = torch.full(
+            [1, 1, seq_length, seq_length],
+            torch.finfo(inputs_tensor.dtype).min,
+            device=inputs_tensor.device,
+            dtype=inputs_tensor.dtype,
+        )
+        for i in range(1, len(cu_seqlens)):
+            attention_mask[..., cu_seqlens[i - 1] : cu_seqlens[i], cu_seqlens[i - 1] : cu_seqlens[i]] = 0
+        return attention_mask
+
+    @auto_docstring
+    def forward(
+        self,
+        input_features,
+        feature_lens=None,
+        aftercnn_lens=None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ):
+        r"""
+        feature_lens (`torch.LongTensor` of shape `(batch_size,)`):
+            mel length
+        aftercnn_lens (`torch.LongTensor` of shape `(batch_size,)`):
+            mel length after cnn
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+        """
+        return_dict = return_dict if return_dict is not None else getattr(self.config, 'use_return_dict', False)
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else getattr(self.config, 'output_hidden_states', False)
+        )
+
+        aftercnn_lens = _get_feat_extract_output_lengths(feature_lens)
+        chunk_num = torch.ceil(feature_lens / (self.n_window * 2)).long()
+
+        chunk_lengths = torch.tensor(
+            [self.n_window * 2] * chunk_num.sum(),
+            dtype=torch.long,
+            device=feature_lens.device,
+        )
+        tail_chunk_index = F.pad(chunk_num, (1, 0), value=-1).cumsum(0)[1:]
+        chunk_lengths[tail_chunk_index] = feature_lens % (self.n_window * 2)
+        chunk_lengths[chunk_lengths == 0] = self.n_window * 2
+
+        chunk_list = input_features.T.split(chunk_lengths.tolist(), dim=0)
+        padded_feature = nn.utils.rnn.pad_sequence(chunk_list, batch_first=True).transpose(1, 2)
+        feature_lens_after_cnn = _get_feat_extract_output_lengths(chunk_lengths)
+        padded_mask_after_cnn = nn.utils.rnn.pad_sequence(
+            [torch.ones(length, dtype=torch.bool, device=padded_feature.device) for length in feature_lens_after_cnn],
+            batch_first=True,
+        )
+        padded_feature = padded_feature.unsqueeze(1)
+        # Split to chunk to avoid OOM during convolution
+        padded_embeds = []
+        for chunk in padded_feature.split(self.conv_chunksize, dim=0):
+            padded_embed = F.gelu(self.conv2d1(chunk))
+            padded_embed = F.gelu(self.conv2d2(padded_embed))
+            padded_embed = F.gelu(self.conv2d3(padded_embed))
+            padded_embeds.append(padded_embed)
+        padded_embed = torch.cat(padded_embeds, dim=0)
+        b, c, f, t = padded_embed.size()
+        padded_embed = self.conv_out(padded_embed.permute(0, 3, 1, 2).contiguous().view(b, t, c * f))
+
+        positional_embedding = (
+            self.positional_embedding.positional_embedding[: padded_embed.shape[1], :]
+            .unsqueeze(0)
+            .to(padded_embed.dtype)
+        )
+        padded_embed = padded_embed + positional_embedding
+        hidden_states = padded_embed[padded_mask_after_cnn]
+        cu_chunk_lens = [0]
+        window_aftercnn = padded_mask_after_cnn.shape[-1] * (self.n_window_infer // (self.n_window * 2))
+        for cnn_len in aftercnn_lens:
+            cu_chunk_lens += [window_aftercnn] * (cnn_len // window_aftercnn)
+            remainder = cnn_len % window_aftercnn
+            if remainder != 0:
+                cu_chunk_lens += [remainder]
+        cu_seqlens = torch.tensor(cu_chunk_lens, device=aftercnn_lens.device).cumsum(-1, dtype=torch.int32)
+
+        all_hidden_states = () if output_hidden_states else None
+        if output_hidden_states:
+            all_hidden_states = (hidden_states,)
+
+        for layer_idx, encoder_layer in enumerate(self.layers):
+            layer_outputs = encoder_layer(
+                hidden_states,
+                cu_seqlens,
+            )
+
+            hidden_states = layer_outputs[0]
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+        hidden_states = self.ln_post(hidden_states)
+
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, all_hidden_states] if v is not None)
+
+        return BaseModelOutput(
+            last_hidden_state=hidden_states,
+            hidden_states=all_hidden_states,
+        )
+
+    def padded_and_mask_function(self, tensor_list, tensor_len, padding_value=0, padding_side="right"):
+        """
+        Pads a sequence of tensors to their maximum length on indicated `padding_side`.
+        Then prepares a mask so that pad tokens are not attended to.
+        """
+        max_len = tensor_len.max()
+        dim = tensor_list[0].shape[0]
+        padded_tensor = torch.full(
+            size=(len(tensor_list), dim, max_len),
+            fill_value=padding_value,
+            dtype=self.dtype,
+            device=tensor_list[0].device,
+        )
+
+        batch_mask = torch.zeros(
+            (len(tensor_len), max_len),
+            dtype=torch.long,
+            device=padded_tensor.device,
+        )
+        for i, length in enumerate(tensor_len):
+            batch_mask[i, :length] = 1
+            padded_tensor[i, :, :length] = tensor_list[i]
+
+        feature_lens_after_cnn = (tensor_len - 1) // 2 + 1
+        max_len_after_cnn = feature_lens_after_cnn.max()
+        batch_mask_after_cnn = torch.zeros(
+            (len(tensor_len), max_len_after_cnn),
+            dtype=torch.long,
+            device=padded_tensor.device,
+        )
+        for i, length in enumerate(feature_lens_after_cnn):
+            batch_mask_after_cnn[i, :length] = 1
+        return (
+            padded_tensor,
+            batch_mask.unsqueeze(1),
+            batch_mask_after_cnn.bool(),
+        )
+
+
+class Adapter(nn.Module):
+    def __init__(
+        self,
+        d_model: int,
+        n_embd: int,
+    ):
+        super().__init__()
+        self.audio_projector = torch.nn.Sequential(
+            torch.nn.Linear(d_model, n_embd),
+            torch.nn.GELU(),
+            torch.nn.Linear(n_embd, n_embd)
+        )
+
+    def forward(self, x: Tensor) -> Tensor:
+        x = self.audio_projector(x)
+        return x
+
+
+class UASAudioForCausalLM(PreTrainedModel, GenerationMixin):
+    config_class = UASAudioConfig
+    main_input_name = "input_ids"
+    supports_gradient_checkpointing = True
+    def __init__(self, config: UASAudioConfig):
+        super().__init__(config)
+        if isinstance(config.dtype, str):
+            dtype = getattr(torch, config.dtype)
+        else:
+            dtype = config.dtype
+        self.bf16 = dtype == torch.bfloat16
+
+        self.llm = Qwen2ForCausalLM(config.text_config)
+        self.audio_encoder = UASAudioEncoder(config.audio_encoder_config)
+
+        d_model = config.audio_encoder_config.d_model
+
+        self.adapter = Adapter(
+            d_model,
+            config.text_config.hidden_size,
+        )
+        self.audio_token = config.audio_token
+
+        if self.bf16:
+            self.audio_encoder = self.audio_encoder.bfloat16()
+            self.adapter = self.adapter.bfloat16()
+
+        self.post_init()
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        mels=None,
+        mel_masks=None,
+        past_key_values=None,
+        **kwargs
+    ):
+        # If past_key_values are provided, we are in the generation phase and should not process audio inputs again
+        if past_key_values is not None:
+            outputs = self.llm(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                past_key_values=past_key_values,
+                **kwargs
+            )
+        else:
+            # First get the text embeddings for the input_ids, then replace audio token positions with audio embeddings
+            hidden_states = self.embedding_with_audio_tokens(input_ids, mels, mel_masks)
+            outputs = self.llm(
+                inputs_embeds=hidden_states,
+                attention_mask=attention_mask,
+                past_key_values=None,
+                **kwargs
+            )
+        return outputs
+
+    def embedding_with_audio_tokens(
+        self,
+        input_ids,
+        mels,
+        mel_masks
+    ):
+        """
+        Get input embeddings for the LLM, replacing audio token positions with audio features from the audio encoder.
+        """
+        hidden_states = self.embeddings(input_ids)
+        if mels is None:
+            return hidden_states
+
+        audio_embeddings = self.audio_encoding(mels, mel_masks)     # data -> feature
+        audio_embeddings = self.adapter(audio_embeddings)
+        audio_mask = input_ids == self.audio_token
+        hidden_states[audio_mask] = audio_embeddings
+        return hidden_states
+
+    def audio_encoding(
+        self,
+        audio_features: torch.Tensor,
+        audio_features_mask: torch.Tensor,
+        output_hidden_states: bool = False
+    ):
+        """
+        Encode audio features into embeddings.
+
+        Args:
+            audio_features: Audio features tensor
+            audio_features_mask: Audio features mask
+            output_hidden_states: Whether to return hidden states from all encoder layers
+
+        Returns:
+            If output_hidden_states=False: audio_features_encoded tensor
+            If output_hidden_states=True: BaseModelOutput with last_hidden_state and hidden_states
+        """
+        feature_lens = audio_features_mask.sum(-1).long()  # [batch_size]
+        input_features = audio_features.permute(0, 2, 1)[audio_features_mask.bool()].permute(1, 0)
+
+        audio_encoder_outputs = self.audio_encoder(
+            input_features,
+            feature_lens=feature_lens,
+            output_hidden_states=output_hidden_states,
+            return_dict=output_hidden_states,  # Only return dict when we need hidden states
+        )
+
+        if output_hidden_states:
+            # When output_hidden_states=True, we get BaseModelOutput
+            return audio_encoder_outputs
+        else:
+            # When output_hidden_states=False, we get tuple (hidden_states, ...)
+            # Extract the first element (hidden_states tensor) for backward compatibility
+            if isinstance(audio_encoder_outputs, tuple):
+                return audio_encoder_outputs[0]
+            return audio_encoder_outputs
+
+    @property
+    def embeddings(self):
+        """Return the model's input embeddings - required for GenerationMixin"""
+        return self.llm.model.embed_tokens
+
+    def forward_with_detailed_outputs(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        mels=None,
+        mel_masks=None,
+        past_key_values=None,
+        output_hidden_states: bool = True,
+        **kwargs
+    ):
+        """
+        Forward pass that returns detailed outputs including:
+        - Audio encoder final output
+        - Audio features after projector (adapter)
+        - Text embedding features
+        - Hidden states from each layer (separated for audio and text)
+
+        Args:
+            input_ids: Input token ids
+            attention_mask: Attention mask
+            mels: Audio mel features
+            mel_masks: Audio mel masks
+            past_key_values: Past key values for generation
+            output_hidden_states: Whether to return hidden states from all layers
+            **kwargs: Additional arguments
+
+        Returns:
+            dict containing:
+                - audio_encoder_output: Final output from audio encoder
+                - audio_features_after_adapter: Audio features after projector/adapter
+                - text_embeddings: Text embedding features (before audio replacement)
+                - audio_encoder_hidden_states: Tuple of hidden states from each audio encoder layer
+                - llm_hidden_states: Tuple of hidden states from each LLM layer (mixed audio+text)
+                - llm_hidden_states_text_only: Tuple of text-only hidden states from each LLM layer
+                - llm_hidden_states_audio_only: Tuple of audio-only hidden states from each LLM layer
+                - llm_outputs: Full LLM outputs (CausalLMOutputWithPast)
+        """
+        # Get text embeddings (pure text, before audio replacement)
+        # Save original text embeddings for return (will have audio parts removed)
+        text_embeddings_pure = self.embeddings(input_ids)
+
+        # Process audio if provided
+        audio_encoder_output = None
+        audio_features_after_adapter = None
+        audio_encoder_hidden_states = None
+        audio_mask = None
+
+        # Create embeddings for LLM forward pass (may include audio features)
+        input_embeddings_for_llm = text_embeddings_pure.clone()
+
+        # Identify audio token positions (even if no audio is provided, audio tokens may exist in input_ids)
+        audio_mask = input_ids == self.audio_token
+
+        if mels is not None:
+            # Get audio encoder outputs with hidden states
+            audio_encoder_outputs = self.audio_encoding(
+                mels,
+                mel_masks,
+                output_hidden_states=output_hidden_states
+            )
+
+            if output_hidden_states:
+                audio_encoder_output = audio_encoder_outputs.last_hidden_state
+                audio_encoder_hidden_states = audio_encoder_outputs.hidden_states
+            else:
+                audio_encoder_output = audio_encoder_outputs
+                audio_encoder_hidden_states = None
+
+            # Apply adapter
+            audio_features_after_adapter = self.adapter(audio_encoder_output)
+
+            # Replace audio token positions with audio embeddings in the LLM input
+            input_embeddings_for_llm[audio_mask] = audio_features_after_adapter
+
+        # Remove audio parts from text_embeddings_pure (delete, not set to zero)
+        # This ensures returned text_embeddings strictly contains no audio features
+        if audio_mask.any():
+            # Process each batch separately since audio positions may differ
+            batch_size = text_embeddings_pure.shape[0]
+            text_embeddings_list = []
+
+            for i in range(batch_size):
+                # Get text-only mask for this batch (inverse of audio_mask)
+                text_mask = ~audio_mask[i]  # shape: (seq_len,)
+                # Extract only text embeddings
+                text_emb = text_embeddings_pure[i][text_mask]  # shape: (text_seq_len, hidden_size)
+                text_embeddings_list.append(text_emb)
+
+            # Pad sequences to the same length for batching
+            # Use the maximum text sequence length across batches
+            max_text_len = max(emb.shape[0] for emb in text_embeddings_list) if text_embeddings_list else 0
+            if max_text_len > 0:
+                hidden_size = text_embeddings_pure.shape[2]
+                text_embeddings_pure = torch.zeros(
+                    (batch_size, max_text_len, hidden_size),
+                    dtype=text_embeddings_pure.dtype,
+                    device=text_embeddings_pure.device
+                )
+                for i, emb in enumerate(text_embeddings_list):
+                    text_len = emb.shape[0]
+                    text_embeddings_pure[i, :text_len] = emb
+            else:
+                # No text embeddings (all are audio tokens)
+                hidden_size = text_embeddings_pure.shape[2]
+                text_embeddings_pure = torch.zeros(
+                    (batch_size, 0, hidden_size),
+                    dtype=text_embeddings_pure.dtype,
+                    device=text_embeddings_pure.device
+                )
+        # If no audio tokens, text_embeddings_pure remains unchanged
+
+        # Forward through LLM
+        if past_key_values is not None:
+            # Incremental decoding
+            llm_outputs = self.llm(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                past_key_values=past_key_values,
+                output_hidden_states=output_hidden_states,
+                return_dict=True,
+                **kwargs
+            )
+        else:
+            # First step: use combined embeddings (may include audio features)
+            llm_outputs = self.llm(
+                inputs_embeds=input_embeddings_for_llm,
+                attention_mask=attention_mask,
+                past_key_values=None,
+                output_hidden_states=output_hidden_states,
+                return_dict=True,
+                **kwargs
+            )
+
+        # Extract LLM hidden states
+        llm_hidden_states = llm_outputs.hidden_states if output_hidden_states else None
+
+        # Separate audio and text hidden states if audio is present
+        llm_hidden_states_text_only = None
+        llm_hidden_states_audio_only = None
+
+        if output_hidden_states and llm_hidden_states is not None and audio_mask is not None:
+            # Separate each layer's hidden states into text and audio parts
+            llm_hidden_states_text_only = tuple()
+            llm_hidden_states_audio_only = tuple()
+
+            batch_size = llm_hidden_states[0].shape[0]
+            hidden_size = llm_hidden_states[0].shape[2]
+
+            for layer_hidden_states in llm_hidden_states:
+                # layer_hidden_states shape: (batch_size, seq_len, hidden_size)
+                # audio_mask shape: (batch_size, seq_len)
+
+                # Process text-only hidden states: delete audio positions, not set to zero
+                text_hidden_list = []
+                audio_hidden_list = []
+
+                for i in range(batch_size):
+                    # Get text-only mask for this batch (inverse of audio_mask)
+                    text_mask = ~audio_mask[i]  # shape: (seq_len,)
+                    audio_mask_i = audio_mask[i]  # shape: (seq_len,)
+
+                    # Extract only text hidden states (delete audio positions)
+                    text_hidden_i = layer_hidden_states[i][text_mask]  # shape: (text_seq_len, hidden_size)
+                    text_hidden_list.append(text_hidden_i)
+
+                    # Extract only audio hidden states (delete text positions)
+                    audio_hidden_i = layer_hidden_states[i][audio_mask_i]  # shape: (audio_seq_len, hidden_size)
+                    audio_hidden_list.append(audio_hidden_i)
+
+                # Pad sequences to the same length for batching
+                # Use the maximum text sequence length across batches
+                max_text_len = max(emb.shape[0] for emb in text_hidden_list) if text_hidden_list else 0
+                max_audio_len = max(emb.shape[0] for emb in audio_hidden_list) if audio_hidden_list else 0
+
+                if max_text_len > 0:
+                    text_hidden = torch.zeros(
+                        (batch_size, max_text_len, hidden_size),
+                        dtype=layer_hidden_states.dtype,
+                        device=layer_hidden_states.device
+                    )
+                    for i, emb in enumerate(text_hidden_list):
+                        text_len = emb.shape[0]
+                        text_hidden[i, :text_len] = emb
+                else:
+                    # No text hidden states (all are audio tokens)
+                    text_hidden = torch.zeros(
+                        (batch_size, 0, hidden_size),
+                        dtype=layer_hidden_states.dtype,
+                        device=layer_hidden_states.device
+                    )
+
+                if max_audio_len > 0:
+                    audio_hidden = torch.zeros(
+                        (batch_size, max_audio_len, hidden_size),
+                        dtype=layer_hidden_states.dtype,
+                        device=layer_hidden_states.device
+                    )
+                    for i, emb in enumerate(audio_hidden_list):
+                        audio_len = emb.shape[0]
+                        audio_hidden[i, :audio_len] = emb
+                else:
+                    # No audio hidden states (all are text tokens)
+                    audio_hidden = torch.zeros(
+                        (batch_size, 0, hidden_size),
+                        dtype=layer_hidden_states.dtype,
+                        device=layer_hidden_states.device
+                    )
+
+                llm_hidden_states_text_only += (text_hidden,)
+                llm_hidden_states_audio_only += (audio_hidden,)
+
+        return {
+            "audio_encoder_output": audio_encoder_output,
+            "audio_features_after_adapter": audio_features_after_adapter,
+            "text_embeddings": text_embeddings_pure,  # Return text embeddings with audio parts removed
+            "audio_encoder_hidden_states": audio_encoder_hidden_states,
+            "llm_hidden_states": llm_hidden_states,
+            "llm_hidden_states_text_only": llm_hidden_states_text_only,
+            "llm_hidden_states_audio_only": llm_hidden_states_audio_only,
+            "llm_outputs": llm_outputs,
+        }
+
+    def generate(
+        self,
+        input_ids,
+        attention_mask=None,
+        mels=None,
+        mel_masks=None,
+        generation_config=None,
+        **generate_kwargs
+    ):
+        """
+        New implementation of the generate method to support audio inputs.
+
+        This method will:
+        1. Handle the initial processing of audio inputs;
+        2. Call the underlying LLM's generate method with the appropriate embeddings;
+        3. The incremental decoding will be handled by the LLM's generate method using past_key_values.
+        """
+        # Process audio inputs and get combined embeddings for the initial step
+        input_embeddings = self.embedding_with_audio_tokens(input_ids, mels, mel_masks)
+
+        # Call the underlying LLM's generate method with inputs_embeds instead of input_ids
+        # The LLM's generate method will handle the generation loop.
+        # During incremental decoding, it will use past_key_values to avoid re-processing audio inputs.
+        outputs = self.llm.generate(
+            inputs_embeds=input_embeddings,
+            attention_mask=attention_mask,
+            generation_config=generation_config,
+            use_cache=True,
+            **generate_kwargs
+        )
+
+        return outputs
+
+
+class UASAudioEncoderOnly(PreTrainedModel):
+    """
+    UASAudio encoder-only model that contains only the audio encoder and adapter.
+    Input: audio features
+    Output: features processed by encoder and adapter
+    """
+    config_class = UASAudioEncoderOnlyConfig
+    main_input_name = "input_features"
+    input_modalities = "audio"
+
+    def __init__(self, config: UASAudioEncoderOnlyConfig):
+        super().__init__(config)
+        if isinstance(config.dtype, str):
+            dtype = getattr(torch, config.dtype)
+        else:
+            dtype = getattr(config, "dtype", torch.bfloat16)
+        self.bf16 = dtype == torch.bfloat16
+
+        self.audio_encoder = UASAudioEncoder(config.audio_encoder_config)
+
+        d_model = config.audio_encoder_config.d_model
+
+        self.adapter = Adapter(
+            d_model,
+            config.hidden_size,
+        )
+
+        if self.bf16:
+            self.audio_encoder = self.audio_encoder.bfloat16()
+            self.adapter = self.adapter.bfloat16()
+
+        self.post_init()
+
+    def forward(
+        self,
+        input_features: torch.Tensor,
+        feature_lens: Optional[torch.Tensor] = None,
+        **kwargs
+    ):
+        """
+        Forward pass through audio encoder and adapter.
+
+        Args:
+            input_features: Audio features tensor of shape (seq_len, num_mel_bins) or (batch, seq_len, num_mel_bins)
+            feature_lens: Optional tensor of shape (batch_size,) indicating the length of each sequence
+            **kwargs: Additional arguments passed to audio encoder
+
+        Returns:
+            torch.Tensor: Features processed by encoder and adapter
+        """
+        # Encode audio features
+        audio_features_encoded = self.audio_encoder(
+            input_features,
+            feature_lens=feature_lens,
+            **kwargs
+        )
+
+        # Handle tuple output from encoder (backward compatibility)
+        if isinstance(audio_features_encoded, tuple):
+            audio_features_encoded = audio_features_encoded[0]
+        elif hasattr(audio_features_encoded, "last_hidden_state"):
+            audio_features_encoded = audio_features_encoded.last_hidden_state
+
+        # Apply adapter (projector)
+        output_features = self.adapter(audio_features_encoded)
+
+        return output_features

preprocessor_config.json

@@ -0,0 +1,15 @@
+{
+  "chunk_length": 30,
+  "dither": 0.0,
+  "feature_extractor_type": "WhisperFeatureExtractor",
+  "feature_size": 128,
+  "hop_length": 160,
+  "n_fft": 400,
+  "n_samples": 480000,
+  "nb_max_frames": 3000,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "processor_class": "UASAudioProcessor",
+  "return_attention_mask": false,
+  "sampling_rate": 16000
+}

special_tokens_map.json

@@ -0,0 +1,29 @@
+{
+  "additional_special_tokens": [
+    "<|begin_of_audio|>",
+    "<|end_of_audio|>",
+    "<|begin_of_transcription|>",
+    "<|end_of_transcription|>"
+  ],
+  "audio_bos_token": "<|audio_bos|>",
+  "audio_eos_token": "<|audio_eos|>",
+  "audio_token": "<|AUDIO|>",
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "image_token": "<|IMAGE|>",
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "video_token": "<|VIDEO|>",
+  "vision_bos_token": "<|vision_bos|>",
+  "vision_eos_token": "<|vision_eos|>"
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d975563e03e50528046a6dc124556b02c284b5f88974a515d9474383839168a
+size 14557594