Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,8 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+assets/general_audio_bar.png filter=lfs diff=lfs merge=lfs -text
+assets/moss-audio-2.png filter=lfs diff=lfs merge=lfs -text
+assets/moss-audio-image.png filter=lfs diff=lfs merge=lfs -text
+assets/moss-audio-logo.png filter=lfs diff=lfs merge=lfs -text
+assets/speech_caption_radar.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,3 +1,612 @@
----
-license: apache-2.0
----
+# MOSS-Audio
+
+
+<p align="center">
+  <img src="./assets/moss-audio-logo.png" width="55%" />
+</p>
+
+
+
+<div align="center">
+  <a href="https://huggingface.co/collections/OpenMOSS-Team/moss-audio"><img src="https://img.shields.io/badge/Huggingface-Models-orange?logo=huggingface&amp"></a>
+  <img src="https://img.shields.io/badge/Blog-Coming_Soon-blue?logo=internet-explorer&amp">
+  <img src="https://img.shields.io/badge/Arxiv-Coming_Soon-red?logo=Arxiv&amp">
+
+  <a href="https://x.com/Open_MOSS"><img src="https://img.shields.io/badge/Twitter-Follow-black?logo=x&amp"></a>
+  <a href="https://discord.gg/Xf3aXddCjc"><img src="https://img.shields.io/badge/Discord-Join-5865F2?logo=discord&amp"></a>
+  <a href="./assets/wechat.jpg"><img src="https://img.shields.io/badge/WeChat-Join-07C160?logo=wechat&amp;logoColor=white" alt="WeChat"></a>
+</div>
+
+<p align="center">
+  <a href="./README.md">English</a> | <a href="./README_zh.md">简体中文</a>
+</p>
+
+
+
+
+MOSS-Audio is an open-source **audio understanding model** from [MOSI.AI](https://mosi.cn/#hero), the [OpenMOSS team](https://www.open-moss.com/), and [Shanghai Innovation Institute](https://www.sii.edu.cn/). It performs unified modeling over complex real-world audio, supporting **speech understanding, environmental sound understanding, music understanding, audio captioning, time-aware QA, and complex reasoning**. In this release, we provide **four models**: **MOSS-Audio-4B-Instruct**, **MOSS-Audio-4B-Thinking**, **MOSS-Audio-8B-Instruct**, and **MOSS-Audio-8B-Thinking**. The Instruct variants are optimized for direct instruction following, while the Thinking variants provide stronger chain-of-thought reasoning capabilities.
+
+
+## News
+* 2026.4.13: 🎉🎉🎉 We have released [MOSS-Audio](https://huggingface.co/collections/OpenMOSS-Team/moss-audio). Blog and paper coming soon!
+
+
+## Contents
+
+- [Introduction](#introduction)
+- [Model Architecture](#model-architecture)
+  - [DeepStack Cross-Layer Feature Injection](#deepstack-cross-layer-feature-injection)
+  - [Time-Aware Representation](#time-aware-representation)
+- [Released Models](#released-models)
+- [Evaluation](#evaluation)
+- [Quickstart](#quickstart)
+  - [Environment Setup](#environment-setup)
+  - [Basic Usage](#basic-usage)
+  - [Gradio App](#gradio-app)
+  - [SGLang Serving](#sglang-serving)
+- [More Information](#more-information)
+- [Citation](#citation)
+
+
+## Introduction
+
+<p align="center">
+  <img src="./assets/moss-audio-image.png" width="95%" />
+</p>
+
+
+
+Understanding audio requires more than simply transcribing words — it demands the ability to perceive acoustic cues, recognize speakers and emotions, interpret environmental sounds, reason over temporal context, and handle complex multi-step inference. **MOSS-Audio** is built to unify these capabilities within a single model.
+
+- **Speech & Content Understanding**: Accurately recognizes and transcribes spoken content from audio inputs, producing clean and well-structured text outputs. Supports both word-level and sentence-level timestamp alignment.
+- **Speaker, Emotion & Event Analysis**: Identifies speaker characteristics, analyzes emotional states based on tone, timbre, and context, and detects key acoustic events within the audio.
+- **Scene & Sound Cue Extraction**: Extracts meaningful cues from background sounds, environmental noise, music, and non-speech signals to infer scene context and atmosphere.
+- **Music Understanding**: Analyzes musical style, emotional progression, instrumentation, and salient acoustic features in music segments.
+- **Audio Question Answering & Summarization**: Answers questions and generates summaries about speech, podcasts, meetings, interviews, and environmental recordings, helping users efficiently extract key information.
+- **Time-Aware QA**: Supports time-aware questions, including word-level and sentence-level timestamp ASR.
+- **Complex Reasoning**: Performs multi-hop reasoning over audio content, powered by chain-of-thought training and reinforcement learning.
+
+## Model Architecture
+
+<p align="center">
+  <img src="./assets/moss-audio-architecture.svg" width="95%" />
+</p>
+
+MOSS-Audio follows a modular design comprising three components: an audio encoder, a modality adapter, and a large language model. Raw audio is first encoded by **MOSS-Audio-Encoder** into continuous temporal representations at **12.5 Hz**, which are then projected into the language model's embedding space through the adapter and finally consumed by the LLM for auto-regressive text generation. 
+
+Rather than relying on off-the-shelf audio frontends, we train a dedicated encoder from scratch to obtain more robust speech representations, tighter temporal alignment, and better extensibility across acoustic domains.
+
+
+### DeepStack Cross-Layer Feature Injection
+
+Using only the encoder's top-layer features tends to lose low-level prosody, transient events, and local time-frequency structure. To address this, we design a **DeepStack**-inspired cross-layer injection module between the encoder and the language model: in addition to the encoder's final-layer output, features from earlier and intermediate layers are selected, independently projected, and injected into the language model's early layers, preserving multi-granularity information from low-level acoustic details to high-level semantic abstractions.
+
+This design is especially well-suited for audio understanding tasks, as it helps retain rhythm, timbre, transients, and background structure — information that a single high-level representation cannot fully capture.
+
+### Time-Aware Representation
+
+Time is a critical dimension in audio understanding. To enhance explicit temporal awareness, we adopt a **time-marker insertion** strategy during pretraining: explicit time tokens are inserted between audio frame representations at fixed time intervals to indicate temporal positions. This design enables the model to learn "what happened when" within a unified text generation framework, naturally supporting timestamp ASR, event localization, time-based QA, and long-audio retrospection.
+
+
+## Released Models
+
+
+| Model | Audio Encoder | LLM Backbone | Total Size | Hugging Face |
+|---|---|---|---:|---|
+| **MOSS-Audio-4B-Instruct** | MOSS-Audio-Encoder | Qwen3-4B | ~4.6B | [![Hugging Face](https://img.shields.io/badge/Huggingface-Model-orange?logo=huggingface)](https://huggingface.co/OpenMOSS-Team/MOSS-Audio-4B-Instruct) 
+| **MOSS-Audio-4B-Thinking** | MOSS-Audio-Encoder | Qwen3-4B | ~4.6B | [![Hugging Face](https://img.shields.io/badge/Huggingface-Model-orange?logo=huggingface)](https://huggingface.co/OpenMOSS-Team/MOSS-Audio-4B-Thinking) 
+| **MOSS-Audio-8B-Instruct** | MOSS-Audio-Encoder | Qwen3-8B | ~8.6B | [![Hugging Face](https://img.shields.io/badge/Huggingface-Model-orange?logo=huggingface)](https://huggingface.co/OpenMOSS-Team/MOSS-Audio-8B-Instruct) 
+| **MOSS-Audio-8B-Thinking** | MOSS-Audio-Encoder | Qwen3-8B | ~8.6B | [![Hugging Face](https://img.shields.io/badge/Huggingface-Model-orange?logo=huggingface)](https://huggingface.co/OpenMOSS-Team/MOSS-Audio-8B-Thinking) 
+
+> More model families, sizes, and variants will be released in the future. Stay tuned!
+
+
+## Evaluation
+
+We evaluate MOSS-Audio on a comprehensive set of audio understanding benchmarks. Key results:
+
+- **General Audio Understanding**: MOSS-Audio-8B-Thinking achieves an average accuracy of **70.80**, outperforming all of the open-source models.
+- **Speech Captioning**: MOSS-Audio-Instruct variants lead across **11 out of 13** fine-grained speech description dimensions, with **MOSS-Audio-8B-Instruct** achieving the best overall average score (**3.7252**).
+- **ASR**: On a diverse ASR benchmark suite spanning 12 evaluation dimensions, MOSS-Audio achieves the **lowest overall CER (11.30)**, with particular strength in health-condition, code-switching, dialect, singing, and non-speech scenarios.
+- **Timestamp ASR**: MOSS-Audio-8B-Instruct achieves **35.77 AAS** on AISHELL-1 and **131.61 AAS** on LibriSpeech, dramatically outperforming Qwen3-Omni (833.66) and Gemini-3.1-Pro (708.24) in timestamp asr accuracy.
+
+### General Audio Understanding (Accuracy↑)
+
+<p align="center">
+  <img src="./assets/general_audio_bar.svg" width="75%" />
+</p>
+
+<table>
+  <thead>
+    <tr>
+      <th>Model</th>
+      <th>Model Size</th>
+      <th>MMAU</th>
+      <th>MMAU-Pro</th>
+      <th>MMAR</th>
+      <th>MMSU</th>
+      <th>Avg</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr><td colspan="7"><em><strong>Open Source (small)</strong></em></td></tr>
+    <tr>
+      <td>Kimi-Audio</td><td>7B</td><td>72.41</td><td>56.58</td><td>60.82</td><td>54.74</td><td>61.14</td>
+    </tr>
+    <tr>
+      <td>Qwen2.5-Omni</td><td>7B</td><td>65.60</td><td>52.20</td><td>56.70</td><td>61.32</td><td>58.96</td>
+    </tr>
+    <tr>
+      <td>Audio Flamingo 3</td><td>7B</td><td>61.23</td><td>51.70</td><td>57.96</td><td>60.04</td><td>57.73</td>
+    </tr>
+    <tr>
+      <td>MiMo-Audio-7B</td><td>7B</td><td>74.90</td><td>53.35</td><td>61.70</td><td>61.94</td><td>62.97</td>
+    </tr>
+    <tr>
+      <td>MiniCPM-o-4.5</td><td>9B</td><td>70.97</td><td>39.65</td><td>55.75</td><td>60.96</td><td>56.83</td>
+    </tr>
+    <tr>
+      <td><strong>MOSS-Audio-4B-Instruct</strong></td><td><strong>4B</strong></td><td>75.79</td><td>58.16</td><td>59.68</td><td>59.68</td><td>64.04</td>
+    </tr>
+    <tr>
+      <td><strong>MOSS-Audio-4B-Thinking</strong></td><td><strong>4B</strong></td><td><strong>77.64</strong></td><td>60.75</td><td>63.91</td><td>71.20</td><td>68.37</td>
+    </tr>
+    <tr>
+      <td><strong>MOSS-Audio-8B-Instruct</strong></td><td><strong>8B</strong></td><td>77.03</td><td>57.48</td><td>64.42</td><td>66.36</td><td>66.32</td>
+    </tr>
+    <tr>
+      <td><strong>MOSS-Audio-8B-Thinking</strong></td><td><strong>8B</strong></td><td>77.13</td><td><strong>64.29</strong></td><td><strong>65.73</strong></td><td><strong>76.06</strong></td><td><strong>70.80</strong></td>
+    </tr>
+    <tr><td colspan="7"><em><strong>Open Source (large)</strong></em></td></tr>
+    <tr>
+      <td>Qwen3-Omni-30B-A3B-Instruct</td><td>30B</td><td>75.00</td><td><strong>61.22</strong></td><td>66.40</td><td>69.00</td><td>67.91</td>
+    </tr>
+    <tr>
+      <td>Step-Audio-R1.1</td><td>33B</td><td>72.18</td><td>60.80</td><td>68.75</td><td>64.18</td><td>66.48</td>
+    </tr>
+    <tr>
+      <td>Step-Audio-R1</td><td>33B</td><td><strong>78.67</strong></td><td>59.68</td><td><strong>69.15</strong></td><td><strong>75.18</strong></td><td><strong>70.67</strong></td>
+    </tr>
+    <tr><td colspan="7"><em><strong>Closed Source</strong></em></td></tr>
+    <tr>
+      <td>GPT4o-Audio</td><td>-</td><td>65.66</td><td>52.30</td><td>59.78</td><td>58.76</td><td>59.13</td>
+    </tr>
+    <tr>
+      <td>Gemini-3-Pro</td><td>-</td><td>80.15</td><td>68.28</td><td>81.73</td><td>81.28</td><td>77.86</td>
+    </tr>
+    <tr>
+      <td>Gemini-3.1-Pro</td><td>-</td><td><strong>81.10</strong></td><td><strong>73.47</strong></td><td><strong>83.70</strong></td><td><strong>81.30</strong></td><td><strong>79.89</strong></td>
+    </tr>
+  </tbody>
+</table>
+
+### Speech Captioning (LLM-as-a-Judge Score↑)
+
+<p align="center">
+  <img src="./assets/speech_caption_radar.png" width="70%" />
+</p>
+
+<details>
+<summary><strong>Speech Captioning (click to expand)</strong></summary>
+
+
+| Model | gender | age | accent | pitch | volume | speed | texture | clarity | fluency | emotion | tone | personality | summary | Avg |
+|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| Qwen3-Omni-30B-A3B-Instruct | 4.436 | 3.936 | 4.356 | 3.590 | 3.682 | 3.614 | 3.093 | 3.521 | 3.531 | 3.328 | 3.224 | 3.292 | 3.179 | 3.5986 |
+| Qwen3-Omni-30B-A3B-Thinking | 4.419 | **4.026** | 4.327 | 3.610 | 3.577 | 3.610 | 3.179 | 3.403 | 3.526 | 3.232 | 3.154 | 3.197 | 3.107 | 3.5667 |
+| Gemini-3-Pro | 4.191 | 3.835 | 4.181 | 3.392 | 3.254 | 3.320 | 2.998 | 3.347 | 3.524 | 3.055 | 2.997 | 3.023 | 2.775 | 3.3763 |
+| Gemini-3.1-Pro| 4.436 | 3.936 | 4.356 | 3.590 | 3.682 | 3.614 | 3.093 | 3.521 | 3.531 | **3.328** | 3.224 | 3.292 | 3.179 | 3.5986 |
+| MOSS-Audio-4B-Instruct | **4.697** | 3.980 | 4.497 | 3.628 | **3.722** | 3.564 | **3.407** | 3.841 | 3.744 | 3.311 | **3.282** | **3.305** | 3.259 | 3.7105 |
+| MOSS-Audio-8B-Instruct | 4.683 | 3.979 | **4.572** | **3.682** | 3.709 | **3.638** | 3.403 | **3.869** | **3.747** | 3.314 | 3.253 | 3.272 | **3.307** | **3.7252** |
+
+</details>
+
+### ASR 
+
+| Model | Overall | Health Condition | Dialect | Singing | Non-Speech Vocalizations | Code-Switching | Acoustic Environment (Clean) | Acoustic Environment (Noisy) | Acoustic Characteristics: Whisper | Acoustic Characteristics: Far-Field / Near-Field | Multi-Speaker | Age | Semantic Content |
+|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| Paraformer-Large | 15.77 | 22.18 | 43.45 | 32.34 | 4.95 | 12.65 | 3.11 | 4.67 | 5.02 | 17.46 | 20.33 | 14.96 | 7.14 |
+| GLM-ASR-Nano | 17.29 | 24.49 | 22.39 | 51.95 | 4.65 | 11.88 | 3.68 | 5.02 | 4.94 | 27.51 | 28.02 | 17.19 | 7.32 |
+| Fun-ASR-Nano | 12.04 | 21.99 | 7.80 | 19.35 | 4.76 | 11.23 | 2.98 | 3.46 | 3.78 | 18.38 | 19.82 | **14.95** | 6.08 |
+| SenseVoice-Small | 14.50 | 24.04 | 8.89 | 23.79 | 4.92 | 13.90 | 4.13 | 4.93 | 5.57 | 26.66 | 24.06 | 17.63 | 7.55 |
+| Kimi-Audio-7B-Instruct | 14.12 | 21.11 | 29.34 | 21.76 | 4.68 | 16.38 | **2.20** | **2.15** | 2.66 | 21.02 | 20.61 | 16.74 | 6.12 |
+| Qwen2.5-Omni-3B | 15.26 | 24.65 | 33.87 | 24.24 | 5.54 | 11.66 | 2.76 | 3.56 | 4.32 | 22.15 | 22.91 | 15.17 | 7.24 |
+| Qwen2.5-Omni-7B | 15.05 | 23.85 | 31.91 | 22.69 | 4.56 | 12.97 | 2.52 | 3.16 | 3.64 | 25.38 | 21.01 | 16.13 | 6.78 |
+| Qwen3-Omni-30B-A3B-Instruct | 11.39 | 20.73 | 15.63 | 16.01 | 4.73 | 11.30 | 2.23 | 2.47 | **1.90** | **17.08** | **18.15** | **11.46** | **5.74** |
+| **MOSS-Audio-4B-Instruct** | 11.58 | 21.11 | 11.84 | 10.79 | **4.01** | **10.11** | 3.11 | 3.72 | 3.29 | 18.48 | 20.33 | 15.09 | 8.15 |
+| **MOSS-Audio-8B-Instruct** | **11.30** | **19.18** | **8.76** | **9.81** | 4.31 | 10.18 | 2.70 | 3.20 | 2.75 | 24.04 | 24.36 | 15.26 | 7.69 |
+
+<details>
+<summary><strong>Detailed ASR Results (click to expand)</strong></summary>
+
+<table>
+  <tr>
+    <th rowspan="2">Model</th>
+    <th colspan="3">Acoustic Environment (Clean)</th>
+    <th colspan="1">Acoustic Environment (Noisy)</th>
+    <th colspan="1">Acoustic Characteristics: Whisper</th>
+    <th colspan="1">Acoustic Characteristics: Far-Field / Near-Field</th>
+    <th colspan="1">Multi-Speaker</th>
+    <th colspan="2">Age</th>
+    <th colspan="2">Health Condition</th>
+    <th colspan="2">Semantic Content</th>
+    <th colspan="3">Code-Switching</th>
+    <th colspan="2">Dialect</th>
+    <th colspan="2">Singing</th>
+    <th colspan="1">Non-Speech Vocalizations</th>
+  </tr>
+  <tr>
+    <th>AISHELL-1<br><em>test</em></th>
+    <th>AISHELL-2<br><em>Android | IOS | Mic</em></th>
+    <th>THCHS-30<br><em>test</em></th>
+    <th>MAGICDATA-READ<br><em>test</em></th>
+    <th>AISHELL6-Whisper<br><em>normal | whisper</em></th>
+    <th>AliMeeting<br><em>Test_Ali_far | Test_Ali_near</em></th>
+    <th>AISHELL-4<br><em>test</em></th>
+    <th>SeniorTalk<br><em>sentence</em></th>
+    <th>ChildMandarin<br><em>test</em></th>
+    <th>AISHELL-6A<br><em>mild | moderate | severe | StutteringSpeech</em></th>
+    <th>AISHELL_6B<br><em>LRDWWS | Uncontrol</em></th>
+    <th>WenetSpeech<br><em>test-meeting</em></th>
+    <th>Fleurs<br><em>cmn_hans_cn</em></th>
+    <th>CS-Dialogue<br><em>test</em></th>
+    <th>TALCS<br><em>test</em></th>
+    <th>ASCEND<br><em>test</em></th>
+    <th>KeSpeech<br><em>test</em></th>
+    <th>WSYue-ASR-eval<br><em>short</em></th>
+    <th>MIR-1K<br><em>test</em></th>
+    <th>openc-pop<br><em>test</em></th>
+    <th>MNV_17</th>
+  </tr>
+  <tr>
+    <td>Paraformer-Large</td>
+    <td>1.98</td>
+    <td>3.28 | 3.21 | 3.00</td>
+    <td>4.07</td>
+    <td>4.67</td>
+    <td>1.11 | 8.92</td>
+    <td><strong>25.64</strong> | 9.27</td>
+    <td>20.33</td>
+    <td>17.31</td>
+    <td>12.60</td>
+    <td>6.98 | 9.30 | 13.34 | 10.74</td>
+    <td>47.59 | 45.08</td>
+    <td>7.88</td>
+    <td>6.40</td>
+    <td>10.64</td>
+    <td>10.77</td>
+    <td>16.55</td>
+    <td>11.48</td>
+    <td>75.42</td>
+    <td>57.70</td>
+    <td>6.98</td>
+    <td>4.95</td>
+  </tr>
+  <tr>
+    <td>GLM-ASR-Nano</td>
+    <td>2.89</td>
+    <td>3.75 | 3.73 | 3.78</td>
+    <td>4.23</td>
+    <td>5.02</td>
+    <td>0.83 | 9.06</td>
+    <td>40.27 | 14.76</td>
+    <td>28.02</td>
+    <td>20.33</td>
+    <td>14.06</td>
+    <td>8.74 | 12.11 | 14.38 | 12.29</td>
+    <td>50.34 | 49.09</td>
+    <td>9.70</td>
+    <td>4.94</td>
+    <td>11.06</td>
+    <td>11.07</td>
+    <td>13.50</td>
+    <td>9.72</td>
+    <td>35.07</td>
+    <td>95.87</td>
+    <td>8.03</td>
+    <td>4.65</td>
+  </tr>
+  <tr>
+    <td>Fun-ASR-Nano</td>
+    <td>2.16</td>
+    <td>3.04 | 2.99 | 3.07</td>
+    <td>3.65</td>
+    <td>3.46</td>
+    <td>0.81 | 6.76</td>
+    <td>27.21 | 9.55</td>
+    <td>19.82</td>
+    <td>16.96</td>
+    <td>12.94</td>
+    <td>6.60 | <strong>8.81</strong> | 12.98 | 10.30</td>
+    <td>47.42 | 45.84</td>
+    <td>7.39</td>
+    <td><strong>4.76</strong></td>
+    <td>10.47</td>
+    <td><strong>8.09</strong></td>
+    <td>15.13</td>
+    <td>7.43</td>
+    <td>8.17</td>
+    <td>35.85</td>
+    <td>2.84</td>
+    <td>4.76</td>
+  </tr>
+  <tr>
+    <td>SenseVoice-Small</td>
+    <td>3.23</td>
+    <td>4.16 | 4.02 | 3.96</td>
+    <td>5.26</td>
+    <td>4.93</td>
+    <td>1.25 | 9.88</td>
+    <td>37.01 | 16.31</td>
+    <td>24.06</td>
+    <td>21.07</td>
+    <td>14.18</td>
+    <td>7.62 | 9.85 | 14.39 | 11.47</td>
+    <td>52.92 | 47.97</td>
+    <td>8.35</td>
+    <td>6.75</td>
+    <td>12.81</td>
+    <td>10.52</td>
+    <td>18.38</td>
+    <td>10.45</td>
+    <td><strong>7.34</strong></td>
+    <td>39.51</td>
+    <td>8.07</td>
+    <td>4.92</td>
+  </tr>
+  <tr>
+    <td>Kimi-Audio-7B-Instruct</td>
+    <td><strong>0.79</strong></td>
+    <td>2.91 | 3.03 | 2.88</td>
+    <td><strong>1.39</strong></td>
+    <td><strong>2.15</strong></td>
+    <td>0.69 | 4.63</td>
+    <td>28.22 | 13.82</td>
+    <td>20.61</td>
+    <td>19.70</td>
+    <td>13.79</td>
+    <td>7.00 | 9.34 | 12.56 | 10.75</td>
+    <td>44.44 | 42.57</td>
+    <td>7.15</td>
+    <td>5.10</td>
+    <td>14.56</td>
+    <td>12.74</td>
+    <td>21.83</td>
+    <td><strong>5.51</strong></td>
+    <td>53.17</td>
+    <td>38.35</td>
+    <td>5.17</td>
+    <td>4.68</td>
+  </tr>
+  <tr>
+    <td>Qwen2.5-Omni-3B</td>
+    <td>1.51</td>
+    <td>3.10 | 2.94 | 2.93</td>
+    <td>3.32</td>
+    <td>3.56</td>
+    <td>0.82 | 7.82</td>
+    <td>32.14 | 12.16</td>
+    <td>22.91</td>
+    <td>17.38</td>
+    <td>12.96</td>
+    <td>6.87 | 10.55 | 14.57 | 11.33</td>
+    <td>54.54 | 50.03</td>
+    <td>9.04</td>
+    <td>5.45</td>
+    <td>10.78</td>
+    <td>10.94</td>
+    <td>13.25</td>
+    <td>7.67</td>
+    <td>60.06</td>
+    <td>45.00</td>
+    <td>3.47</td>
+    <td>5.54</td>
+  </tr>
+  <tr>
+    <td>Qwen2.5-Omni-7B</td>
+    <td>1.16</td>
+    <td>2.88 | 2.77 | 2.73</td>
+    <td>3.06</td>
+    <td>3.16</td>
+    <td>0.71 | 6.57</td>
+    <td>32.03 | 18.73</td>
+    <td>21.01</td>
+    <td>19.96</td>
+    <td>12.29</td>
+    <td>7.27 | 10.94 | 12.92 | 10.53</td>
+    <td>51.99 | 49.45</td>
+    <td>8.43</td>
+    <td>5.13</td>
+    <td>14.02</td>
+    <td>10.46</td>
+    <td>14.42</td>
+    <td>6.40</td>
+    <td>57.43</td>
+    <td>42.62</td>
+    <td>2.75</td>
+    <td>4.56</td>
+  </tr>
+  <tr>
+    <td>Qwen3-Omni-30B-A3B-Instruct</td>
+    <td>0.95</td>
+    <td><strong>2.70</strong> | <strong>2.72</strong> | <strong>2.57</strong></td>
+    <td>2.21</td>
+    <td>2.47</td>
+    <td><strong>0.59</strong> | <strong>3.22</strong></td>
+    <td>25.72 | <strong>8.44</strong></td>
+    <td><strong>18.15</strong></td>
+    <td><strong>14.13</strong></td>
+    <td><strong>8.79</strong></td>
+    <td>6.20 | 8.88 | 11.59 | 10.25</td>
+    <td>45.80 | 41.65</td>
+    <td><strong>6.64</strong></td>
+    <td>4.84</td>
+    <td>12.94</td>
+    <td>8.33</td>
+    <td><strong>12.64</strong></td>
+    <td>5.87</td>
+    <td>25.39</td>
+    <td>30.81</td>
+    <td><strong>1.21</strong></td>
+    <td>4.73</td>
+  </tr>
+  <tr>
+    <td><strong>MOSS-Audio-4B-Instruct</strong></td>
+    <td>2.26</td>
+    <td>3.22 | 3.20 | 3.33</td>
+    <td>3.53</td>
+    <td>3.72</td>
+    <td>0.73 | 5.86</td>
+    <td>27.27 | 9.68</td>
+    <td>20.33</td>
+    <td>16.93</td>
+    <td>13.25</td>
+    <td>6.36 | 9.77 | 12.68 | 10.28</td>
+    <td>43.35 | 44.25</td>
+    <td>8.17</td>
+    <td>8.13</td>
+    <td>9.14</td>
+    <td>8.37</td>
+    <td>12.83</td>
+    <td>14.65</td>
+    <td>9.04</td>
+    <td>18.47</td>
+    <td>3.10</td>
+    <td><strong>4.01</strong></td>
+  </tr>
+  <tr>
+    <td><strong>MOSS-Audio-8B-Instruct</strong></td>
+    <td>1.82</td>
+    <td>2.97 | 2.95 | 2.91</td>
+    <td>2.82</td>
+    <td>3.20</td>
+    <td>0.69 | 4.80</td>
+    <td>36.82 | 11.25</td>
+    <td>24.36</td>
+    <td>17.42</td>
+    <td>13.10</td>
+    <td><strong>5.84</strong> | 8.94 | <strong>11.52</strong> | <strong>9.72</strong></td>
+    <td><strong>39.76</strong> | <strong>39.27</strong></td>
+    <td>7.86</td>
+    <td>7.52</td>
+    <td><strong>9.07</strong></td>
+    <td>8.22</td>
+    <td>13.26</td>
+    <td>9.18</td>
+    <td>8.33</td>
+    <td><strong>17.24</strong></td>
+    <td>2.39</td>
+    <td>4.31</td>
+  </tr>
+</table>
+
+</details>
+
+
+### Timestamp ASR (AAS↓)
+
+| Model | AISHELL-1(zh)  | LibriSpeech(en) |
+|---|---:|---:|
+| Qwen3-Omni-30B-A3B-Instruct | 833.66 | 646.95 |
+| Gemini-3.1-Pro| 708.24 | 871.19 |
+| MOSS-Audio-4B-Instruct | 76.96 | 358.13 |
+| **MOSS-Audio-8B-Instruct** | **35.77** | **131.61** |
+
+
+## Quickstart
+
+### Environment Setup
+
+We recommend Python 3.12 with a clean Conda environment. The commands below are enough for local inference.
+
+#### Recommended setup
+
+```bash
+git clone https://github.com/OpenMOSS/MOSS-Audio.git
+cd MOSS-Audio
+
+conda create -n moss-audio python=3.12 -y
+conda activate moss-audio
+
+conda install -c conda-forge "ffmpeg=7" -y
+pip install --extra-index-url https://download.pytorch.org/whl/cu128 -e ".[torch-runtime]"
+```
+
+#### Optional: FlashAttention 2
+
+If your GPU supports FlashAttention 2, you can replace the last install command with:
+
+```bash
+pip install --extra-index-url https://download.pytorch.org/whl/cu128 -e ".[torch-runtime,flash-attn]"
+```
+
+
+### Basic Usage
+
+Download the model first:
+
+```bash
+huggingface-cli download OpenMOSS-Team/MOSS-Audio --local-dir ./weights/MOSS-Audio
+huggingface-cli download OpenMOSS-Team/MOSS-Audio-Instruct --local-dir ./weights/MOSS-Audio-Instruct
+```
+
+Then edit `MODEL_PATH` / `AUDIO_PATH` in `infer.py` as needed, and run:
+
+```bash
+python infer.py
+```
+
+The default prompt in `infer.py` is `Describe this audio.` You can directly edit that line if you want to try transcription, audio QA, or speech captioning.
+
+### Gradio App
+
+Start the Gradio demo with:
+
+```bash
+python app.py
+```
+
+
+
+### SGLang Serving
+
+If you want to serve MOSS-Audio with SGLang, see the full guide in `moss_audio_usage_guide.md`.
+
+The shortest setup is:
+
+```bash
+git clone -b moss-audio https://github.com/OpenMOSS/sglang.git
+cd sglang
+pip install -e "python[all]"
+pip install nvidia-cudnn-cu12==9.16.0.29
+cd ..
+sglang serve --model-path ./weights/MOSS-Audio --trust-remote-code
+```
+
+If you use the default `torch==2.9.1+cu128` runtime, installing `nvidia-cudnn-cu12==9.16.0.29` is recommended before starting `sglang serve`.
+
+
+<a id="more-information"></a>
+
+## More Information
+- **MOSI.AI**: [https://mosi.cn](https://mosi.cn)
+- **OpenMOSS**: [https://www.open-moss.com](https://www.open-moss.com)
+
+
+## LICENSE
+
+Models in MOSS-Audio are licensed under the Apache License 2.0.
+
+
+## Citation
+
+```bibtex
+@misc{mossaudio2026,
+      title={MOSS-Audio Technical Report},
+      author={OpenMOSS Team},
+      year={2026},
+      howpublished={\url{https://github.com/OpenMOSS/MOSS-Audio}},
+      note={GitHub repository}
+}
+```
+

added_tokens.json

@@ -0,0 +1,32 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|assistant|>": 151671,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|eot|>": 151672,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|system|>": 151669,
+  "<|user|>": 151670,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

chat_template.jinja

@@ -0,0 +1,85 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set content = message.content %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in message.content %}
+                {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
+                {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {%- if loop.last or (not loop.last and reasoning_content) %}
+                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+            {%- else %}
+                {{- '<|im_start|>' + message.role + '\n' + content }}
+            {%- endif %}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

config.json

@@ -0,0 +1,109 @@
+{
+  "adapter_hidden_size": 8192,
+  "architectures": [
+    "MossAudioModel"
+  ],
+  "audio_config": {
+    "_attn_implementation": "eager",
+    "activation_dropout": 0.0,
+    "activation_function": "gelu",
+    "attention_dropout": 0.1,
+    "d_model": 1280,
+    "deepstack_encoder_layer_indexes": [
+      8,
+      16,
+      24
+    ],
+    "downsample_hidden_size": 480,
+    "downsample_rate": 8,
+    "dropout": 0.1,
+    "encoder_attention_heads": 20,
+    "encoder_attention_window_size": 100,
+    "encoder_ffn_dim": 5120,
+    "encoder_layers": 32,
+    "layer_norm_eps": 1e-05,
+    "max_source_positions": 1500,
+    "num_mel_bins": 128,
+    "output_dim": 1280,
+    "pretrained_path": ""
+  },
+  "auto_map": {
+    "AutoConfig": "configuration_moss_audio.MossAudioConfig",
+    "AutoProcessor": "processing_moss_audio.MossAudioProcessor"
+  },
+  "bos_token_id": 151643,
+  "deepstack_num_inject_layers": 3,
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "ignore_index": -100,
+  "language_config": {
+    "architectures": [
+      "Qwen3ForCausalLM"
+    ],
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "bos_token_id": 151643,
+    "eos_token_id": 151645,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 4096,
+    "initializer_range": 0.02,
+    "intermediate_size": 12288,
+    "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",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention"
+    ],
+    "max_position_embeddings": 40960,
+    "max_window_layers": 36,
+    "model_type": "qwen3",
+    "num_attention_heads": 32,
+    "num_hidden_layers": 36,
+    "num_key_value_heads": 8,
+    "rms_norm_eps": 1e-06,
+    "rope_scaling": null,
+    "rope_theta": 1000000,
+    "sliding_window": null,
+    "use_cache": true,
+    "use_sliding_window": false,
+    "vocab_size": 151936
+  },
+  "model_type": "moss_audio",
+  "num_hidden_layers": 36,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.57.1",
+  "vocab_size": 151936
+}

configuration_moss_audio.py

@@ -0,0 +1,31 @@
+from transformers import PretrainedConfig, Qwen3Config
+
+
+class MossAudioConfig(PretrainedConfig):
+    model_type = "moss_audio"
+    is_composition = True
+
+    def __init__(
+        self,
+        audio_config=None,
+        language_config=None,
+        adapter_hidden_size=8192,
+        ignore_index=-100,
+        deepstack_num_inject_layers=None,
+        **kwargs,
+    ):
+        if isinstance(language_config, dict):
+            language_config = Qwen3Config(**language_config)
+        elif language_config is None:
+            language_config = Qwen3Config()
+
+        self.audio_config = audio_config
+        self.language_config = language_config
+        self.adapter_hidden_size = adapter_hidden_size
+        self.ignore_index = ignore_index
+        self.deepstack_num_inject_layers = deepstack_num_inject_layers
+
+        for key in ("num_hidden_layers", "eos_token_id", "bos_token_id", "vocab_size"):
+            kwargs.setdefault(key, getattr(language_config, key, None))
+
+        super().__init__(**kwargs)

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "transformers_version": "4.57.1"
+}

model-00001-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5444e22367fe81bc56f5fff3693e6e788984452325cf90f645724d7f85fe2ee6
+size 4931218512

model-00002-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6dd174db4db22bc72a4e7aeb99bb6a3de5d8af6c53036a1ce215c621eb8252b
+size 4983069720

model-00003-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a49edfcecb911b9b65c5f57b0596a935e867a56358587c39b612705cf90ec5b7
+size 4999847608

model-00004-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9078bd9267da1452ece38bc7ea4400a34b69a9a3fa6f467e79390acc096855ba
+size 3190899536

model.safetensors.index.json

@@ -0,0 +1,909 @@
+{
+  "metadata": {
+    "total_parameters": 9052463488,
+    "total_size": 18104926976
+  },
+  "weight_map": {
+    "audio_adapter.down_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_adapter.gate_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_adapter.up_proj.weight": "model-00004-of-00004.safetensors",
+    "audio_encoder.conv1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.conv1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.conv2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.conv2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.conv3.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.conv3.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.0.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.1.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.10.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.11.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.12.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.13.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.14.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.15.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.16.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.17.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.18.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.19.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.2.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.20.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.21.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.22.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.23.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.24.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.25.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.26.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.27.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.28.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.29.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.3.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.30.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.31.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.4.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.5.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.6.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.7.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.8.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.fc1.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.fc1.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.fc2.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.fc2.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.final_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.final_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.out_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.out_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn_layer_norm.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.layers.9.self_attn_layer_norm.weight": "model-00001-of-00004.safetensors",
+    "audio_encoder.stem_proj.bias": "model-00001-of-00004.safetensors",
+    "audio_encoder.stem_proj.weight": "model-00001-of-00004.safetensors",
+    "deepstack_audio_merger_list.0.down_proj.weight": "model-00004-of-00004.safetensors",
+    "deepstack_audio_merger_list.0.gate_proj.weight": "model-00004-of-00004.safetensors",
+    "deepstack_audio_merger_list.0.up_proj.weight": "model-00004-of-00004.safetensors",
+    "deepstack_audio_merger_list.1.down_proj.weight": "model-00004-of-00004.safetensors",
+    "deepstack_audio_merger_list.1.gate_proj.weight": "model-00004-of-00004.safetensors",
+    "deepstack_audio_merger_list.1.up_proj.weight": "model-00004-of-00004.safetensors",
+    "deepstack_audio_merger_list.2.down_proj.weight": "model-00004-of-00004.safetensors",
+    "deepstack_audio_merger_list.2.gate_proj.weight": "model-00004-of-00004.safetensors",
+    "deepstack_audio_merger_list.2.up_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.embed_tokens.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.self_attn.k_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.self_attn.q_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.0.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.self_attn.k_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.self_attn.q_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.1.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.10.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.10.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.11.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.12.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.13.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.14.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.15.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.16.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.17.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.18.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.19.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.19.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.19.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.19.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.19.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.19.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.19.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.19.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.19.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.19.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.19.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.2.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.self_attn.k_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.self_attn.q_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.2.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.20.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.20.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.21.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.22.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.23.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.24.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.25.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.26.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.27.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.28.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.29.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.3.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.self_attn.k_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.self_attn.q_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.3.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.30.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.30.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.self_attn.k_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.self_attn.q_norm.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.31.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.32.input_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.32.mlp.down_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.32.mlp.gate_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.32.mlp.up_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.32.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.32.self_attn.k_norm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.32.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.32.self_attn.o_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.32.self_attn.q_norm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.32.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.layers.32.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.input_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.mlp.down_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.mlp.gate_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.mlp.up_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.self_attn.k_norm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.self_attn.o_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.self_attn.q_norm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.33.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.input_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.mlp.down_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.mlp.gate_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.mlp.up_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.self_attn.k_norm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.self_attn.o_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.self_attn.q_norm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.34.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.input_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.mlp.down_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.mlp.gate_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.mlp.up_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.self_attn.k_norm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.self_attn.o_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.self_attn.q_norm.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.35.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.layers.4.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.self_attn.k_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.self_attn.q_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.4.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.self_attn.k_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.self_attn.q_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.5.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.6.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.6.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.6.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.6.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.6.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.6.self_attn.k_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.6.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.6.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.6.self_attn.q_norm.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.6.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.6.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.layers.7.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.7.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.8.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.self_attn.k_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.self_attn.q_norm.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.layers.9.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.norm.weight": "model-00004-of-00004.safetensors",
+    "lm_head.weight": "model-00004-of-00004.safetensors"
+  }
+}

preprocessor_config.json

@@ -0,0 +1,6 @@
+{
+  "processor_class": "PalomarProcessor",
+  "auto_map": {
+    "AutoProcessor": "processing_palomar_myaut_mel_whisper.PalomarProcessor"
+  }
+}

processing_moss_audio.py

@@ -0,0 +1,407 @@
+import importlib.util
+import os
+import re
+import sys
+import types
+from dataclasses import dataclass
+from typing import List, Optional, Sequence, Union
+
+import numpy as np
+import torch
+import torchaudio
+from transformers import AutoTokenizer, BatchEncoding
+
+
+@dataclass
+class MelConfig:
+    mel_sr: int = 16000
+    mel_dim: int = 128
+    mel_n_fft: int = 400
+    mel_hop_length: int = 160
+    mel_dtype: torch.dtype = torch.bfloat16
+    use_whisper_feature_extractor: bool = True
+
+
+def load_chat_template(template_path: str, mossflux_path: str = None) -> List:
+    if mossflux_path is None:
+        template_dir = os.path.dirname(os.path.abspath(template_path))
+        current = template_dir
+        while current and os.path.basename(current) != "mossLite":
+            parent = os.path.dirname(current)
+            if parent == current:
+                break
+            current = parent
+        if os.path.basename(current) == "mossLite":
+            mossflux_path = os.path.join(current, "mossflux")
+
+    if mossflux_path and mossflux_path not in sys.path:
+        sys.path.insert(0, mossflux_path)
+
+    spec = importlib.util.spec_from_file_location("chat_template_module", template_path)
+    module = importlib.util.module_from_spec(spec)
+    sys.modules["chat_template_module"] = module
+    spec.loader.exec_module(module)
+    return module.chat_template
+
+
+class MossAudioProcessor:
+    _AUDIO_SPAN_RE = re.compile(r"<\|audio_bos\|>(?:<\|AUDIO\|>)+<\|audio_eos\|>")
+    _auto_class = None
+
+    @classmethod
+    def register_for_auto_class(cls, auto_class="AutoProcessor"):
+        if not isinstance(auto_class, str):
+            auto_class = auto_class.__name__
+        cls._auto_class = auto_class
+
+    def __init__(
+        self,
+        tokenizer,
+        *,
+        mel_config: Optional[MelConfig] = None,
+        template_path: Optional[str] = None,
+        enable_time_marker: bool = True,
+        audio_token_id: int = 151654,
+        audio_start_id: int = 151669,
+        audio_end_id: int = 151670,
+    ):
+        self._base_tokenizer = tokenizer
+        self.tokenizer = tokenizer
+        self.audio_token_id = int(audio_token_id)
+        self.audio_start_id = int(audio_start_id)
+        self.audio_end_id = int(audio_end_id)
+        self.chat_template = (
+            None if template_path is None else load_chat_template(template_path)
+        )
+        self.custom_texts = {}
+        self.enable_time_marker = bool(enable_time_marker)
+        self.config = mel_config or MelConfig()
+        self._whisper_feature_extractor = None
+
+        alias_map = {
+            "<|AUDIO|>": self.audio_token_id,
+            "<|audio_bos|>": self.audio_start_id,
+            "<|audio_eos|>": self.audio_end_id,
+        }
+        orig_convert_tokens_to_ids = self.tokenizer.convert_tokens_to_ids
+
+        def _patched_convert_tokens_to_ids(tokenizer_self, tokens):
+            if isinstance(tokens, (list, tuple)):
+                converted = [
+                    _patched_convert_tokens_to_ids(tokenizer_self, token)
+                    for token in tokens
+                ]
+                return converted if isinstance(tokens, list) else tuple(converted)
+            if isinstance(tokens, str) and tokens in alias_map:
+                return alias_map[tokens]
+            return orig_convert_tokens_to_ids(tokens)
+
+        self.tokenizer.convert_tokens_to_ids = types.MethodType(
+            _patched_convert_tokens_to_ids, self.tokenizer
+        )
+
+        self._digit_token_ids = {
+            "0": 15,
+            "1": 16,
+            "2": 17,
+            "3": 18,
+            "4": 19,
+            "5": 20,
+            "6": 21,
+            "7": 22,
+            "8": 23,
+            "9": 24,
+        }
+        self.audio_tokens_per_second = 12.5
+        self.time_marker_every_seconds = 2
+        self.time_marker_every_audio_tokens = int(
+            self.audio_tokens_per_second * self.time_marker_every_seconds
+        )
+        self.model_input_names = [
+            "input_ids",
+            "attention_mask",
+            "audio_data",
+            "audio_data_seqlens",
+        ]
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_name_or_path, **kwargs):
+        tokenizer_kwargs = {}
+        for key in ["cache_dir", "revision", "token", "local_files_only"]:
+            if key in kwargs:
+                tokenizer_kwargs[key] = kwargs[key]
+
+        tokenizer = AutoTokenizer.from_pretrained(
+            pretrained_model_name_or_path,
+            use_fast=False,
+            **tokenizer_kwargs,
+        )
+
+        mel_config = kwargs.pop("mel_config", None)
+        template_path = kwargs.pop("template_path", None)
+        enable_time_marker = kwargs.pop("enable_time_marker", False)
+        audio_token_id = kwargs.pop("audio_token_id", 151654)
+        audio_start_id = kwargs.pop("audio_start_id", 151669)
+        audio_end_id = kwargs.pop("audio_end_id", 151670)
+
+        return cls(
+            tokenizer,
+            mel_config=mel_config,
+            template_path=template_path,
+            enable_time_marker=enable_time_marker,
+            audio_token_id=audio_token_id,
+            audio_start_id=audio_start_id,
+            audio_end_id=audio_end_id,
+        )
+
+    def load_template(self, template_path: str):
+        self.chat_template = load_chat_template(template_path)
+        return self
+
+    def set_custom_text(self, key: str, text: str):
+        self.custom_texts[key] = text
+        return self
+
+    def clear_custom_text(self, key: Optional[str] = None):
+        if key is None:
+            self.custom_texts.clear()
+        else:
+            self.custom_texts.pop(key, None)
+        return self
+
+    def _template_requires_audio(self) -> bool:
+        if self.chat_template is None:
+            return False
+        for segment in self.chat_template:
+            if segment.type in {"audio_contiguous", "audio_token"}:
+                return True
+        return False
+
+    @staticmethod
+    def _conv3_downsample_len(raw_mel_len: int) -> int:
+        def conv_out_len(length: int) -> int:
+            return (length - 1) // 2 + 1
+
+        length1 = conv_out_len(int(raw_mel_len))
+        length2 = conv_out_len(length1)
+        length3 = conv_out_len(length2)
+        return int(length3)
+
+    def _get_whisper_feature_extractor(self):
+        if self._whisper_feature_extractor is not None:
+            return self._whisper_feature_extractor
+
+        from transformers.models.whisper.feature_extraction_whisper import (
+            WhisperFeatureExtractor,
+        )
+
+        self._whisper_feature_extractor = WhisperFeatureExtractor(
+            feature_size=int(self.config.mel_dim),
+            sampling_rate=int(self.config.mel_sr),
+            hop_length=int(self.config.mel_hop_length),
+            n_fft=int(self.config.mel_n_fft),
+        )
+        return self._whisper_feature_extractor
+
+    def _extract_mel(self, audio: Union[np.ndarray, torch.Tensor]) -> torch.Tensor:
+        if isinstance(audio, np.ndarray):
+            wav = torch.from_numpy(audio)
+        else:
+            wav = audio
+        wav = wav.to(dtype=torch.float32)
+        if wav.dim() == 1:
+            wav = wav.unsqueeze(0)
+
+        if bool(getattr(self.config, "use_whisper_feature_extractor", False)):
+            fe = self._get_whisper_feature_extractor()
+            wav_np = wav.detach().to("cpu", torch.float32).contiguous().numpy()
+            if wav_np.ndim == 2:
+                wav_np = wav_np[0]
+            feats = fe._np_extract_fbank_features(wav_np[None, ...], device="cpu")
+            mel = torch.from_numpy(feats[0])
+
+        return mel.to(dtype=self.config.mel_dtype)
+
+    def _get_time_marker_token_ids(self, second: int) -> List[int]:
+        return [self._digit_token_ids[digit] for digit in str(second)]
+
+    def _build_audio_tokens_with_time_markers(self, audio_seq_len: int) -> List[int]:
+        total_duration_seconds = audio_seq_len / self.audio_tokens_per_second
+        num_full_seconds = int(total_duration_seconds)
+
+        token_ids: List[int] = []
+        audio_tokens_consumed = 0
+        for second in range(
+            self.time_marker_every_seconds,
+            num_full_seconds + 1,
+            self.time_marker_every_seconds,
+        ):
+            marker_pos = (
+                second // self.time_marker_every_seconds
+            ) * self.time_marker_every_audio_tokens
+            audio_segment_len = marker_pos - audio_tokens_consumed
+            if audio_segment_len > 0:
+                token_ids.extend([self.audio_token_id] * audio_segment_len)
+                audio_tokens_consumed += audio_segment_len
+            token_ids.extend(self._get_time_marker_token_ids(second))
+
+        remaining = audio_seq_len - audio_tokens_consumed
+        if remaining > 0:
+            token_ids.extend([self.audio_token_id] * remaining)
+        return token_ids
+
+    def _build_audio_placeholder_ids(self, num_audio_tokens: int) -> List[int]:
+        if self.enable_time_marker:
+            return self._build_audio_tokens_with_time_markers(num_audio_tokens)
+        return [self.audio_token_id] * num_audio_tokens
+
+    def _build_input_from_template(
+        self, num_audio_tokens: int, include_answer: bool = False
+    ) -> List[int]:
+        if self.chat_template is None:
+            raise ValueError("Chat template not loaded.")
+
+        input_ids: List[int] = []
+        for segment in self.chat_template:
+            seg_type = segment.type
+            if seg_type == "constant_text_token":
+                input_ids.extend(segment.text_ids.tolist())
+            elif seg_type in {"audio_contiguous", "audio_token"}:
+                input_ids.extend(self._build_audio_placeholder_ids(num_audio_tokens))
+            elif seg_type == "text_token":
+                text_token_key = segment.text_token_key
+                if "answer" in text_token_key.lower() and not include_answer:
+                    break
+                if text_token_key not in self.custom_texts:
+                    break
+                text_ids = self._base_tokenizer.encode(
+                    self.custom_texts[text_token_key], add_special_tokens=False
+                )
+                input_ids.extend(text_ids)
+
+        return input_ids
+
+    def _build_default_prompt(self, text: str, has_audio: bool) -> str:
+        if has_audio:
+            return (
+                "<|im_start|>system\n"
+                "You are a helpful assistant.<|im_end|>\n"
+                "<|im_start|>user\n"
+                "<|audio_bos|><|AUDIO|><|audio_eos|>\n"
+                f"{text}<|im_end|>\n"
+                "<|im_start|>assistant\n"
+            )
+        return (
+            "<|im_start|>system\n"
+            "You are a helpful assistant.<|im_end|>\n"
+            "<|im_start|>user\n"
+            f"{text}<|im_end|>\n"
+            "<|im_start|>assistant\n"
+        )
+
+    def _build_input_from_prompt(self, prompt: str, token_lens: List[int]) -> List[int]:
+        spans = list(self._AUDIO_SPAN_RE.finditer(prompt))
+        if len(spans) != len(token_lens):
+            raise ValueError(
+                f"Audio placeholder count mismatch: found {len(spans)} spans in text, "
+                f"but got {len(token_lens)} audio inputs."
+            )
+
+        input_ids: List[int] = []
+        cursor = 0
+        for index, match in enumerate(spans):
+            prefix = prompt[cursor : match.start()]
+            if prefix:
+                input_ids.extend(
+                    self._base_tokenizer.encode(prefix, add_special_tokens=False)
+                )
+
+            input_ids.append(self.audio_start_id)
+            input_ids.extend(self._build_audio_placeholder_ids(int(token_lens[index])))
+            input_ids.append(self.audio_end_id)
+            cursor = match.end()
+
+        suffix = prompt[cursor:]
+        if suffix:
+            input_ids.extend(
+                self._base_tokenizer.encode(suffix, add_special_tokens=False)
+            )
+        return input_ids
+
+    def __call__(
+        self,
+        *,
+        text: Union[str, Sequence[str], None] = None,
+        audios: Optional[Sequence[Union[np.ndarray, torch.Tensor]]] = None,
+        audio: Optional[Sequence[Union[np.ndarray, torch.Tensor]]] = None,
+        return_tensors: str = "pt",
+        **kwargs,
+    ):
+        if isinstance(text, (list, tuple)):
+            if len(text) != 1:
+                raise ValueError(f"Expected text batch size 1, got {len(text)}")
+            prompt_text = text[0]
+        else:
+            prompt_text = text
+
+        audio_list = audios if audios is not None else audio
+        audio_list = [] if audio_list is None else list(audio_list)
+
+        mels: List[torch.Tensor] = []
+        raw_lengths: List[int] = []
+        token_lens: List[int] = []
+        for one_audio in audio_list:
+            mel = self._extract_mel(one_audio)
+            raw_len = int(mel.shape[-1])
+            mels.append(mel)
+            raw_lengths.append(raw_len)
+            token_lens.append(self._conv3_downsample_len(raw_len))
+
+        if mels:
+            max_length = max(raw_lengths)
+            audio_batch = torch.zeros(
+                (len(mels), self.config.mel_dim, max_length),
+                dtype=self.config.mel_dtype,
+            )
+            for index, mel in enumerate(mels):
+                audio_batch[index, :, : mel.shape[-1]] = mel
+            seqlens_tensor = torch.tensor(raw_lengths, dtype=torch.long)
+        else:
+            audio_batch = None
+            seqlens_tensor = None
+
+        if prompt_text is not None:
+            if self._AUDIO_SPAN_RE.search(prompt_text) is None and audio_list:
+                prompt_text = self._build_default_prompt(prompt_text, has_audio=True)
+            elif self._AUDIO_SPAN_RE.search(prompt_text) is None and not audio_list:
+                prompt_text = self._build_default_prompt(prompt_text, has_audio=False)
+            input_ids_list = self._build_input_from_prompt(prompt_text, token_lens)
+        elif self.chat_template is not None:
+            input_ids_list = self._build_input_from_template(
+                token_lens[0] if token_lens else 0
+            )
+        else:
+            raise ValueError(
+                "Either provide text or load a chat_template before calling the processor."
+            )
+
+        input_ids_tensor = torch.tensor([input_ids_list], dtype=torch.long)
+        attention_mask_tensor = torch.ones_like(input_ids_tensor)
+
+        data = {
+            "input_ids": input_ids_tensor,
+            "attention_mask": attention_mask_tensor,
+        }
+        if audio_batch is not None:
+            data["audio_data"] = audio_batch
+            data["audio_data_seqlens"] = seqlens_tensor
+        return BatchEncoding(data=data, tensor_type=return_tensors)
+
+    def batch_decode(self, *args, **kwargs):
+        return self._base_tokenizer.batch_decode(*args, **kwargs)
+
+    def decode(self, *args, **kwargs):
+        return self._base_tokenizer.decode(*args, **kwargs)
+
+
+__all__ = ["MelConfig", "MossAudioProcessor"]

special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,271 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151669": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151670": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151671": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151672": {
+      "content": "<|eot|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}