mattChrisP commited on
Commit
a8d652b
·
0 Parent(s):

Release Myna-Hokkien

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +76 -0
  2. README.md +149 -0
  3. config.json +685 -0
  4. configuration_mynahokkien.py +41 -0
  5. demo_samples/01/01_gemini.wav +3 -0
  6. demo_samples/01/01_glmvoice.wav +3 -0
  7. demo_samples/01/01_gptaudio.wav +3 -0
  8. demo_samples/01/01_input.wav +3 -0
  9. demo_samples/01/01_mynahokkien.wav +3 -0
  10. demo_samples/01/01_qwen.wav +3 -0
  11. demo_samples/02/02_gemini.wav +3 -0
  12. demo_samples/02/02_glmvoice.wav +3 -0
  13. demo_samples/02/02_gptaudio.wav +3 -0
  14. demo_samples/02/02_input.wav +3 -0
  15. demo_samples/02/02_mynahokkien.wav +3 -0
  16. demo_samples/02/02_qwen.wav +3 -0
  17. demo_samples/03/03_gemini.wav +3 -0
  18. demo_samples/03/03_glmvoice.wav +3 -0
  19. demo_samples/03/03_gptaudio.wav +3 -0
  20. demo_samples/03/03_input.wav +3 -0
  21. demo_samples/03/03_mynahokkien.wav +3 -0
  22. demo_samples/03/03_qwen.wav +3 -0
  23. demo_samples/04/04_gemini.wav +3 -0
  24. demo_samples/04/04_glmvoice.wav +3 -0
  25. demo_samples/04/04_gptaudio.wav +3 -0
  26. demo_samples/04/04_input.wav +3 -0
  27. demo_samples/04/04_mynahokkien.wav +3 -0
  28. demo_samples/04/04_qwen.wav +3 -0
  29. demo_samples/05/05_gemini.wav +3 -0
  30. demo_samples/05/05_glmvoice.wav +3 -0
  31. demo_samples/05/05_gptaudio.wav +3 -0
  32. demo_samples/05/05_input.wav +3 -0
  33. demo_samples/05/05_mynahokkien.wav +3 -0
  34. demo_samples/05/05_qwen.wav +3 -0
  35. demo_samples/06/06_gemini.wav +3 -0
  36. demo_samples/06/06_glmvoice.wav +3 -0
  37. demo_samples/06/06_gptaudio.wav +3 -0
  38. demo_samples/06/06_input.wav +3 -0
  39. demo_samples/06/06_mynahokkien.wav +3 -0
  40. demo_samples/06/06_qwen.wav +3 -0
  41. demo_samples/07/07_gemini.wav +3 -0
  42. demo_samples/07/07_glmvoice.wav +3 -0
  43. demo_samples/07/07_gptaudio.wav +3 -0
  44. demo_samples/07/07_input.wav +3 -0
  45. demo_samples/07/07_mynahokkien.wav +3 -0
  46. demo_samples/07/07_qwen.wav +3 -0
  47. demo_samples/08_en_text/08_gemini.wav +3 -0
  48. demo_samples/08_en_text/08_glmvoice.wav +3 -0
  49. demo_samples/08_en_text/08_gptaudio.wav +3 -0
  50. demo_samples/08_en_text/08_mynahokkien.wav +3 -0
.gitattributes ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar filter=lfs diff=lfs merge=lfs -text
29
+ *.tflite filter=lfs diff=lfs merge=lfs -text
30
+ *.tgz filter=lfs diff=lfs merge=lfs -text
31
+ *.wasm filter=lfs diff=lfs merge=lfs -text
32
+ *.wav filter=lfs diff=lfs merge=lfs -text
33
+ *.xz filter=lfs diff=lfs merge=lfs -text
34
+ *.zip filter=lfs diff=lfs merge=lfs -text
35
+ *.zst filter=lfs diff=lfs merge=lfs -text
36
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
37
+ q25_processor_config/tokenizer.json filter=lfs diff=lfs merge=lfs -text
38
+ q3_processor_config/tokenizer.json filter=lfs diff=lfs merge=lfs -text
39
+ demo_samples/01/01_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
40
+ demo_samples/01/01_input.wav filter=lfs diff=lfs merge=lfs -text
41
+ demo_samples/01/01_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
42
+ demo_samples/02/02_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
43
+ demo_samples/02/02_input.wav filter=lfs diff=lfs merge=lfs -text
44
+ demo_samples/02/02_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
45
+ demo_samples/02/02_qwen.wav filter=lfs diff=lfs merge=lfs -text
46
+ demo_samples/03/03_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
47
+ demo_samples/03/03_input.wav filter=lfs diff=lfs merge=lfs -text
48
+ demo_samples/03/03_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
49
+ demo_samples/03/03_qwen.wav filter=lfs diff=lfs merge=lfs -text
50
+ demo_samples/04/04_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
51
+ demo_samples/04/04_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
52
+ demo_samples/04/04_qwen.wav filter=lfs diff=lfs merge=lfs -text
53
+ demo_samples/05/05_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
54
+ demo_samples/05/05_input.wav filter=lfs diff=lfs merge=lfs -text
55
+ demo_samples/05/05_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
56
+ demo_samples/05/05_qwen.wav filter=lfs diff=lfs merge=lfs -text
57
+ demo_samples/06/06_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
58
+ demo_samples/06/06_input.wav filter=lfs diff=lfs merge=lfs -text
59
+ demo_samples/06/06_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
60
+ demo_samples/06/06_qwen.wav filter=lfs diff=lfs merge=lfs -text
61
+ demo_samples/07/07_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
62
+ demo_samples/07/07_input.wav filter=lfs diff=lfs merge=lfs -text
63
+ demo_samples/07/07_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
64
+ demo_samples/07/07_qwen.wav filter=lfs diff=lfs merge=lfs -text
65
+ demo_samples/08_en_text/08_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
66
+ demo_samples/08_en_text/08_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
67
+ demo_samples/08_en_text/08_qwen.wav filter=lfs diff=lfs merge=lfs -text
68
+ demo_samples/09_en_text/09_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
69
+ demo_samples/09_en_text/09_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
70
+ demo_samples/09_en_text/09_qwen.wav filter=lfs diff=lfs merge=lfs -text
71
+ demo_samples/10_zh_text/10_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
72
+ demo_samples/10_zh_text/10_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
73
+ demo_samples/10_zh_text/10_qwen.wav filter=lfs diff=lfs merge=lfs -text
74
+ demo_samples/04/04_input.wav filter=lfs diff=lfs merge=lfs -text
75
+ processors/generation/tokenizer.json filter=lfs diff=lfs merge=lfs -text
76
+ processors/reasoning/tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,149 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - nan
4
+ library_name: mynahokkien
5
+ pipeline_tag: audio-to-audio
6
+ tags:
7
+ - hokkien
8
+ - singapore-hokkien
9
+ - speech-to-speech
10
+ - audio
11
+ ---
12
+
13
+ # Myna-Hokkien
14
+
15
+ Myna-Hokkien is an open-source end-to-end conversational speech model for Singapore Hokkien.
16
+ Users speak to the model in Hokkien and the model replies in natural-sounding Hokkien speech -
17
+ no text round-trip required, though text input/output is also supported.
18
+
19
+ Hokkien is the native language of tens of millions of speakers across Asia,
20
+ yet it remains almost entirely absent from mainstream speech AI: no existing omni-style model (either open or closed) natively supports Hokkien.
21
+ Myna-Hokkien is an attempt to close that gap, and to leave behind a reusable recipe for other low-resource language communities to do the same.
22
+
23
+ Proudly built from [iNLP Lab](https://huggingface.co/iNLP-Lab)
24
+
25
+
26
+ ## Model Details
27
+ * Languages: Hokkien / Minnan (闽南语), primarily Singapore-accent in this release
28
+ * Modalities: Audio in → Audio out: full spoken dialogue, text input/output also supported
29
+ * Architecture: Built on two generations of Qwen-Omni via model editing: Qwen2.5-Omni-3B Thinker + trained Transformer Bridge + Qwen3-Omni Talker
30
+ * License:
31
+
32
+
33
+ ## Demo Samples
34
+ Examples showcasing the model's Hokkien understanding & generation capabilities.
35
+
36
+ For comparison, we also ran the same inputs through four frontier audio-to-audio models — GPT Audio, Qwen3.5-Omni-Plus, Gemini, and GLM-Voice —
37
+ none of which officially supports Hokkien;
38
+ we include them to illustrate how far mainstream omni models are from handling the language, not as an apples-to-apples benchmark.
39
+
40
+ | Input | Myna-Hokkien | GPT Audio | Qwen3.5-Omni-Plus | Gemini | GLM-Voice |
41
+ |-------|-------------|-----------|--------------------|--------|-----------|
42
+ | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/01/01_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/01/01_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/01/01_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/01/01_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/01/01_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/01/01_glmvoice.wav"></audio> |
43
+ | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/02/02_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/02/02_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/02/02_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/02/02_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/02/02_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/02/02_glmvoice.wav"></audio> |
44
+ | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/03/03_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/03/03_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/03/03_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/03/03_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/03/03_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/03/03_glmvoice.wav"></audio> |
45
+ | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/04/04_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/04/04_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/04/04_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/04/04_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/04/04_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/04/04_glmvoice.wav"></audio> |
46
+ | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/05/05_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/05/05_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/05/05_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/05/05_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/05/05_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/05/05_glmvoice.wav"></audio> |
47
+ | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/06/06_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/06/06_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/06/06_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/06/06_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/06/06_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/06/06_glmvoice.wav"></audio> |
48
+ | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/07/07_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/07/07_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/07/07_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/07/07_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/07/07_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/07/07_glmvoice.wav"></audio> |
49
+ | Draft an email for me to request to take a day off at work. | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/08_en_text/08_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/08_en_text/08_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/08_en_text/08_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/08_en_text/08_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/08_en_text/08_glmvoice.wav"></audio> |
50
+ | I'm feeling awful today. | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/09_en_text/09_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/09_en_text/09_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/09_en_text/09_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/09_en_text/09_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/09_en_text/09_glmvoice.wav"></audio> |
51
+ | 我今天心情有点不好。 | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/10_zh_text/10_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/10_zh_text/10_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/10_zh_text/10_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/10_zh_text/10_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/10_zh_text/10_glmvoice.wav"></audio> |
52
+
53
+
54
+ ## Installation
55
+
56
+ Download MynaHokkien into the standard Hugging Face cache, then install the included inference
57
+ runtime:
58
+
59
+ ```bash
60
+ pip install --upgrade huggingface_hub
61
+
62
+ MODEL_DIR="$(hf download iNLP-Lab/MynaHokkien --quiet)"
63
+ pip install "$MODEL_DIR"
64
+ ```
65
+
66
+ ## Basic usage — spoken question
67
+
68
+ ```python
69
+ import torch
70
+ import soundfile as sf
71
+ from mynahokkien import MynaHokkien
72
+
73
+ model = MynaHokkien.from_pretrained(
74
+ "iNLP-Lab/MynaHokkien",
75
+ device_map="cuda:0",
76
+ dtype=torch.float16,
77
+ )
78
+
79
+ output = model.generate(
80
+ audio="question.wav",
81
+ prompt="Listen to this audio and reply naturally in Singaporean Hokkien.",
82
+ language="nan",
83
+ return_text=True,
84
+ return_audio=True,
85
+ )
86
+
87
+ print(output.text)
88
+ sf.write("output.wav", output.audio, output.sampling_rate)
89
+ ```
90
+
91
+ ## Text-query usage
92
+
93
+ Text is treated as a question or instruction to the Hokkien assistant; it is not treated as a TTS
94
+ transcript.
95
+
96
+ ```python
97
+ output = model.generate(
98
+ text="講一個新加坡福建話的笑話。",
99
+ language="nan",
100
+ )
101
+ sf.write("output.wav", output.audio, output.sampling_rate)
102
+ print(output.text)
103
+ ```
104
+
105
+ Exactly one of `audio=` and `text=` must be supplied. This release currently supports
106
+ `language="nan"` and the `Ethan` voice.
107
+
108
+ To return only one modality:
109
+
110
+ ```python
111
+ text_only = model.generate(text="你會曉講福建話無?", return_text=True, return_audio=False)
112
+ audio_only = model.generate(text="講一句歡迎詞。", return_text=False, return_audio=True)
113
+ ```
114
+
115
+ ## Prompt behavior
116
+
117
+ If `prompt=` is omitted for audio input, MynaHokkien uses this built-in prompt:
118
+
119
+ ```text
120
+ Answer or respond to what is said in the audio, in spoken Singapore Hokkien. Do NOT repeat or transcribe it.
121
+ ```
122
+
123
+ To override it, pass a different instruction through `prompt=`:
124
+
125
+ ```python
126
+ output = model.generate(
127
+ audio="question.wav",
128
+ prompt="Listen to this audio and reply naturally in Singaporean Hokkien.",
129
+ language="nan",
130
+ )
131
+ ```
132
+
133
+ For text input, put the instruction directly in `text=`.
134
+
135
+
136
+ ## Citation
137
+ ```
138
+ @misc{myna-hokkien-2026,
139
+ title = {Myna-Hokkien: An Open-Source End-to-End Hokkien Spoken Dialogue Model},
140
+ author = {Matt, Ryner, Wenxuan},
141
+ year = {2026},
142
+ howpublished = {\url{xxx}}
143
+ }
144
+ ```
145
+
146
+ ## Contact & collaboration
147
+
148
+ This is an active, ongoing project — we're continuing to improve accent coverage, prosody, and expressiveness.
149
+ We'd love to hear from you if you want to collaborate, have feedback, or run into issues: [Wenxuan Zhang](https://isakzhang.github.io/).
config.json ADDED
@@ -0,0 +1,685 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "MynaHokkienForConditionalGeneration"
4
+ ],
5
+ "bridge_config": {
6
+ "ffn": 6144,
7
+ "heads": 12,
8
+ "inner": 1536,
9
+ "layers": 4,
10
+ "q25_dim": 2048
11
+ },
12
+ "default_language": "nan",
13
+ "default_speaker": "Ethan",
14
+ "generation_embedding_config": {
15
+ "embedding_dim": 2048,
16
+ "num_embeddings": 152064
17
+ },
18
+ "generation_processor_path": "processors/generation",
19
+ "input_sampling_rate": 16000,
20
+ "model_type": "mynahokkien",
21
+ "reasoning_adapter_config": {
22
+ "alora_invocation_tokens": null,
23
+ "alpha_pattern": {},
24
+ "arrow_config": null,
25
+ "auto_mapping": null,
26
+ "base_model_name_or_path": "Qwen2.5-Omni-3B/thinker",
27
+ "bias": "none",
28
+ "corda_config": null,
29
+ "ensure_weight_tying": false,
30
+ "eva_config": null,
31
+ "exclude_modules": null,
32
+ "fan_in_fan_out": false,
33
+ "inference_mode": true,
34
+ "init_lora_weights": true,
35
+ "layer_replication": null,
36
+ "layers_pattern": null,
37
+ "layers_to_transform": null,
38
+ "loftq_config": {},
39
+ "lora_alpha": 32,
40
+ "lora_bias": false,
41
+ "lora_dropout": 0.05,
42
+ "lora_ga_config": null,
43
+ "megatron_config": null,
44
+ "megatron_core": "megatron.core",
45
+ "modules_to_save": null,
46
+ "peft_type": "LORA",
47
+ "peft_version": "0.19.1",
48
+ "qalora_group_size": 16,
49
+ "r": 16,
50
+ "rank_pattern": {},
51
+ "revision": null,
52
+ "target_modules": [
53
+ "q_proj",
54
+ "gate_proj",
55
+ "up_proj",
56
+ "down_proj",
57
+ "k_proj",
58
+ "v_proj",
59
+ "o_proj"
60
+ ],
61
+ "target_parameters": null,
62
+ "task_type": "CAUSAL_LM",
63
+ "trainable_token_indices": null,
64
+ "use_bdlora": null,
65
+ "use_dora": false,
66
+ "use_qalora": false,
67
+ "use_rslora": false
68
+ },
69
+ "reasoning_model_config": {
70
+ "_attn_implementation_autoset": true,
71
+ "_name_or_path": "Qwen2.5-Omni-3B/thinker",
72
+ "architectures": [
73
+ "Qwen2OmniNaViTThinkerForConditionalGeneration"
74
+ ],
75
+ "audio_config": {
76
+ "_attn_implementation_autoset": true,
77
+ "_name_or_path": "",
78
+ "activation_dropout": 0.0,
79
+ "activation_function": "gelu",
80
+ "add_cross_attention": false,
81
+ "architectures": null,
82
+ "attention_dropout": 0.0,
83
+ "bad_words_ids": null,
84
+ "begin_suppress_tokens": null,
85
+ "bos_token_id": null,
86
+ "chunk_size_feed_forward": 0,
87
+ "cross_attention_hidden_size": null,
88
+ "d_model": 1280,
89
+ "decoder_start_token_id": null,
90
+ "diversity_penalty": 0.0,
91
+ "do_sample": false,
92
+ "dropout": 0.0,
93
+ "dtype": null,
94
+ "early_stopping": false,
95
+ "encoder_attention_heads": 20,
96
+ "encoder_ffn_dim": 5120,
97
+ "encoder_layerdrop": 0.0,
98
+ "encoder_layers": 32,
99
+ "encoder_no_repeat_ngram_size": 0,
100
+ "eos_token_id": null,
101
+ "exponential_decay_length_penalty": null,
102
+ "finetuning_task": null,
103
+ "forced_bos_token_id": null,
104
+ "forced_eos_token_id": null,
105
+ "id2label": {
106
+ "0": "LABEL_0",
107
+ "1": "LABEL_1"
108
+ },
109
+ "init_std": 0.02,
110
+ "initializer_range": 0.02,
111
+ "is_decoder": false,
112
+ "is_encoder_decoder": false,
113
+ "label2id": {
114
+ "LABEL_0": 0,
115
+ "LABEL_1": 1
116
+ },
117
+ "length_penalty": 1.0,
118
+ "max_length": 20,
119
+ "max_source_positions": 1500,
120
+ "min_length": 0,
121
+ "model_type": "qwen2_5_omni_audio_encoder",
122
+ "n_window": 100,
123
+ "no_repeat_ngram_size": 0,
124
+ "num_beam_groups": 1,
125
+ "num_beams": 1,
126
+ "num_hidden_layers": 32,
127
+ "num_mel_bins": 128,
128
+ "num_return_sequences": 1,
129
+ "output_attentions": false,
130
+ "output_dim": 2048,
131
+ "output_hidden_states": false,
132
+ "output_scores": false,
133
+ "pad_token_id": null,
134
+ "prefix": null,
135
+ "problem_type": null,
136
+ "pruned_heads": {},
137
+ "remove_invalid_values": false,
138
+ "repetition_penalty": 1.0,
139
+ "return_dict": true,
140
+ "return_dict_in_generate": false,
141
+ "scale_embedding": false,
142
+ "sep_token_id": null,
143
+ "suppress_tokens": null,
144
+ "task_specific_params": null,
145
+ "temperature": 1.0,
146
+ "tf_legacy_loss": false,
147
+ "tie_encoder_decoder": false,
148
+ "tie_word_embeddings": true,
149
+ "tokenizer_class": null,
150
+ "top_k": 50,
151
+ "top_p": 1.0,
152
+ "torchscript": false,
153
+ "typical_p": 1.0,
154
+ "use_bfloat16": false
155
+ },
156
+ "audio_end_token_id": 151648,
157
+ "audio_start_token_id": 151647,
158
+ "audio_token_index": 151646,
159
+ "bos_token_id": 151644,
160
+ "dtype": "bfloat16",
161
+ "eos_token_id": 151645,
162
+ "ignore_index": -100,
163
+ "image_token_index": 151655,
164
+ "init_std": 0.02,
165
+ "initializer_range": 0.02,
166
+ "model_type": "qwen2_5_omni_thinker",
167
+ "pad_token_id": 151643,
168
+ "position_id_per_seconds": 25,
169
+ "seconds_per_chunk": 2,
170
+ "text_config": {
171
+ "_name_or_path": "",
172
+ "add_cross_attention": false,
173
+ "architectures": null,
174
+ "attention_dropout": 0.0,
175
+ "bad_words_ids": null,
176
+ "begin_suppress_tokens": null,
177
+ "bos_token_id": null,
178
+ "chunk_size_feed_forward": 0,
179
+ "cross_attention_hidden_size": null,
180
+ "decoder_start_token_id": null,
181
+ "diversity_penalty": 0.0,
182
+ "do_sample": false,
183
+ "dtype": null,
184
+ "early_stopping": false,
185
+ "encoder_no_repeat_ngram_size": 0,
186
+ "eos_token_id": null,
187
+ "exponential_decay_length_penalty": null,
188
+ "finetuning_task": null,
189
+ "forced_bos_token_id": null,
190
+ "forced_eos_token_id": null,
191
+ "hidden_act": "silu",
192
+ "hidden_size": 2048,
193
+ "id2label": {
194
+ "0": "LABEL_0",
195
+ "1": "LABEL_1"
196
+ },
197
+ "init_std": 0.02,
198
+ "initializer_range": 0.02,
199
+ "intermediate_size": 11008,
200
+ "is_decoder": false,
201
+ "is_encoder_decoder": false,
202
+ "label2id": {
203
+ "LABEL_0": 0,
204
+ "LABEL_1": 1
205
+ },
206
+ "layer_types": [
207
+ "full_attention",
208
+ "full_attention",
209
+ "full_attention",
210
+ "full_attention",
211
+ "full_attention",
212
+ "full_attention",
213
+ "full_attention",
214
+ "full_attention",
215
+ "full_attention",
216
+ "full_attention",
217
+ "full_attention",
218
+ "full_attention",
219
+ "full_attention",
220
+ "full_attention",
221
+ "full_attention",
222
+ "full_attention",
223
+ "full_attention",
224
+ "full_attention",
225
+ "full_attention",
226
+ "full_attention",
227
+ "full_attention",
228
+ "full_attention",
229
+ "full_attention",
230
+ "full_attention",
231
+ "full_attention",
232
+ "full_attention",
233
+ "full_attention",
234
+ "full_attention",
235
+ "full_attention",
236
+ "full_attention",
237
+ "full_attention",
238
+ "full_attention",
239
+ "full_attention",
240
+ "full_attention",
241
+ "full_attention",
242
+ "full_attention"
243
+ ],
244
+ "length_penalty": 1.0,
245
+ "max_length": 20,
246
+ "max_position_embeddings": 32768,
247
+ "max_window_layers": 70,
248
+ "min_length": 0,
249
+ "model_type": "qwen2_5_omni_text",
250
+ "no_repeat_ngram_size": 0,
251
+ "num_attention_heads": 16,
252
+ "num_beam_groups": 1,
253
+ "num_beams": 1,
254
+ "num_hidden_layers": 36,
255
+ "num_key_value_heads": 2,
256
+ "num_return_sequences": 1,
257
+ "output_attentions": false,
258
+ "output_hidden_states": false,
259
+ "output_scores": false,
260
+ "pad_token_id": null,
261
+ "prefix": null,
262
+ "problem_type": null,
263
+ "pruned_heads": {},
264
+ "remove_invalid_values": false,
265
+ "repetition_penalty": 1.0,
266
+ "return_dict": true,
267
+ "return_dict_in_generate": false,
268
+ "rms_norm_eps": 1e-06,
269
+ "rope_parameters": {
270
+ "interleaved": true,
271
+ "mrope_interleaved": true,
272
+ "mrope_section": [
273
+ 24,
274
+ 20,
275
+ 20
276
+ ],
277
+ "rope_theta": 1000000,
278
+ "rope_type": "default",
279
+ "type": "default"
280
+ },
281
+ "rope_scaling": {
282
+ "mrope_section": [
283
+ 16,
284
+ 24,
285
+ 24
286
+ ],
287
+ "rope_type": "default",
288
+ "type": "default"
289
+ },
290
+ "rope_theta": 1000000.0,
291
+ "sep_token_id": null,
292
+ "sliding_window": null,
293
+ "suppress_tokens": null,
294
+ "task_specific_params": null,
295
+ "temperature": 1.0,
296
+ "tf_legacy_loss": false,
297
+ "tie_encoder_decoder": false,
298
+ "tie_word_embeddings": false,
299
+ "tokenizer_class": null,
300
+ "top_k": 50,
301
+ "top_p": 1.0,
302
+ "torchscript": false,
303
+ "typical_p": 1.0,
304
+ "use_bfloat16": false,
305
+ "use_cache": true,
306
+ "use_sliding_window": false,
307
+ "vocab_size": 151936
308
+ },
309
+ "tie_word_embeddings": false,
310
+ "use_cache": true,
311
+ "user_token_id": 872,
312
+ "video_token_index": 151656,
313
+ "vision_config": {
314
+ "_attn_implementation_autoset": true,
315
+ "_name_or_path": "",
316
+ "add_cross_attention": false,
317
+ "architectures": null,
318
+ "bad_words_ids": null,
319
+ "begin_suppress_tokens": null,
320
+ "bos_token_id": null,
321
+ "chunk_size_feed_forward": 0,
322
+ "cross_attention_hidden_size": null,
323
+ "decoder_start_token_id": null,
324
+ "depth": 32,
325
+ "diversity_penalty": 0.0,
326
+ "do_sample": false,
327
+ "dtype": null,
328
+ "early_stopping": false,
329
+ "embed_dim": 1280,
330
+ "encoder_no_repeat_ngram_size": 0,
331
+ "eos_token_id": null,
332
+ "exponential_decay_length_penalty": null,
333
+ "finetuning_task": null,
334
+ "forced_bos_token_id": null,
335
+ "forced_eos_token_id": null,
336
+ "fullatt_block_indexes": [
337
+ 7,
338
+ 15,
339
+ 23,
340
+ 31
341
+ ],
342
+ "hidden_act": "silu",
343
+ "hidden_size": 1280,
344
+ "id2label": {
345
+ "0": "LABEL_0",
346
+ "1": "LABEL_1"
347
+ },
348
+ "in_channels": 3,
349
+ "in_chans": 3,
350
+ "init_std": 0.02,
351
+ "initializer_range": 0.02,
352
+ "intermediate_size": 3420,
353
+ "is_decoder": false,
354
+ "is_encoder_decoder": false,
355
+ "label2id": {
356
+ "LABEL_0": 0,
357
+ "LABEL_1": 1
358
+ },
359
+ "length_penalty": 1.0,
360
+ "max_length": 20,
361
+ "min_length": 0,
362
+ "model_type": "qwen2_5_omni_vision_encoder",
363
+ "no_repeat_ngram_size": 0,
364
+ "num_beam_groups": 1,
365
+ "num_beams": 1,
366
+ "num_heads": 16,
367
+ "num_return_sequences": 1,
368
+ "out_hidden_size": 2048,
369
+ "output_attentions": false,
370
+ "output_hidden_states": false,
371
+ "output_scores": false,
372
+ "pad_token_id": null,
373
+ "patch_size": 14,
374
+ "prefix": null,
375
+ "problem_type": null,
376
+ "pruned_heads": {},
377
+ "remove_invalid_values": false,
378
+ "repetition_penalty": 1.0,
379
+ "return_dict": true,
380
+ "return_dict_in_generate": false,
381
+ "sep_token_id": null,
382
+ "spatial_merge_size": 2,
383
+ "spatial_patch_size": 14,
384
+ "suppress_tokens": null,
385
+ "task_specific_params": null,
386
+ "temperature": 1.0,
387
+ "temporal_patch_size": 2,
388
+ "tf_legacy_loss": false,
389
+ "tie_encoder_decoder": false,
390
+ "tie_word_embeddings": true,
391
+ "tokenizer_class": null,
392
+ "tokens_per_second": 25,
393
+ "top_k": 50,
394
+ "top_p": 1.0,
395
+ "torchscript": false,
396
+ "typical_p": 1.0,
397
+ "use_bfloat16": false,
398
+ "window_size": 112
399
+ },
400
+ "vision_end_token_id": 151653,
401
+ "vision_start_token_id": 151652,
402
+ "vision_token_id": 151654
403
+ },
404
+ "reasoning_processor_path": "processors/reasoning",
405
+ "release_epoch": 3,
406
+ "sampling_rate": 24000,
407
+ "speech_generator_config": {
408
+ "accept_hidden_layer": 24,
409
+ "audio_end_token_id": 151670,
410
+ "audio_start_token_id": 151669,
411
+ "audio_token_id": 151675,
412
+ "audio_tower_config": {},
413
+ "code_predictor_config": {
414
+ "_name_or_path": "",
415
+ "add_cross_attention": false,
416
+ "architectures": null,
417
+ "attention_bias": false,
418
+ "attention_dropout": 0,
419
+ "bad_words_ids": null,
420
+ "begin_suppress_tokens": null,
421
+ "bos_token_id": null,
422
+ "chunk_size_feed_forward": 0,
423
+ "cross_attention_hidden_size": null,
424
+ "decoder_start_token_id": null,
425
+ "diversity_penalty": 0.0,
426
+ "do_sample": false,
427
+ "dtype": "bfloat16",
428
+ "early_stopping": false,
429
+ "encoder_no_repeat_ngram_size": 0,
430
+ "eos_token_id": null,
431
+ "exponential_decay_length_penalty": null,
432
+ "finetuning_task": null,
433
+ "forced_bos_token_id": null,
434
+ "forced_eos_token_id": null,
435
+ "head_dim": 128,
436
+ "hidden_act": "silu",
437
+ "hidden_size": 1024,
438
+ "id2label": {
439
+ "0": "LABEL_0",
440
+ "1": "LABEL_1"
441
+ },
442
+ "initializer_range": 0.02,
443
+ "intermediate_size": 3072,
444
+ "is_decoder": false,
445
+ "is_encoder_decoder": false,
446
+ "label2id": {
447
+ "LABEL_0": 0,
448
+ "LABEL_1": 1
449
+ },
450
+ "layer_types": [
451
+ "full_attention",
452
+ "full_attention",
453
+ "full_attention",
454
+ "full_attention",
455
+ "full_attention"
456
+ ],
457
+ "length_penalty": 1.0,
458
+ "max_length": 20,
459
+ "max_position_embeddings": 32768,
460
+ "max_window_layers": 28,
461
+ "min_length": 0,
462
+ "model_type": "qwen3_omni_moe_talker_code_predictor",
463
+ "no_repeat_ngram_size": 0,
464
+ "num_attention_heads": 16,
465
+ "num_beam_groups": 1,
466
+ "num_beams": 1,
467
+ "num_code_groups": 16,
468
+ "num_hidden_layers": 5,
469
+ "num_key_value_heads": 8,
470
+ "num_return_sequences": 1,
471
+ "output_attentions": false,
472
+ "output_hidden_states": false,
473
+ "output_scores": false,
474
+ "pad_token_id": null,
475
+ "prefix": null,
476
+ "problem_type": null,
477
+ "pruned_heads": {},
478
+ "remove_invalid_values": false,
479
+ "repetition_penalty": 1.0,
480
+ "return_dict": true,
481
+ "return_dict_in_generate": false,
482
+ "rms_norm_eps": 1e-06,
483
+ "rope_parameters": {
484
+ "rope_theta": 1000000,
485
+ "rope_type": "default"
486
+ },
487
+ "rope_scaling": null,
488
+ "rope_theta": 10000,
489
+ "sep_token_id": null,
490
+ "sliding_window": null,
491
+ "suppress_tokens": null,
492
+ "task_specific_params": null,
493
+ "temperature": 1.0,
494
+ "tf_legacy_loss": false,
495
+ "tie_encoder_decoder": false,
496
+ "tie_word_embeddings": false,
497
+ "tokenizer_class": null,
498
+ "top_k": 50,
499
+ "top_p": 1.0,
500
+ "torchscript": false,
501
+ "typical_p": 1.0,
502
+ "use_bfloat16": false,
503
+ "use_cache": true,
504
+ "use_sliding_window": false,
505
+ "vocab_size": 2048
506
+ },
507
+ "codec_bos_id": 2149,
508
+ "codec_config": {},
509
+ "codec_eos_token_id": 2150,
510
+ "codec_nothink_id": 2155,
511
+ "codec_pad_id": 2148,
512
+ "codec_think_bos_id": 2156,
513
+ "codec_think_eos_id": 2157,
514
+ "dtype": "bfloat16",
515
+ "image_token_id": 151655,
516
+ "initializer_range": 0.02,
517
+ "model_type": "",
518
+ "num_code_groups": 16,
519
+ "output_router_logits": false,
520
+ "position_id_per_seconds": 13,
521
+ "seconds_per_chunk": 2,
522
+ "spatial_merge_size": 2,
523
+ "speaker_id": {
524
+ "aiden": 2303,
525
+ "chelsie": 2301,
526
+ "ethan": 2302
527
+ },
528
+ "text_config": {
529
+ "_name_or_path": "",
530
+ "add_cross_attention": false,
531
+ "architectures": null,
532
+ "attention_bias": false,
533
+ "attention_dropout": 0,
534
+ "bad_words_ids": null,
535
+ "begin_suppress_tokens": null,
536
+ "bos_token_id": null,
537
+ "chunk_size_feed_forward": 0,
538
+ "cross_attention_hidden_size": null,
539
+ "decoder_sparse_step": 1,
540
+ "decoder_start_token_id": null,
541
+ "diversity_penalty": 0.0,
542
+ "do_sample": false,
543
+ "dtype": "bfloat16",
544
+ "early_stopping": false,
545
+ "encoder_no_repeat_ngram_size": 0,
546
+ "eos_token_id": null,
547
+ "exponential_decay_length_penalty": null,
548
+ "finetuning_task": null,
549
+ "forced_bos_token_id": null,
550
+ "forced_eos_token_id": null,
551
+ "head_dim": 128,
552
+ "hidden_act": "silu",
553
+ "hidden_size": 1024,
554
+ "id2label": {
555
+ "0": "LABEL_0",
556
+ "1": "LABEL_1"
557
+ },
558
+ "initializer_range": 0.02,
559
+ "intermediate_size": 2048,
560
+ "is_decoder": false,
561
+ "is_encoder_decoder": false,
562
+ "label2id": {
563
+ "LABEL_0": 0,
564
+ "LABEL_1": 1
565
+ },
566
+ "length_penalty": 1.0,
567
+ "max_length": 20,
568
+ "max_position_embeddings": 65536,
569
+ "min_length": 0,
570
+ "mlp_only_layers": [],
571
+ "model_type": "qwen3_omni_moe_talker_text",
572
+ "moe_intermediate_size": 384,
573
+ "no_repeat_ngram_size": 0,
574
+ "norm_topk_prob": true,
575
+ "num_attention_heads": 16,
576
+ "num_beam_groups": 1,
577
+ "num_beams": 1,
578
+ "num_experts": 128,
579
+ "num_experts_per_tok": 6,
580
+ "num_hidden_layers": 20,
581
+ "num_key_value_heads": 2,
582
+ "num_local_experts": 128,
583
+ "num_return_sequences": 1,
584
+ "output_attentions": false,
585
+ "output_hidden_states": false,
586
+ "output_router_logits": false,
587
+ "output_scores": false,
588
+ "pad_token_id": null,
589
+ "prefix": null,
590
+ "problem_type": null,
591
+ "pruned_heads": {},
592
+ "remove_invalid_values": false,
593
+ "repetition_penalty": 1.0,
594
+ "return_dict": true,
595
+ "return_dict_in_generate": false,
596
+ "rms_norm_eps": 1e-06,
597
+ "rope_parameters": {
598
+ "interleaved": true,
599
+ "mrope_section": [
600
+ 24,
601
+ 20,
602
+ 20
603
+ ],
604
+ "rope_theta": 1000000,
605
+ "rope_type": "default",
606
+ "type": "default"
607
+ },
608
+ "rope_scaling": {
609
+ "interleaved": true,
610
+ "mrope_section": [
611
+ 24,
612
+ 20,
613
+ 20
614
+ ],
615
+ "rope_type": "default",
616
+ "type": "default"
617
+ },
618
+ "rope_theta": 10000,
619
+ "router_aux_loss_coef": 0.001,
620
+ "sep_token_id": null,
621
+ "shared_expert_intermediate_size": 768,
622
+ "sliding_window": null,
623
+ "suppress_tokens": null,
624
+ "task_specific_params": null,
625
+ "temperature": 1.0,
626
+ "tf_legacy_loss": false,
627
+ "tie_encoder_decoder": false,
628
+ "tie_word_embeddings": false,
629
+ "tokenizer_class": null,
630
+ "top_k": 50,
631
+ "top_p": 1.0,
632
+ "torchscript": false,
633
+ "typical_p": 1.0,
634
+ "use_bfloat16": false,
635
+ "use_cache": true,
636
+ "use_sliding_window": false,
637
+ "vocab_size": 3072
638
+ },
639
+ "thinker_hidden_size": 2048,
640
+ "tie_word_embeddings": false,
641
+ "video_token_id": 151656,
642
+ "vision_start_token_id": 151652
643
+ },
644
+ "torch_dtype": "float16",
645
+ "transformers_version": "4.57.1",
646
+ "waveform_decoder_config": {
647
+ "attention_bias": false,
648
+ "attention_dropout": 0.0,
649
+ "codebook_dim": 512,
650
+ "codebook_size": 2048,
651
+ "decoder_dim": 1536,
652
+ "dtype": "bfloat16",
653
+ "hidden_act": "silu",
654
+ "hidden_size": 1024,
655
+ "initializer_range": 0.02,
656
+ "intermediate_size": 3072,
657
+ "layer_scale_initial_scale": 0.01,
658
+ "max_position_embeddings": 8000,
659
+ "model_type": "",
660
+ "num_attention_heads": 16,
661
+ "num_hidden_layers": 8,
662
+ "num_key_value_heads": 16,
663
+ "num_quantizers": 16,
664
+ "num_semantic_quantizers": 1,
665
+ "rms_norm_eps": 1e-05,
666
+ "rope_parameters": {
667
+ "rope_theta": 10000,
668
+ "rope_type": "default"
669
+ },
670
+ "rope_theta": 10000,
671
+ "semantic_codebook_size": 4096,
672
+ "sliding_window": 72,
673
+ "upsample_rates": [
674
+ 8,
675
+ 5,
676
+ 4,
677
+ 3
678
+ ],
679
+ "upsampling_ratios": [
680
+ 2,
681
+ 2
682
+ ],
683
+ "vector_quantization_hidden_dimension": 512
684
+ }
685
+ }
configuration_mynahokkien.py ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ from transformers import PretrainedConfig
4
+
5
+
6
+ class MynaHokkienConfig(PretrainedConfig):
7
+ """Configuration for the unified MynaHokkien inference checkpoint."""
8
+
9
+ model_type = "mynahokkien"
10
+
11
+ def __init__(
12
+ self,
13
+ *,
14
+ reasoning_model_config=None,
15
+ reasoning_adapter_config=None,
16
+ speech_generator_config=None,
17
+ waveform_decoder_config=None,
18
+ generation_embedding_config=None,
19
+ bridge_config=None,
20
+ reasoning_processor_path="processors/reasoning",
21
+ generation_processor_path="processors/generation",
22
+ sampling_rate=24000,
23
+ input_sampling_rate=16000,
24
+ default_language="nan",
25
+ default_speaker="Ethan",
26
+ **kwargs,
27
+ ):
28
+ super().__init__(**kwargs)
29
+ self.reasoning_model_config = reasoning_model_config or {}
30
+ self.reasoning_adapter_config = reasoning_adapter_config or {}
31
+ self.speech_generator_config = speech_generator_config or {}
32
+ self.waveform_decoder_config = waveform_decoder_config or {}
33
+ self.generation_embedding_config = generation_embedding_config or {}
34
+ self.bridge_config = bridge_config or {}
35
+ self.reasoning_processor_path = reasoning_processor_path
36
+ self.generation_processor_path = generation_processor_path
37
+ self.sampling_rate = sampling_rate
38
+ self.input_sampling_rate = input_sampling_rate
39
+ self.default_language = default_language
40
+ self.default_speaker = default_speaker
41
+
demo_samples/01/01_gemini.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b4947de9b7d40b99a7f1185b340f22ab4e51b46bb073038114ac2a4c9a33e6b5
3
+ size 424366
demo_samples/01/01_glmvoice.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:54ba51eeee9ce255b8ef8a7c4d2d4fad14bd353203d4cbc84abe0f7a99a28973
3
+ size 1890860
demo_samples/01/01_gptaudio.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d51f604229433f703f773645a735b09b141f7a8853997a2dd5051fbb260f55a7
3
+ size 1370444
demo_samples/01/01_input.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:219bb25e83f5738074fbc6f049e4288138cbb0204bd9238df9e660558ffb99d6
3
+ size 598094
demo_samples/01/01_mynahokkien.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:794fd7c9935c5d3fed2362c90b0c2844b252d0781a688e7a91d2b28b21d326ed
3
+ size 114134
demo_samples/01/01_qwen.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a57642c6aeda75e577dbeb469240fd523a1743bafcf37cdf0f0b060494d4f11e
3
+ size 69164
demo_samples/02/02_gemini.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd3f422e4c2dcf2fed091f88fb22d49c29e441176941f6da52e44e87d8fb9538
3
+ size 419086
demo_samples/02/02_glmvoice.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:496df0db2bfeb33d38e8875f86d525f9be202048e187ad1892cc20dbd798fa68
3
+ size 472620
demo_samples/02/02_gptaudio.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e84ea3a842a893957715c90d8a426dc572981334ae5fee5864c2036188d9495f
3
+ size 345644
demo_samples/02/02_input.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ee23b89cc6546151fd72f25a767131c64818f7e7b92bc9b421ec439d2ac6faec
3
+ size 1294414
demo_samples/02/02_mynahokkien.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3b449ddf28c4bb722b36958fe625e47998aabb163d5b36bc2b53ff39fc5772d6
3
+ size 179414
demo_samples/02/02_qwen.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a021427dd029ed591479c9e18718fe8a034cced82c926c47dcd02dedd63596f0
3
+ size 126764
demo_samples/03/03_gemini.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:db23c52e56a7f9e1e2b4aa720178d302cdd2ee33a1c8254d2d63aabec694236e
3
+ size 666286
demo_samples/03/03_glmvoice.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8bb303cf57c8e4c5a5e9a9bcddf2599e44f9207e86ebc0b4ebeedfc6d6e2f1e8
3
+ size 1562668
demo_samples/03/03_gptaudio.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d822eebd35ed59ad18a7270fd43bb756077eab8fbe8b990b437d8a22d3c209c9
3
+ size 1833644
demo_samples/03/03_input.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eabd9e01ac1b943ba07934f8fde1c5450c2153e64caaa559e995070394f482d0
3
+ size 761934
demo_samples/03/03_mynahokkien.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:09ac764e9756aa6fe5bbdf673d7e5fe20284a510bb1993cdbe01dc2f3568ec7d
3
+ size 179414
demo_samples/03/03_qwen.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3720f57a19ccd29d38628d55667b57c30c078d1759bb9b6c8100efa896d4032
3
+ size 430124
demo_samples/04/04_gemini.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b1f2f0ebae3a968fe41a0ec991f26fd7ebea92dcbbe2d4b49bc61a78a36e903e
3
+ size 313006
demo_samples/04/04_glmvoice.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:64e125e026597b22b339c6c0d2dfe4bc57bbe3630c3c8577f25302e487f22a33
3
+ size 1432108
demo_samples/04/04_gptaudio.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7adbe59113e792672115d7cf0fd7225628c1275c0c476ab1310607f72d4484df
3
+ size 842444
demo_samples/04/04_input.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f171401b9b4674972877b73a439895ff59664952da1f06dab0b43538b954596d
3
+ size 598094
demo_samples/04/04_mynahokkien.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce69f759d9fead2c0f7c52f81de53ac85e9c615c33df46f2551cd854ad7fe99c
3
+ size 240854
demo_samples/04/04_qwen.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e60638f088e4ce7cdd768ca3cf61747359de1b4128cf66899990fbf999237c0d
3
+ size 115244
demo_samples/05/05_gemini.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:59f7478fe346dbdb40b253dfeaed548cf1f8843343d3708ffd7c7dfb588470f1
3
+ size 915914
demo_samples/05/05_glmvoice.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9c4c9ab6c8726e942ee1eecba9cdd29f25bb404ed16c7ea1acf2284256eed946
3
+ size 588844
demo_samples/05/05_gptaudio.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c599238c7dc534260d3b61330d8cc6b6f02675b1309c98ca8783815a91151031
3
+ size 1663244
demo_samples/05/05_input.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe21f26b571496c1cfff99abdfe107bf390be38761b3b58d586e2b9bd0018397
3
+ size 737358
demo_samples/05/05_mynahokkien.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:28faf6f08fa43c95736aade84fa07ecd1ff105c4580e94e7419e9153e808f1eb
3
+ size 252374
demo_samples/05/05_qwen.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7e2e0c24a760d5d95b6fcf7b37a9d1082cdb05bddcdc295e4d28c5cd4fc3a0a8
3
+ size 103724
demo_samples/06/06_gemini.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f238a6f2fa8dcdcbd017b579ea62d5d8ef2d3298a0e7db555d35b1abb37ec3b2
3
+ size 670154
demo_samples/06/06_glmvoice.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d9ff1b557a01fa3d9d3f414e053dfa20e0ae27446760b106815db50a566ae74
3
+ size 260652
demo_samples/06/06_gptaudio.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f2b6d70f6350126fdb2968faf4d3e11fc70234f3635ad419ac09339f1c60fc15
3
+ size 832844
demo_samples/06/06_input.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:afd1a802932f2f99d58b687f69d4c7123cfa650a592cd61976fc181d3c530eb0
3
+ size 819278
demo_samples/06/06_mynahokkien.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:134d1e110eaa589424b50b83139c755d90dab4737ee5e99133728e48244c9933
3
+ size 275414
demo_samples/06/06_qwen.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:319c45c76d4f991d6e19b97dd7b56f425b4be1d5cd745ced316e88eed07a3463
3
+ size 188204
demo_samples/07/07_gemini.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:11a9fe4a8bf7aefd93344e881d682e90cc52ad5a95644980811e4aaffc454082
3
+ size 464206
demo_samples/07/07_glmvoice.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:60e948718db900fdee3a38485d94db805dfe61f45c541beebb10e278805e0f9a
3
+ size 6748716
demo_samples/07/07_gptaudio.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aa362ab3d90c80964ea1a183f453b5355ab2bfd7e29d928e24a250f081708ef0
3
+ size 1504844
demo_samples/07/07_input.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c6b89eda545ffc8fa95ed71c55b5d6382645a58c6a7cea65317cdb5cc1d79d97
3
+ size 876622
demo_samples/07/07_mynahokkien.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fdfc846ab7dfe4eb4e969fec56ba9c8ae26d6b228f199b888707b9b9eea1855e
3
+ size 152534
demo_samples/07/07_qwen.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:367eea846173800a96ccb645170a2d3126dd1595220521f11226ba10535b1dd8
3
+ size 192044
demo_samples/08_en_text/08_gemini.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:504b74c947ffd3252b10ea454a6b4433e8d7958ab5594d649cfd9bd2c0f44309
3
+ size 673006
demo_samples/08_en_text/08_glmvoice.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:02c45cda78f77c00e03aa168184bc11bdf00cf37fbf633e9b26dd7e2528e5a94
3
+ size 1428524
demo_samples/08_en_text/08_gptaudio.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:52ef86bd4acf84409217f44d3d4ab06d542a266824552171427f6711f6c93cd5
3
+ size 2200844
demo_samples/08_en_text/08_mynahokkien.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:abc55668f39b38b3fac7e4b81b816129fe3fa8274f4870ec72457843c6e4c5ca
3
+ size 294614