release myna

.gitattributes

@@ -1,76 +1,5 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
 *.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
 *.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
 *.wav filter=lfs diff=lfs merge=lfs -text
-*.xz 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
-q25_processor_config/tokenizer.json filter=lfs diff=lfs merge=lfs -text
-q3_processor_config/tokenizer.json filter=lfs diff=lfs merge=lfs -text
-demo_samples/01/01_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/01/01_input.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/01/01_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/02/02_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/02/02_input.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/02/02_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/02/02_qwen.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/03/03_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/03/03_input.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/03/03_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/03/03_qwen.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/04/04_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/04/04_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/04/04_qwen.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/05/05_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/05/05_input.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/05/05_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/05/05_qwen.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/06/06_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/06/06_input.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/06/06_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/06/06_qwen.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/07/07_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/07/07_input.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/07/07_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/07/07_qwen.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/08_en_text/08_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/08_en_text/08_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/08_en_text/08_qwen.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/09_en_text/09_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/09_en_text/09_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/09_en_text/09_qwen.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/10_zh_text/10_gptaudio.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/10_zh_text/10_mynahokkien.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/10_zh_text/10_qwen.wav filter=lfs diff=lfs merge=lfs -text
-demo_samples/04/04_input.wav filter=lfs diff=lfs merge=lfs -text
-processors/generation/tokenizer.json filter=lfs diff=lfs merge=lfs -text
-processors/reasoning/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,7 @@
+.cache/
+__pycache__/
+*.py[cod]
+build/
+dist/
+*.egg-info/
+merge_info.json

MODEL_LINEAGE.json

@@ -0,0 +1,22 @@
+{
+  "base_model": "Qwen/Qwen3-Omni-30B-A3B-Instruct",
+  "release_family": "release_v2",
+  "thinker": {
+    "stage": 2,
+    "experiment": "thinker_stage2_meralion_asrqa_fullft_v1",
+    "training": "full-parameter continuation",
+    "state_during_talker_stage4": "frozen"
+  },
+  "talker": {
+    "stage": 4,
+    "experiment": "talker_stage4_currentbest_humanchat_s2s_v1",
+    "epoch": 1,
+    "training": "full Talker and MTP continuation"
+  },
+  "native_hidden_routing": {
+    "text_and_assistant_positions": 0,
+    "user_audio_positions": 24,
+    "runtime_ablation_passed": true
+  },
+  "speaker": "Ethan"
+}

README.md

@@ -1,59 +1,52 @@
 ---
 language:
 - nan
-library_name: mynahokkien
+license: apache-2.0
+library_name: transformers
 pipeline_tag: audio-to-audio
+base_model: Qwen/Qwen3-Omni-30B-A3B-Instruct
 tags:
 - hokkien
 - singapore-hokkien
 - speech-to-speech
 - audio
+- qwen3-omni
 ---
 
 # Myna-Hokkien
 
-Myna-Hokkien is an open-source end-to-end conversational speech model for Singapore Hokkien. 
-Users speak to the model in Hokkien and the model replies in natural-sounding Hokkien speech - 
-no text round-trip required, though text input/output is also supported.
+Myna-Hokkien is an open-source conversational speech model for Singapore Hokkien. It accepts a
+spoken or written question and returns Singapore-Hokkien text and 24 kHz speech.
 
-Hokkien is the native language of tens of millions of speakers across Asia, 
-yet it remains almost entirely absent from mainstream speech AI: no existing omni-style model (either open or closed) natively supports Hokkien. 
-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.
+This release is the native Qwen3-Omni `release_v2` model: the selected Stage-2 Thinker together
+with the Stage-4 epoch-1 Talker and MTP. The public Python API is
+`mynahokkien.MynaHokkien`.
 
-Proudly built from [iNLP Lab](https://huggingface.co/iNLP-Lab)
+Proudly built by [iNLP Lab](https://huggingface.co/iNLP-Lab) at SUTD.
 
+## Model details
 
-## Model Details
-* Languages: Hokkien / Minnan (闽南语), primarily Singapore-accent in this release
-* Modalities: Audio in → Audio out: full spoken dialogue, text input/output also supported
-* Architecture: Built on two generations of Qwen-Omni via model editing: Qwen2.5-Omni-3B Thinker + trained Transformer Bridge + Qwen3-Omni Talker
-* License: 
+- **Base architecture:** Qwen3-Omni-30B-A3B-Instruct, native Thinker–Talker
+- **Thinker:** `thinker_stage2_meralion_asrqa_fullft_v1`
+- **Talker:** `talker_stage4_currentbest_humanchat_s2s_v1`, epoch 1
+- **Language:** Singapore Hokkien / Minnan (`nan`)
+- **Voice:** Ethan
+- **Input:** 16 kHz mono audio or text
+- **Output:** text and 24 kHz mono audio
+- **License:** Apache-2.0
 
+The model is approximately 70.5 GB in BF16/FP16 weights. A large-memory CUDA GPU is strongly
+recommended; the release workflow is validated on an H200.
 
-## Demo Samples
-Examples showcasing the model's Hokkien understanding & generation capabilities.
-
-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 — 
-none of which officially supports Hokkien; 
-we include them to illustrate how far mainstream omni models are from handling the language, not as an apples-to-apples benchmark.
-
-| Input | Myna-Hokkien | GPT Audio | Qwen3.5-Omni-Plus | Gemini | GLM-Voice |
-|-------|-------------|-----------|--------------------|--------|-----------|
-| <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/01/01_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/01/01_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/01/01_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/01/01_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/01/01_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/01/01_glmvoice.wav"></audio> |
-| <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/02/02_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/02/02_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/02/02_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/02/02_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/02/02_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/02/02_glmvoice.wav"></audio> |
-| <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/03/03_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/03/03_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/03/03_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/03/03_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/03/03_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/03/03_glmvoice.wav"></audio> |
-| <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/04/04_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/04/04_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/04/04_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/04/04_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/04/04_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/04/04_glmvoice.wav"></audio> |
-| <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/05/05_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/05/05_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/05/05_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/05/05_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/05/05_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/05/05_glmvoice.wav"></audio> |
-| <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/06/06_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/06/06_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/06/06_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/06/06_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/06/06_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/06/06_glmvoice.wav"></audio> |
-| <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/07/07_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/07/07_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/07/07_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/07/07_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/07/07_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/07/07_glmvoice.wav"></audio> |
-| <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/08/08_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/08/08_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/08/08_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/08/08_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/08/08_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/08/08_glmvoice.wav"></audio> |
-| <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/09/09_input.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/09/09_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/09/09_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/09/09_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/09/09_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/09/09_glmvoice.wav"></audio> |
-| 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/release_csv/10/10_mynahokkien.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/10/10_gptaudio.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/10/10_qwen.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/10/10_gemini.wav"></audio> | <audio controls src="https://huggingface.co/iNLP-Lab/MynaHokkien/resolve/main/demo_samples/release_csv/10/10_glmvoice.wav"></audio> |
+## Qualitative evaluation
 
+The [playable unseen-audio probe](https://mattchrisp.github.io/omni-hokkien/) contains the
+Stage-4 model outputs, prompt text and raw per-sample results. It is a qualitative listening
+probe, not a scored benchmark: the recordings have no reference answers.
 
 ## Installation
 
-Download MynaHokkien into the standard Hugging Face cache, then install the included inference
+Download Myna-Hokkien into the standard Hugging Face cache, then install the included inference
 runtime:
 
 ```bash
@@ -63,6 +56,9 @@ MODEL_DIR="$(hf download iNLP-Lab/MynaHokkien --quiet)"
 pip install "$MODEL_DIR"
 ```
 
+The runtime uses `transformers==4.57.1`, matching the version recorded in the released model
+configuration.
+
 ## Basic usage — spoken question
 
 ```python
@@ -78,7 +74,6 @@ model = MynaHokkien.from_pretrained(
 
 output = model.generate(
     audio="question.wav",
-    prompt="Listen to this audio and reply naturally in Singaporean Hokkien.",
     language="nan",
     return_text=True,
     return_audio=True,
@@ -88,10 +83,13 @@ print(output.text)
 sf.write("output.wav", output.audio, output.sampling_rate)
 ```
 
+Omitting `prompt=` is intentional: the release runtime supplies the default forced-Hokkien
+instruction shown below.
+
 ## Text-query usage
 
 Text is treated as a question or instruction to the Hokkien assistant; it is not treated as a TTS
-transcript.
+transcript. The runtime prefixes it with the corresponding forced-Hokkien text instruction.
 
 ```python
 output = model.generate(
@@ -102,8 +100,8 @@ sf.write("output.wav", output.audio, output.sampling_rate)
 print(output.text)
 ```
 
-Exactly one of `audio=` and `text=` must be supplied. This release currently supports
-`language="nan"` and the `Ethan` voice.
+Exactly one of `audio=` and `text=` must be supplied. This release supports `language="nan"` and
+the `Ethan` voice.
 
 To return only one modality:
 
@@ -114,36 +112,57 @@ audio_only = model.generate(text="講一句歡迎詞。", return_text=False, ret
 
 ## Prompt behavior
 
-If `prompt=` is omitted for audio input, MynaHokkien uses this built-in prompt:
+The stock Qwen system turn is left unchanged. For audio input, the runtime places the following
+instruction **after the audio in the same user turn**:
 
 ```text
-Answer or respond to what is said in the audio, in spoken Singapore Hokkien. Do NOT repeat or transcribe it.
+Listen to the spoken Hokkien and reply naturally in concise Singapore Hokkien. Always answer in colloquial Singapore Hokkien written in Hanji. Never answer in Mandarin or English. Do not repeat or transcribe the input; respond to it directly.
 ```
 
-To override it, pass a different instruction through `prompt=`:
+To override the audio instruction, pass the complete replacement through `prompt=`:
 
 ```python
 output = model.generate(
     audio="question.wav",
-    prompt="Listen to this audio and reply naturally in Singaporean Hokkien.",
+    prompt="Listen to the spoken Hokkien and reply naturally in concise Singapore Hokkien.",
     language="nan",
 )
 ```
 
-For text input, put the instruction directly in `text=`.
+For text queries, the runtime uses the corresponding direct-response text instruction recorded in
+`release_config.json`, then appends the caller's `text=` in the same user turn. These are inference
+instructions; their added constraint clauses were not separate training targets.
 
+## Limitations
 
-## Citation
+- This is an active research release, not a production assistant.
+- The model can mirror or restate the speaker instead of answering them.
+- The prompt strongly requests Hokkien but does not provide a hard decoding guarantee.
+- Automatic language-marker screens are incomplete and should not replace listening evaluation.
+- The model is large and has only been validated in the release workflow on H200-class hardware.
+
+## Included CLI
+
+After installation, or from the downloaded model directory:
+
+```bash
+python inference.py \
+  --audio question.wav \
+  --output-wav answer.wav \
+  --output-text answer.txt
 ```
+
+## Citation
+
+```bibtex
 @misc{myna-hokkien-2026,
   title  = {Myna-Hokkien: An Open-Source End-to-End Hokkien Spoken Dialogue Model},
-  author = {Matt, Ryner, Wenxuan},
-  year   = {2026},
-  howpublished = {\url{xxx}}
+  author = {Matt and Ryner and Wenxuan},
+  year   = {2026}
 }
 ```
 
-## Contact & collaboration
+## Contact and collaboration
 
-This is an active, ongoing project — we're continuing to improve accent coverage, prosody, and expressiveness. 
-We'd love to hear from you if you want to collaborate, have feedback, or run into issues: [Wenxuan Zhang](https://isakzhang.github.io/).
+This is an ongoing project. For collaboration, feedback or runtime issues, contact
+[Wenxuan Zhang](https://isakzhang.github.io/).

config.json

@@ -1,133 +1,112 @@
 {
   "architectures": [
-    "MynaHokkienForConditionalGeneration"
+    "Qwen3OmniMoeForConditionalGeneration"
   ],
-  "bridge_config": {
-    "ffn": 6144,
-    "heads": 12,
-    "inner": 1536,
-    "layers": 4,
-    "q25_dim": 2048
-  },
-  "default_language": "nan",
-  "default_speaker": "Ethan",
-  "generation_embedding_config": {
-    "embedding_dim": 2048,
-    "num_embeddings": 152064
-  },
-  "generation_processor_path": "processors/generation",
-  "input_sampling_rate": 16000,
-  "model_type": "mynahokkien",
-  "reasoning_adapter_config": {
-    "alora_invocation_tokens": null,
-    "alpha_pattern": {},
-    "arrow_config": null,
-    "auto_mapping": null,
-    "base_model_name_or_path": "Qwen2.5-Omni-3B/thinker",
-    "bias": "none",
-    "corda_config": null,
-    "ensure_weight_tying": false,
-    "eva_config": null,
-    "exclude_modules": null,
-    "fan_in_fan_out": false,
-    "inference_mode": true,
-    "init_lora_weights": true,
-    "layer_replication": null,
-    "layers_pattern": null,
-    "layers_to_transform": null,
-    "loftq_config": {},
-    "lora_alpha": 32,
-    "lora_bias": false,
-    "lora_dropout": 0.05,
-    "lora_ga_config": null,
-    "megatron_config": null,
-    "megatron_core": "megatron.core",
-    "modules_to_save": null,
-    "peft_type": "LORA",
-    "peft_version": "0.19.1",
-    "qalora_group_size": 16,
-    "r": 16,
-    "rank_pattern": {},
-    "revision": null,
-    "target_modules": [
-      "q_proj",
-      "gate_proj",
-      "up_proj",
-      "down_proj",
-      "k_proj",
-      "v_proj",
-      "o_proj"
+  "assistant_token_id": 77091,
+  "code2wav_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "codebook_dim": 512,
+    "codebook_size": 2048,
+    "decoder_dim": 1536,
+    "dtype": "bfloat16",
+    "hidden_act": "silu",
+    "hidden_size": 1024,
+    "intermediate_size": 3072,
+    "layer_scale_initial_scale": 0.01,
+    "max_position_embeddings": 8000,
+    "model_type": "",
+    "num_attention_heads": 16,
+    "num_hidden_layers": 8,
+    "num_key_value_heads": 16,
+    "num_quantizers": 16,
+    "num_semantic_quantizers": 1,
+    "rms_norm_eps": 1e-05,
+    "rope_theta": 10000,
+    "semantic_codebook_size": 4096,
+    "sliding_window": 72,
+    "upsample_rates": [
+      8,
+      5,
+      4,
+      3
     ],
-    "target_parameters": null,
-    "task_type": "CAUSAL_LM",
-    "trainable_token_indices": null,
-    "use_bdlora": null,
-    "use_dora": false,
-    "use_qalora": false,
-    "use_rslora": false
-  },
-  "reasoning_model_config": {
-    "_attn_implementation_autoset": true,
-    "_name_or_path": "Qwen2.5-Omni-3B/thinker",
-    "architectures": [
-      "Qwen2OmniNaViTThinkerForConditionalGeneration"
+    "upsampling_ratios": [
+      2,
+      2
     ],
-    "audio_config": {
-      "_attn_implementation_autoset": true,
+    "vector_quantization_hidden_dimension": 512
+  },
+  "dtype": "bfloat16",
+  "enable_audio_output": true,
+  "im_end_token_id": 151645,
+  "im_start_token_id": 151644,
+  "model_type": "qwen3_omni_moe",
+  "system_token_id": 8948,
+  "talker_config": {
+    "accept_hidden_layer": 24,
+    "audio_end_token_id": 151670,
+    "audio_start_token_id": 151669,
+    "audio_token_id": 151675,
+    "code_predictor_config": {
       "_name_or_path": "",
-      "activation_dropout": 0.0,
-      "activation_function": "gelu",
       "add_cross_attention": false,
       "architectures": null,
-      "attention_dropout": 0.0,
+      "attention_bias": false,
+      "attention_dropout": 0,
       "bad_words_ids": null,
       "begin_suppress_tokens": null,
       "bos_token_id": null,
       "chunk_size_feed_forward": 0,
       "cross_attention_hidden_size": null,
-      "d_model": 1280,
       "decoder_start_token_id": null,
       "diversity_penalty": 0.0,
       "do_sample": false,
-      "dropout": 0.0,
       "dtype": null,
       "early_stopping": false,
-      "encoder_attention_heads": 20,
-      "encoder_ffn_dim": 5120,
-      "encoder_layerdrop": 0.0,
-      "encoder_layers": 32,
       "encoder_no_repeat_ngram_size": 0,
       "eos_token_id": null,
       "exponential_decay_length_penalty": null,
       "finetuning_task": null,
       "forced_bos_token_id": null,
       "forced_eos_token_id": null,
+      "head_dim": 128,
+      "hidden_act": "silu",
+      "hidden_size": 1024,
       "id2label": {
         "0": "LABEL_0",
         "1": "LABEL_1"
       },
-      "init_std": 0.02,
       "initializer_range": 0.02,
+      "intermediate_size": 3072,
       "is_decoder": false,
       "is_encoder_decoder": false,
       "label2id": {
         "LABEL_0": 0,
         "LABEL_1": 1
       },
+      "layer_types": [
+        "full_attention",
+        "full_attention",
+        "full_attention",
+        "full_attention",
+        "full_attention"
+      ],
       "length_penalty": 1.0,
       "max_length": 20,
-      "max_source_positions": 1500,
+      "max_position_embeddings": 32768,
+      "max_window_layers": 28,
       "min_length": 0,
-      "model_type": "qwen2_5_omni_audio_encoder",
-      "n_window": 100,
+      "model_type": "qwen3_omni_moe_talker_code_predictor",
       "no_repeat_ngram_size": 0,
+      "num_attention_heads": 16,
       "num_beam_groups": 1,
       "num_beams": 1,
-      "num_hidden_layers": 32,
-      "num_mel_bins": 128,
+      "num_code_groups": 16,
+      "num_hidden_layers": 5,
+      "num_key_value_heads": 8,
       "num_return_sequences": 1,
       "output_attentions": false,
-      "output_dim": 2048,
       "output_hidden_states": false,
       "output_scores": false,
       "pad_token_id": null,
@@ -138,45 +117,58 @@
       "repetition_penalty": 1.0,
       "return_dict": true,
       "return_dict_in_generate": false,
-      "scale_embedding": false,
+      "rms_norm_eps": 1e-06,
+      "rope_scaling": null,
+      "rope_theta": 1000000,
       "sep_token_id": null,
+      "sliding_window": null,
       "suppress_tokens": null,
       "task_specific_params": null,
       "temperature": 1.0,
       "tf_legacy_loss": false,
       "tie_encoder_decoder": false,
-      "tie_word_embeddings": true,
+      "tie_word_embeddings": false,
       "tokenizer_class": null,
       "top_k": 50,
       "top_p": 1.0,
       "torchscript": false,
       "typical_p": 1.0,
-      "use_bfloat16": false
+      "use_bfloat16": false,
+      "use_cache": true,
+      "use_sliding_window": false,
+      "vocab_size": 2048
     },
-    "audio_end_token_id": 151648,
-    "audio_start_token_id": 151647,
-    "audio_token_index": 151646,
-    "bos_token_id": 151644,
+    "codec_bos_id": 2149,
+    "codec_eos_token_id": 2150,
+    "codec_nothink_id": 2155,
+    "codec_pad_id": 2148,
+    "codec_think_bos_id": 2156,
+    "codec_think_eos_id": 2157,
     "dtype": "bfloat16",
-    "eos_token_id": 151645,
-    "ignore_index": -100,
-    "image_token_index": 151655,
-    "init_std": 0.02,
-    "initializer_range": 0.02,
-    "model_type": "qwen2_5_omni_thinker",
-    "pad_token_id": 151643,
-    "position_id_per_seconds": 25,
+    "image_token_id": 151655,
+    "model_type": "",
+    "num_code_groups": 16,
+    "output_router_logits": false,
+    "position_id_per_seconds": 13,
     "seconds_per_chunk": 2,
+    "spatial_merge_size": 2,
+    "speaker_id": {
+      "aiden": 2303,
+      "chelsie": 2301,
+      "ethan": 2302
+    },
     "text_config": {
       "_name_or_path": "",
       "add_cross_attention": false,
       "architectures": null,
-      "attention_dropout": 0.0,
+      "attention_bias": false,
+      "attention_dropout": 0,
       "bad_words_ids": null,
       "begin_suppress_tokens": null,
       "bos_token_id": null,
       "chunk_size_feed_forward": 0,
       "cross_attention_hidden_size": null,
+      "decoder_sparse_step": 1,
       "decoder_start_token_id": null,
       "diversity_penalty": 0.0,
       "do_sample": false,
@@ -188,74 +180,41 @@
       "finetuning_task": null,
       "forced_bos_token_id": null,
       "forced_eos_token_id": null,
+      "head_dim": 128,
       "hidden_act": "silu",
-      "hidden_size": 2048,
+      "hidden_size": 1024,
       "id2label": {
         "0": "LABEL_0",
         "1": "LABEL_1"
       },
-      "init_std": 0.02,
       "initializer_range": 0.02,
-      "intermediate_size": 11008,
+      "intermediate_size": 2048,
       "is_decoder": false,
       "is_encoder_decoder": false,
       "label2id": {
         "LABEL_0": 0,
         "LABEL_1": 1
       },
-      "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"
-      ],
       "length_penalty": 1.0,
       "max_length": 20,
-      "max_position_embeddings": 32768,
-      "max_window_layers": 70,
+      "max_position_embeddings": 65536,
       "min_length": 0,
-      "model_type": "qwen2_5_omni_text",
+      "mlp_only_layers": [],
+      "model_type": "qwen3_omni_moe_talker_text",
+      "moe_intermediate_size": 384,
       "no_repeat_ngram_size": 0,
+      "norm_topk_prob": true,
       "num_attention_heads": 16,
       "num_beam_groups": 1,
       "num_beams": 1,
-      "num_hidden_layers": 36,
+      "num_experts": 128,
+      "num_experts_per_tok": 6,
+      "num_hidden_layers": 20,
       "num_key_value_heads": 2,
       "num_return_sequences": 1,
       "output_attentions": false,
       "output_hidden_states": false,
+      "output_router_logits": false,
       "output_scores": false,
       "pad_token_id": null,
       "prefix": null,
@@ -266,29 +225,20 @@
       "return_dict": true,
       "return_dict_in_generate": false,
       "rms_norm_eps": 1e-06,
-      "rope_parameters": {
+      "rope_scaling": {
         "interleaved": true,
-        "mrope_interleaved": true,
         "mrope_section": [
           24,
           20,
           20
         ],
-        "rope_theta": 1000000,
         "rope_type": "default",
         "type": "default"
       },
-      "rope_scaling": {
-        "mrope_section": [
-          16,
-          24,
-          24
-        ],
-        "rope_type": "default",
-        "type": "default"
-      },
-      "rope_theta": 1000000.0,
+      "rope_theta": 1000000,
+      "router_aux_loss_coef": 0.001,
       "sep_token_id": null,
+      "shared_expert_intermediate_size": 768,
       "sliding_window": null,
       "suppress_tokens": null,
       "task_specific_params": null,
@@ -304,52 +254,48 @@
       "use_bfloat16": false,
       "use_cache": true,
       "use_sliding_window": false,
-      "vocab_size": 151936
+      "vocab_size": 3072
     },
-    "tie_word_embeddings": false,
-    "use_cache": true,
-    "user_token_id": 872,
-    "video_token_index": 151656,
-    "vision_config": {
-      "_attn_implementation_autoset": true,
+    "thinker_hidden_size": 2048,
+    "video_token_id": 151656,
+    "vision_start_token_id": 151652
+  },
+  "thinker_config": {
+    "audio_config": {
       "_name_or_path": "",
+      "activation_dropout": 0,
+      "activation_function": "gelu",
       "add_cross_attention": false,
       "architectures": null,
+      "attention_dropout": 0,
       "bad_words_ids": null,
       "begin_suppress_tokens": null,
       "bos_token_id": null,
       "chunk_size_feed_forward": 0,
+      "conv_chunksize": 500,
       "cross_attention_hidden_size": null,
+      "d_model": 1280,
       "decoder_start_token_id": null,
-      "depth": 32,
       "diversity_penalty": 0.0,
       "do_sample": false,
+      "downsample_hidden_size": 480,
+      "dropout": 0,
       "dtype": null,
       "early_stopping": false,
-      "embed_dim": 1280,
+      "encoder_attention_heads": 20,
+      "encoder_ffn_dim": 5120,
+      "encoder_layers": 32,
       "encoder_no_repeat_ngram_size": 0,
       "eos_token_id": null,
       "exponential_decay_length_penalty": null,
       "finetuning_task": null,
       "forced_bos_token_id": null,
       "forced_eos_token_id": null,
-      "fullatt_block_indexes": [
-        7,
-        15,
-        23,
-        31
-      ],
-      "hidden_act": "silu",
-      "hidden_size": 1280,
       "id2label": {
         "0": "LABEL_0",
         "1": "LABEL_1"
       },
-      "in_channels": 3,
-      "in_chans": 3,
-      "init_std": 0.02,
       "initializer_range": 0.02,
-      "intermediate_size": 3420,
       "is_decoder": false,
       "is_encoder_decoder": false,
       "label2id": {
@@ -358,19 +304,22 @@
       },
       "length_penalty": 1.0,
       "max_length": 20,
+      "max_source_positions": 1500,
       "min_length": 0,
-      "model_type": "qwen2_5_omni_vision_encoder",
+      "model_type": "qwen3_omni_moe_audio_encoder",
+      "n_window": 50,
+      "n_window_infer": 800,
       "no_repeat_ngram_size": 0,
       "num_beam_groups": 1,
       "num_beams": 1,
-      "num_heads": 16,
+      "num_hidden_layers": 32,
+      "num_mel_bins": 128,
       "num_return_sequences": 1,
-      "out_hidden_size": 2048,
       "output_attentions": false,
+      "output_dim": 2048,
       "output_hidden_states": false,
       "output_scores": false,
       "pad_token_id": null,
-      "patch_size": 14,
       "prefix": null,
       "problem_type": null,
       "pruned_heads": {},
@@ -378,53 +327,46 @@
       "repetition_penalty": 1.0,
       "return_dict": true,
       "return_dict_in_generate": false,
+      "scale_embedding": false,
       "sep_token_id": null,
-      "spatial_merge_size": 2,
-      "spatial_patch_size": 14,
       "suppress_tokens": null,
       "task_specific_params": null,
       "temperature": 1.0,
-      "temporal_patch_size": 2,
       "tf_legacy_loss": false,
       "tie_encoder_decoder": false,
       "tie_word_embeddings": true,
       "tokenizer_class": null,
-      "tokens_per_second": 25,
       "top_k": 50,
       "top_p": 1.0,
       "torchscript": false,
       "typical_p": 1.0,
-      "use_bfloat16": false,
-      "window_size": 112
+      "use_bfloat16": false
     },
-    "vision_end_token_id": 151653,
-    "vision_start_token_id": 151652,
-    "vision_token_id": 151654
-  },
-  "reasoning_processor_path": "processors/reasoning",
-  "release_epoch": 3,
-  "sampling_rate": 24000,
-  "speech_generator_config": {
-    "accept_hidden_layer": 24,
     "audio_end_token_id": 151670,
     "audio_start_token_id": 151669,
     "audio_token_id": 151675,
-    "audio_tower_config": {},
-    "code_predictor_config": {
+    "dtype": "bfloat16",
+    "image_token_id": 151655,
+    "initializer_range": 0.02,
+    "model_type": "qwen3_omni_moe_thinker",
+    "position_id_per_seconds": 13,
+    "seconds_per_chunk": 2,
+    "text_config": {
       "_name_or_path": "",
       "add_cross_attention": false,
       "architectures": null,
       "attention_bias": false,
-      "attention_dropout": 0,
+      "attention_dropout": 0.0,
       "bad_words_ids": null,
       "begin_suppress_tokens": null,
       "bos_token_id": null,
       "chunk_size_feed_forward": 0,
       "cross_attention_hidden_size": null,
+      "decoder_sparse_step": 1,
       "decoder_start_token_id": null,
       "diversity_penalty": 0.0,
       "do_sample": false,
-      "dtype": "bfloat16",
+      "dtype": null,
       "early_stopping": false,
       "encoder_no_repeat_ngram_size": 0,
       "eos_token_id": null,
@@ -434,42 +376,39 @@
       "forced_eos_token_id": null,
       "head_dim": 128,
       "hidden_act": "silu",
-      "hidden_size": 1024,
+      "hidden_size": 2048,
       "id2label": {
         "0": "LABEL_0",
         "1": "LABEL_1"
       },
       "initializer_range": 0.02,
-      "intermediate_size": 3072,
+      "intermediate_size": 768,
       "is_decoder": false,
       "is_encoder_decoder": false,
       "label2id": {
         "LABEL_0": 0,
         "LABEL_1": 1
       },
-      "layer_types": [
-        "full_attention",
-        "full_attention",
-        "full_attention",
-        "full_attention",
-        "full_attention"
-      ],
       "length_penalty": 1.0,
       "max_length": 20,
-      "max_position_embeddings": 32768,
-      "max_window_layers": 28,
+      "max_position_embeddings": 65536,
       "min_length": 0,
-      "model_type": "qwen3_omni_moe_talker_code_predictor",
+      "mlp_only_layers": [],
+      "model_type": "qwen3_omni_moe_text",
+      "moe_intermediate_size": 768,
       "no_repeat_ngram_size": 0,
-      "num_attention_heads": 16,
+      "norm_topk_prob": true,
+      "num_attention_heads": 32,
       "num_beam_groups": 1,
       "num_beams": 1,
-      "num_code_groups": 16,
-      "num_hidden_layers": 5,
-      "num_key_value_heads": 8,
+      "num_experts": 128,
+      "num_experts_per_tok": 8,
+      "num_hidden_layers": 48,
+      "num_key_value_heads": 4,
       "num_return_sequences": 1,
       "output_attentions": false,
       "output_hidden_states": false,
+      "output_router_logits": false,
       "output_scores": false,
       "pad_token_id": null,
       "prefix": null,
@@ -480,13 +419,21 @@
       "return_dict": true,
       "return_dict_in_generate": false,
       "rms_norm_eps": 1e-06,
-      "rope_parameters": {
-        "rope_theta": 1000000,
-        "rope_type": "default"
+      "rope_scaling": {
+        "interleaved": true,
+        "mrope_interleaved": true,
+        "mrope_section": [
+          24,
+          20,
+          20
+        ],
+        "rope_type": "default",
+        "type": "default"
       },
-      "rope_scaling": null,
-      "rope_theta": 10000,
+      "rope_theta": 1000000,
+      "router_aux_loss_coef": 0.001,
       "sep_token_id": null,
+      "shared_expert_intermediate_size": 0,
       "sliding_window": null,
       "suppress_tokens": null,
       "task_specific_params": null,
@@ -501,46 +448,32 @@
       "typical_p": 1.0,
       "use_bfloat16": false,
       "use_cache": true,
+      "use_qk_norm": true,
       "use_sliding_window": false,
-      "vocab_size": 2048
+      "vocab_size": 152064
     },
-    "codec_bos_id": 2149,
-    "codec_config": {},
-    "codec_eos_token_id": 2150,
-    "codec_nothink_id": 2155,
-    "codec_pad_id": 2148,
-    "codec_think_bos_id": 2156,
-    "codec_think_eos_id": 2157,
-    "dtype": "bfloat16",
-    "image_token_id": 151655,
-    "initializer_range": 0.02,
-    "model_type": "",
-    "num_code_groups": 16,
-    "output_router_logits": false,
-    "position_id_per_seconds": 13,
-    "seconds_per_chunk": 2,
-    "spatial_merge_size": 2,
-    "speaker_id": {
-      "aiden": 2303,
-      "chelsie": 2301,
-      "ethan": 2302
-    },
-    "text_config": {
+    "user_token_id": 872,
+    "video_token_id": 151656,
+    "vision_config": {
       "_name_or_path": "",
       "add_cross_attention": false,
+      "apply_vit_abs_pos_embed": true,
       "architectures": null,
-      "attention_bias": false,
-      "attention_dropout": 0,
       "bad_words_ids": null,
       "begin_suppress_tokens": null,
       "bos_token_id": null,
       "chunk_size_feed_forward": 0,
       "cross_attention_hidden_size": null,
-      "decoder_sparse_step": 1,
       "decoder_start_token_id": null,
+      "deepstack_visual_indexes": [
+        8,
+        16,
+        24
+      ],
+      "depth": 27,
       "diversity_penalty": 0.0,
       "do_sample": false,
-      "dtype": "bfloat16",
+      "dtype": null,
       "early_stopping": false,
       "encoder_no_repeat_ngram_size": 0,
       "eos_token_id": null,
@@ -548,15 +481,17 @@
       "finetuning_task": null,
       "forced_bos_token_id": null,
       "forced_eos_token_id": null,
-      "head_dim": 128,
-      "hidden_act": "silu",
-      "hidden_size": 1024,
+      "hidden_act": "gelu_pytorch_tanh",
+      "hidden_size": 1152,
       "id2label": {
         "0": "LABEL_0",
         "1": "LABEL_1"
       },
+      "image_size": 768,
+      "in_channels": 3,
+      "in_chans": 3,
       "initializer_range": 0.02,
-      "intermediate_size": 2048,
+      "intermediate_size": 4304,
       "is_decoder": false,
       "is_encoder_decoder": false,
       "label2id": {
@@ -565,27 +500,20 @@
       },
       "length_penalty": 1.0,
       "max_length": 20,
-      "max_position_embeddings": 65536,
       "min_length": 0,
-      "mlp_only_layers": [],
-      "model_type": "qwen3_omni_moe_talker_text",
-      "moe_intermediate_size": 384,
+      "model_type": "qwen3_omni_moe_vision_encoder",
       "no_repeat_ngram_size": 0,
-      "norm_topk_prob": true,
-      "num_attention_heads": 16,
       "num_beam_groups": 1,
       "num_beams": 1,
-      "num_experts": 128,
-      "num_experts_per_tok": 6,
-      "num_hidden_layers": 20,
-      "num_key_value_heads": 2,
-      "num_local_experts": 128,
+      "num_heads": 16,
+      "num_position_embeddings": 2304,
       "num_return_sequences": 1,
+      "out_hidden_size": 2048,
       "output_attentions": false,
       "output_hidden_states": false,
-      "output_router_logits": false,
       "output_scores": false,
       "pad_token_id": null,
+      "patch_size": 16,
       "prefix": null,
       "problem_type": null,
       "pruned_heads": {},
@@ -593,93 +521,30 @@
       "repetition_penalty": 1.0,
       "return_dict": true,
       "return_dict_in_generate": false,
-      "rms_norm_eps": 1e-06,
-      "rope_parameters": {
-        "interleaved": true,
-        "mrope_section": [
-          24,
-          20,
-          20
-        ],
-        "rope_theta": 1000000,
-        "rope_type": "default",
-        "type": "default"
-      },
-      "rope_scaling": {
-        "interleaved": true,
-        "mrope_section": [
-          24,
-          20,
-          20
-        ],
-        "rope_type": "default",
-        "type": "default"
-      },
-      "rope_theta": 10000,
-      "router_aux_loss_coef": 0.001,
       "sep_token_id": null,
-      "shared_expert_intermediate_size": 768,
-      "sliding_window": null,
+      "spatial_merge_size": 2,
+      "spatial_patch_size": 16,
       "suppress_tokens": null,
       "task_specific_params": null,
       "temperature": 1.0,
+      "temporal_patch_size": 2,
       "tf_legacy_loss": false,
       "tie_encoder_decoder": false,
-      "tie_word_embeddings": false,
+      "tie_word_embeddings": true,
       "tokenizer_class": null,
+      "tokens_per_second": 2,
       "top_k": 50,
       "top_p": 1.0,
       "torchscript": false,
       "typical_p": 1.0,
-      "use_bfloat16": false,
-      "use_cache": true,
-      "use_sliding_window": false,
-      "vocab_size": 3072
+      "use_bfloat16": false
     },
-    "thinker_hidden_size": 2048,
-    "tie_word_embeddings": false,
-    "video_token_id": 151656,
+    "vision_end_token_id": 151653,
     "vision_start_token_id": 151652
   },
-  "torch_dtype": "float16",
   "transformers_version": "4.57.1",
-  "waveform_decoder_config": {
-    "attention_bias": false,
-    "attention_dropout": 0.0,
-    "codebook_dim": 512,
-    "codebook_size": 2048,
-    "decoder_dim": 1536,
-    "dtype": "bfloat16",
-    "hidden_act": "silu",
-    "hidden_size": 1024,
-    "initializer_range": 0.02,
-    "intermediate_size": 3072,
-    "layer_scale_initial_scale": 0.01,
-    "max_position_embeddings": 8000,
-    "model_type": "",
-    "num_attention_heads": 16,
-    "num_hidden_layers": 8,
-    "num_key_value_heads": 16,
-    "num_quantizers": 16,
-    "num_semantic_quantizers": 1,
-    "rms_norm_eps": 1e-05,
-    "rope_parameters": {
-      "rope_theta": 10000,
-      "rope_type": "default"
-    },
-    "rope_theta": 10000,
-    "semantic_codebook_size": 4096,
-    "sliding_window": 72,
-    "upsample_rates": [
-      8,
-      5,
-      4,
-      3
-    ],
-    "upsampling_ratios": [
-      2,
-      2
-    ],
-    "vector_quantization_hidden_dimension": 512
-  }
+  "tts_bos_token_id": 151672,
+  "tts_eos_token_id": 151673,
+  "tts_pad_token_id": 151671,
+  "user_token_id": 872
 }

configuration_mynahokkien.py

@@ -1,41 +0,0 @@
-from __future__ import annotations
-
-from transformers import PretrainedConfig
-
-
-class MynaHokkienConfig(PretrainedConfig):
-    """Configuration for the unified MynaHokkien inference checkpoint."""
-
-    model_type = "mynahokkien"
-
-    def __init__(
-        self,
-        *,
-        reasoning_model_config=None,
-        reasoning_adapter_config=None,
-        speech_generator_config=None,
-        waveform_decoder_config=None,
-        generation_embedding_config=None,
-        bridge_config=None,
-        reasoning_processor_path="processors/reasoning",
-        generation_processor_path="processors/generation",
-        sampling_rate=24000,
-        input_sampling_rate=16000,
-        default_language="nan",
-        default_speaker="Ethan",
-        **kwargs,
-    ):
-        super().__init__(**kwargs)
-        self.reasoning_model_config = reasoning_model_config or {}
-        self.reasoning_adapter_config = reasoning_adapter_config or {}
-        self.speech_generator_config = speech_generator_config or {}
-        self.waveform_decoder_config = waveform_decoder_config or {}
-        self.generation_embedding_config = generation_embedding_config or {}
-        self.bridge_config = bridge_config or {}
-        self.reasoning_processor_path = reasoning_processor_path
-        self.generation_processor_path = generation_processor_path
-        self.sampling_rate = sampling_rate
-        self.input_sampling_rate = input_sampling_rate
-        self.default_language = default_language
-        self.default_speaker = default_speaker
-

generation_config.json

@@ -1,4 +1,8 @@
 {
-  "max_audio_tokens": 4096,
-  "max_new_tokens": 512
+  "talker_max_new_tokens": 4096,
+  "talker_repetition_penalty": 1.05,
+  "talker_temperature": 0.9,
+  "talker_top_k": 50,
+  "talker_top_p": 1.0,
+  "transformers_version": "4.57.1"
 }

inference.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-"""Simple one-request inference CLI for the unified Myna-Hokkien release."""
+"""Run one request through the released Myna-Hokkien model."""
 from __future__ import annotations
 
 import argparse
@@ -14,16 +14,18 @@ from mynahokkien import MynaHokkien
 def parse_args() -> argparse.Namespace:
     parser = argparse.ArgumentParser()
     parser.add_argument("--model", default="iNLP-Lab/MynaHokkien")
-    modality = parser.add_mutually_exclusive_group(required=True)
-    modality.add_argument("--audio", type=Path)
-    modality.add_argument("--text")
-    parser.add_argument("--prompt")
+    source = parser.add_mutually_exclusive_group(required=True)
+    source.add_argument("--audio", type=Path)
+    source.add_argument("--text")
+    parser.add_argument("--prompt", help="audio-user-turn override")
     parser.add_argument("--output-wav", type=Path, default=Path("output.wav"))
     parser.add_argument("--output-text", type=Path)
     parser.add_argument("--device", default="cuda:0")
+    parser.add_argument("--dtype", choices=("float16", "bfloat16"), default="float16")
     parser.add_argument("--seed", type=int, default=1234)
-    parser.add_argument("--max-new-tokens", type=int, default=512)
+    parser.add_argument("--max-new-tokens", type=int, default=256)
     parser.add_argument("--max-audio-tokens", type=int, default=4096)
+    parser.add_argument("--revision")
     parser.add_argument("--local-files-only", action="store_true")
     return parser.parse_args()
 
@@ -31,14 +33,13 @@ def parse_args() -> argparse.Namespace:
 def main() -> None:
     args = parse_args()
     if args.prompt is not None and args.audio is None:
-        raise ValueError("--prompt is only valid with --audio; include the instruction in --text")
-    if args.audio is not None and not args.audio.is_file():
-        raise FileNotFoundError(args.audio)
-
+        raise ValueError("--prompt is only valid with --audio")
+    dtype = torch.float16 if args.dtype == "float16" else torch.bfloat16
     model = MynaHokkien.from_pretrained(
         args.model,
         device_map=args.device,
-        dtype=torch.float16,
+        dtype=dtype,
+        revision=args.revision,
         local_files_only=args.local_files_only,
     )
     output = model.generate(
@@ -46,20 +47,19 @@ def main() -> None:
         text=args.text,
         prompt=args.prompt,
         language="nan",
+        speaker="Ethan",
         seed=args.seed,
         max_new_tokens=args.max_new_tokens,
         max_audio_tokens=args.max_audio_tokens,
-        return_text=True,
-        return_audio=True,
     )
-    if not output.text or output.audio is None:
-        raise RuntimeError("model did not return both text and audio")
+    if output.text is None or output.audio is None:
+        raise RuntimeError("expected both text and audio")
 
     args.output_wav.parent.mkdir(parents=True, exist_ok=True)
     sf.write(args.output_wav, output.audio, output.sampling_rate)
     if args.output_text is not None:
         args.output_text.parent.mkdir(parents=True, exist_ok=True)
-        args.output_text.write_text(output.text.strip() + "\n", encoding="utf-8")
+        args.output_text.write_text(output.text + "\n", encoding="utf-8")
     print(output.text)
     print(f"[audio] {args.output_wav} ({output.sampling_rate} Hz)")
 

modeling_mynahokkien.py

@@ -1,115 +0,0 @@
-from __future__ import annotations
-
-import inspect
-from pathlib import Path
-
-import torch
-from torch import nn
-from transformers import PreTrainedModel
-
-from configuration_mynahokkien import MynaHokkienConfig
-from mynahokkien.models.bridge1_decoder import Bridge1Decoder
-
-
-class MynaHokkienForConditionalGeneration(PreTrainedModel):
-    """The single module whose state dict is stored in the unified checkpoint."""
-
-    config_class = MynaHokkienConfig
-    base_model_prefix = "mynahokkien"
-    main_input_name = "input_ids"
-    _supports_assign_param_buffer = True
-
-    def __init__(self, config: MynaHokkienConfig):
-        super().__init__(config)
-        from peft import LoraConfig, get_peft_model
-        from transformers.models.qwen3_omni_moe.configuration_qwen3_omni_moe import Qwen3OmniMoeCode2WavConfig
-        from transformers import (
-            Qwen2_5OmniThinkerForConditionalGeneration,
-            Qwen3OmniMoeCode2Wav,
-            Qwen3OmniMoeTalkerForConditionalGeneration,
-        )
-
-        reasoning_config = (
-            Qwen2_5OmniThinkerForConditionalGeneration.config_class.from_dict(
-                config.reasoning_model_config
-            )
-        )
-        reasoning_base = Qwen2_5OmniThinkerForConditionalGeneration(reasoning_config)
-
-        # PEFT config files can be produced by a newer PEFT release than the
-        # runtime. Keep only constructor parameters supported by the pinned
-        # runtime while preserving the actual LoRA topology and scaling.
-        adapter_values = dict(config.reasoning_adapter_config)
-        allowed = set(inspect.signature(LoraConfig.__init__).parameters)
-        adapter_values = {
-            key: value for key, value in adapter_values.items()
-            if key in allowed and key != "self"
-        }
-        adapter_values["inference_mode"] = True
-        self.reasoning_model = get_peft_model(
-            reasoning_base,
-            LoraConfig(**adapter_values),
-        )
-
-        speech_config = Qwen3OmniMoeTalkerForConditionalGeneration.config_class.from_dict(
-            config.speech_generator_config
-        )
-        waveform_config = Qwen3OmniMoeCode2WavConfig.from_dict(
-            config.waveform_decoder_config
-        )
-        self.speech_generator = Qwen3OmniMoeTalkerForConditionalGeneration(speech_config)
-        self.waveform_decoder = Qwen3OmniMoeCode2Wav(waveform_config)
-
-        embedding = config.generation_embedding_config
-        self.generation_embedding = nn.Embedding(
-            int(embedding["num_embeddings"]),
-            int(embedding["embedding_dim"]),
-        )
-        bridge = config.bridge_config
-        self.bridge = Bridge1Decoder(
-            q3_vocab=int(embedding["num_embeddings"]),
-            q3_emb_dim=int(embedding["embedding_dim"]),
-            q25_dim=int(bridge["q25_dim"]),
-            inner=int(bridge["inner"]),
-            heads=int(bridge["heads"]),
-            layers=int(bridge["layers"]),
-            ffn=int(bridge["ffn"]),
-        )
-
-    @classmethod
-    def load_unified(
-        cls,
-        root: Path,
-        *,
-        device: str,
-        dtype: torch.dtype,
-    ) -> "MynaHokkienForConditionalGeneration":
-        """Low-memory load of the root sharded checkpoint onto one device."""
-        from accelerate import init_empty_weights, load_checkpoint_and_dispatch
-
-        config = MynaHokkienConfig.from_pretrained(root, local_files_only=True)
-        with init_empty_weights():
-            model = cls(config)
-        model = load_checkpoint_and_dispatch(
-            model,
-            checkpoint=str(root),
-            device_map={"": device},
-            dtype=None,
-            strict=True,
-        )
-
-        # Match the original release runtime: foundation components use the
-        # requested inference dtype, while Bridge1 remains FP32.
-        model.reasoning_model.to(dtype=dtype)
-        model.speech_generator.to(dtype=dtype)
-        model.waveform_decoder.to(dtype=dtype)
-        model.generation_embedding.to(dtype=dtype)
-        model.bridge.to(dtype=torch.float32)
-        model.eval()
-        return model
-
-    def forward(self, *args, **kwargs):
-        raise NotImplementedError(
-            "Use the high-level MynaHokkien.generate(audio=... | text=...) interface."
-        )
-

myna_hokkien.egg-info/PKG-INFO

@@ -0,0 +1,15 @@
+Metadata-Version: 2.4
+Name: myna-hokkien
+Version: 0.2.0
+Summary: iNLP-Lab Myna-Hokkien native Qwen3-Omni inference runtime
+Project-URL: Homepage, https://huggingface.co/iNLP-Lab/MynaHokkien
+Project-URL: Repository, https://huggingface.co/iNLP-Lab/MynaHokkien
+Requires-Python: >=3.10
+Requires-Dist: torch>=2.6
+Requires-Dist: transformers==4.57.1
+Requires-Dist: accelerate<2,>=1.10
+Requires-Dist: huggingface_hub<2,>=0.36
+Requires-Dist: safetensors>=0.4
+Requires-Dist: numpy<3,>=1.26
+Requires-Dist: librosa<1,>=0.11
+Requires-Dist: soundfile<1,>=0.13

myna_hokkien.egg-info/SOURCES.txt

@@ -0,0 +1,9 @@
+README.md
+pyproject.toml
+myna_hokkien.egg-info/PKG-INFO
+myna_hokkien.egg-info/SOURCES.txt
+myna_hokkien.egg-info/dependency_links.txt
+myna_hokkien.egg-info/requires.txt
+myna_hokkien.egg-info/top_level.txt
+mynahokkien/__init__.py
+mynahokkien/model.py

myna_hokkien.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

myna_hokkien.egg-info/requires.txt

@@ -0,0 +1,8 @@
+torch>=2.6
+transformers==4.57.1
+accelerate<2,>=1.10
+huggingface_hub<2,>=0.36
+safetensors>=0.4
+numpy<3,>=1.26
+librosa<1,>=0.11
+soundfile<1,>=0.13

myna_hokkien.egg-info/top_level.txt

@@ -0,0 +1 @@
+mynahokkien

mynahokkien/__init__.py

@@ -1,15 +1,5 @@
-__all__ = [
-    "MynaHokkien",
-    "MynaHokkienOutput",
-]
+"""Public inference API for the Myna-Hokkien release."""
 
+from .model import MynaHokkien, MynaHokkienOutput
 
-def __getattr__(name):
-    if name in __all__:
-        from .models.mynahokkien import MynaHokkien, MynaHokkienOutput
-
-        return {
-            "MynaHokkien": MynaHokkien,
-            "MynaHokkienOutput": MynaHokkienOutput,
-        }[name]
-    raise AttributeError(name)
+__all__ = ["MynaHokkien", "MynaHokkienOutput"]

mynahokkien/model.py

@@ -0,0 +1,208 @@
+"""Small, stable wrapper around the native Qwen3-Omni release."""
+from __future__ import annotations
+
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Optional, Union
+
+import librosa
+import numpy as np
+import torch
+from transformers import Qwen3OmniMoeForConditionalGeneration, Qwen3OmniMoeProcessor
+
+
+SYSTEM_PROMPT = (
+    "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, "
+    "capable of perceiving auditory inputs and generating text and speech."
+)
+DEFAULT_AUDIO_PROMPT = (
+    "Listen to the spoken Hokkien and reply naturally in concise Singapore Hokkien. "
+    "Always answer in colloquial Singapore Hokkien written in Hanji. Never answer in "
+    "Mandarin or English. Do not repeat or transcribe the input; respond to it directly."
+)
+DEFAULT_TEXT_PROMPT = (
+    "Read the text and reply naturally in concise Singapore Hokkien. Always answer in "
+    "colloquial Singapore Hokkien written in Hanji. Never answer in Mandarin or English. "
+    "Do not repeat the input; respond to it directly."
+)
+
+
+@dataclass
+class MynaHokkienOutput:
+    text: Optional[str]
+    audio: Optional[np.ndarray]
+    sampling_rate: int = 24000
+
+
+class MynaHokkien:
+    """Audio/text question to Singapore-Hokkien text and speech."""
+
+    input_sampling_rate = 16000
+    sampling_rate = 24000
+
+    def __init__(self, model, processor) -> None:
+        self.model = model.eval()
+        self.processor = processor
+        self.last_rendered_prompt: Optional[str] = None
+
+    @classmethod
+    def from_pretrained(
+        cls,
+        model_id_or_path: Union[str, Path],
+        *,
+        device_map: str = "cuda:0",
+        dtype: torch.dtype = torch.float16,
+        token: Optional[str] = None,
+        revision: Optional[str] = None,
+        local_files_only: bool = False,
+        attn_implementation: str = "sdpa",
+    ) -> "MynaHokkien":
+        common = {
+            "token": token,
+            "revision": revision,
+            "local_files_only": local_files_only,
+        }
+        model = Qwen3OmniMoeForConditionalGeneration.from_pretrained(
+            model_id_or_path,
+            device_map=device_map,
+            dtype=dtype,
+            attn_implementation=attn_implementation,
+            **common,
+        )
+        processor = Qwen3OmniMoeProcessor.from_pretrained(model_id_or_path, **common)
+        return cls(model, processor)
+
+    @staticmethod
+    def _check_request(
+        audio: Optional[Union[str, Path]],
+        text: Optional[str],
+        prompt: Optional[str],
+        language: str,
+        speaker: str,
+        return_text: bool,
+        return_audio: bool,
+    ) -> None:
+        if (audio is None) == (text is None):
+            raise ValueError("provide exactly one of audio=... or text=...")
+        if prompt is not None and audio is None:
+            raise ValueError("prompt= is only valid with audio=...")
+        if language != "nan":
+            raise ValueError("this release currently supports language='nan' only")
+        if speaker.lower() != "ethan":
+            raise ValueError("this release currently supports speaker='Ethan' only")
+        if not return_text and not return_audio:
+            raise ValueError("at least one output modality must be requested")
+
+    def _conversation(
+        self,
+        *,
+        audio: Optional[Union[str, Path]],
+        text: Optional[str],
+        prompt: Optional[str],
+    ) -> tuple[list[dict], Optional[np.ndarray]]:
+        waveform = None
+        if audio is not None:
+            audio_path = Path(audio).expanduser()
+            if not audio_path.is_file():
+                raise FileNotFoundError(audio_path)
+            waveform = librosa.load(
+                str(audio_path), sr=self.input_sampling_rate, mono=True
+            )[0].astype(np.float32)
+            user_content = [
+                {"type": "audio", "audio": waveform},
+                {"type": "text", "text": prompt or DEFAULT_AUDIO_PROMPT},
+            ]
+        else:
+            query = text.strip() if text is not None else ""
+            if not query:
+                raise ValueError("text= must not be empty")
+            # This is the exact prefix form used by probe/prompts.json.
+            user_content = [
+                {"type": "text", "text": f"{DEFAULT_TEXT_PROMPT}\n\n{query}"}
+            ]
+        conversation = [
+            {"role": "system", "content": [{"type": "text", "text": SYSTEM_PROMPT}]},
+            {"role": "user", "content": user_content},
+        ]
+        return conversation, waveform
+
+    def generate(
+        self,
+        *,
+        audio: Optional[Union[str, Path]] = None,
+        text: Optional[str] = None,
+        prompt: Optional[str] = None,
+        language: str = "nan",
+        speaker: str = "Ethan",
+        seed: Optional[int] = 1234,
+        max_new_tokens: int = 256,
+        max_audio_tokens: int = 4096,
+        return_text: bool = True,
+        return_audio: bool = True,
+    ) -> MynaHokkienOutput:
+        self._check_request(
+            audio, text, prompt, language, speaker, return_text, return_audio
+        )
+        if seed is not None:
+            torch.manual_seed(seed)
+            if torch.cuda.is_available():
+                torch.cuda.manual_seed_all(seed)
+
+        conversation, waveform = self._conversation(audio=audio, text=text, prompt=prompt)
+        rendered = self.processor.apply_chat_template(
+            conversation, add_generation_prompt=True, tokenize=False
+        )
+        self.last_rendered_prompt = rendered
+        processor_kwargs = {
+            "text": [rendered],
+            "return_tensors": "pt",
+            "padding": True,
+            "use_audio_in_video": False,
+        }
+        if waveform is not None:
+            processor_kwargs.update(
+                {"audio": [waveform], "sampling_rate": self.input_sampling_rate}
+            )
+        inputs = self.processor(**processor_kwargs)
+        inputs = inputs.to(self.model.device).to(self.model.dtype)
+        prompt_tokens = inputs["input_ids"].shape[1]
+
+        generation_kwargs = {
+            **inputs,
+            "speaker": speaker,
+            "return_audio": return_audio,
+            "thinker_do_sample": False,
+            "thinker_return_dict_in_generate": True,
+            "thinker_max_new_tokens": max_new_tokens,
+            "talker_do_sample": False,
+            "talker_max_new_tokens": max_audio_tokens,
+            "use_audio_in_video": False,
+        }
+        with torch.inference_mode():
+            generated = self.model.generate(**generation_kwargs)
+
+        if return_audio:
+            text_result, generated_audio = generated
+        else:
+            text_result, generated_audio = generated, None
+        sequences = (
+            text_result.sequences if hasattr(text_result, "sequences") else text_result
+        )
+        decoded = self.processor.batch_decode(
+            sequences[:, prompt_tokens:],
+            skip_special_tokens=True,
+            clean_up_tokenization_spaces=False,
+        )[0].strip()
+        if not decoded:
+            raise RuntimeError("the Thinker returned an empty answer")
+
+        audio_array = None
+        if generated_audio is not None:
+            audio_array = (
+                generated_audio.reshape(-1).float().cpu().numpy().astype(np.float32)
+            )
+        return MynaHokkienOutput(
+            text=decoded if return_text else None,
+            audio=audio_array,
+            sampling_rate=self.sampling_rate,
+        )

mynahokkien/models/__init__.py

@@ -1,15 +0,0 @@
-__all__ = [
-    "MynaHokkien",
-    "MynaHokkienOutput",
-]
-
-
-def __getattr__(name):
-    if name in __all__:
-        from .mynahokkien import MynaHokkien, MynaHokkienOutput
-
-        return {
-            "MynaHokkien": MynaHokkien,
-            "MynaHokkienOutput": MynaHokkienOutput,
-        }[name]
-    raise AttributeError(name)

mynahokkien/models/bridge1_decoder.py

@@ -1,83 +0,0 @@
-"""Bridge1 runtime used by the standalone MynaHokkien inference package."""
-from __future__ import annotations
-
-import math
-
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-
-
-def sinusoidal_pe(n, dim, device, dtype):
-    pos = torch.arange(n, device=device, dtype=torch.float32).unsqueeze(1)
-    div = torch.exp(torch.arange(0, dim, 2, device=device, dtype=torch.float32) * (-math.log(10000.0) / dim))
-    pe = torch.zeros(n, dim, device=device, dtype=torch.float32)
-    pe[:, 0::2] = torch.sin(pos * div)
-    pe[:, 1::2] = torch.cos(pos * div)
-    return pe.to(dtype)
-
-
-class Bridge1Decoder(nn.Module):
-    def __init__(self, q3_vocab, q3_emb_dim=2048, q25_dim=2048, inner=1536, heads=12, layers=4, ffn=6144):
-        super().__init__()
-        self.in_proj = nn.Linear(q3_emb_dim, inner)
-        self.mem_proj = nn.Linear(q25_dim, inner)
-        self.start = nn.Parameter(torch.zeros(1, 1, inner))
-        nn.init.normal_(self.start, std=0.02)
-        self.blocks = nn.ModuleList(
-            [
-                nn.TransformerDecoderLayer(
-                    d_model=inner,
-                    nhead=heads,
-                    dim_feedforward=ffn,
-                    dropout=0.0,
-                    activation="gelu",
-                    batch_first=True,
-                    norm_first=True,
-                )
-                for _ in range(layers)
-            ]
-        )
-        self.out_norm = nn.LayerNorm(inner)
-        self.out_proj = nn.Linear(inner, q3_emb_dim)
-        self.length_head = nn.Linear(inner, 1)
-
-    def _embed_tokens(self, ids, q3_emb_weight):
-        return self.in_proj(F.embedding(ids, q3_emb_weight).to(self.in_proj.weight.dtype))
-
-    def _logits(self, hidden, q3_emb_weight):
-        projected = self.out_proj(self.out_norm(hidden))
-        return F.linear(projected, q3_emb_weight.to(projected.dtype))
-
-    def _run(self, target, memory):
-        target = target + sinusoidal_pe(
-            target.shape[1], target.shape[-1], target.device, target.dtype
-        ).unsqueeze(0)
-        memory = memory + sinusoidal_pe(
-            memory.shape[1], memory.shape[-1], memory.device, memory.dtype
-        ).unsqueeze(0)
-        causal = torch.triu(
-            torch.full((target.shape[1], target.shape[1]), float("-inf"), device=target.device),
-            diagonal=1,
-        )
-        hidden = target
-        for block in self.blocks:
-            hidden = block(tgt=hidden, memory=memory, tgt_mask=causal)
-        return hidden
-
-    @torch.no_grad()
-    def generate(self, q25_memory, q3_emb_weight, eos_id, max_len=512):
-        memory = self.mem_proj(q25_memory)
-        emitted = []
-        current = self.start.to(memory.dtype)
-        for _ in range(max_len):
-            hidden = self._run(current, memory)
-            token = int(self._logits(hidden[:, -1:], q3_emb_weight)[0, -1].argmax().item())
-            if token == eos_id:
-                break
-            emitted.append(token)
-            next_embed = self._embed_tokens(
-                torch.tensor([[token]], device=memory.device), q3_emb_weight
-            )
-            current = torch.cat([current, next_embed], dim=1)
-        return torch.tensor(emitted, dtype=torch.long, device=memory.device)

mynahokkien/models/mynahokkien.py

@@ -1,314 +0,0 @@
-from __future__ import annotations
-
-from dataclasses import dataclass
-from pathlib import Path
-from typing import Optional, Union
-
-import numpy as np
-import torch
-
-from modeling_mynahokkien import MynaHokkienForConditionalGeneration
-from processing_mynahokkien import MynaHokkienProcessor
-
-
-@dataclass
-class MynaHokkienOutput:
-    text: Optional[str] = None
-    audio: Optional[np.ndarray] = None
-    sampling_rate: int = 24000
-
-
-class MynaHokkien:
-    """Unified inference interface for the implicit Hokkien S2S model."""
-
-    sampling_rate = 24000
-    input_sampling_rate = 16000
-
-    def __init__(self):
-        self.last_text: Optional[str] = None
-
-    @staticmethod
-    def _resolve_model_path(model_id_or_path: Union[str, Path], **hub_kwargs) -> Path:
-        local = Path(model_id_or_path).expanduser()
-        if local.is_dir():
-            return local.resolve()
-        from huggingface_hub import snapshot_download
-
-        return Path(snapshot_download(repo_id=str(model_id_or_path), **hub_kwargs))
-
-    @classmethod
-    def from_pretrained(
-        cls,
-        model_id_or_path: Union[str, Path],
-        *,
-        device_map: str = "cuda:0",
-        dtype: torch.dtype = torch.float16,
-        token: Optional[str] = None,
-        revision: Optional[str] = None,
-        local_files_only: bool = False,
-    ) -> "MynaHokkien":
-        root = cls._resolve_model_path(
-            model_id_or_path,
-            token=token,
-            revision=revision,
-            local_files_only=local_files_only,
-        )
-        self = cls()
-        self.root = root
-        self.device = torch.device(device_map)
-        self.dtype = dtype
-        self.core = MynaHokkienForConditionalGeneration.load_unified(
-            root,
-            device=device_map,
-            dtype=dtype,
-        )
-        self.sampling_rate = int(self.core.config.sampling_rate)
-        self.input_sampling_rate = int(self.core.config.input_sampling_rate)
-        processors = MynaHokkienProcessor.from_pretrained(root)
-        self.q25_processor = processors.reasoning_processor
-        self.q3_processor = processors.generation_processor
-        from transformers import Qwen3OmniMoeConfig
-
-        q3_processor_path = root / self.core.config.generation_processor_path
-        self.q3_omni_config = Qwen3OmniMoeConfig.from_pretrained(
-            q3_processor_path,
-            local_files_only=True,
-        )
-
-        self.q25 = self.core.reasoning_model
-        self.talker = self.core.speech_generator
-        self.code2wav = self.core.waveform_decoder
-        self.q3_embedding = self.core.generation_embedding
-        self.bridge = self.core.bridge
-        self.q3_config = self.q3_processor_config = self.q3_omni_config.to_dict()
-
-        self.q25.eval()
-        self.talker.eval()
-        self.code2wav.eval()
-        self.bridge.eval()
-        return self
-
-    def _q25_answer_memory(
-        self,
-        text: Optional[str],
-        audio: Optional[Union[str, Path]],
-        prompt: Optional[str],
-        max_new_tokens: int,
-    ):
-        system = (
-            "You are a helpful Singapore Hokkien speaker. Always reply in colloquial Singapore "
-            "Hokkien written in Hanji. Keep it short."
-        )
-        sg_prompt = (
-            "Answer or respond to what is said in the audio, in spoken Singapore Hokkien. "
-            "Do NOT repeat or transcribe it."
-        )
-        wav = None
-        if audio is not None:
-            import librosa
-
-            wav = librosa.load(str(audio), sr=self.input_sampling_rate, mono=True)[0].astype(np.float32)
-            content = [{"type": "audio", "audio": wav}, {"type": "text", "text": prompt or sg_prompt}]
-        elif text:
-            content = [{"type": "text", "text": text}]
-        else:
-            raise ValueError("provide exactly one of audio=... or text=...")
-
-        conversation = [
-            {"role": "system", "content": [{"type": "text", "text": system}]},
-            {"role": "user", "content": content},
-        ]
-        rendered = self.q25_processor.apply_chat_template(
-            conversation, add_generation_prompt=True, tokenize=False
-        )
-        if wav is None:
-            inputs = self.q25_processor(text=[rendered], return_tensors="pt", padding=True)
-        else:
-            inputs = self.q25_processor(
-                text=[rendered],
-                audio=[wav],
-                sampling_rate=self.input_sampling_rate,
-                return_tensors="pt",
-                padding=True,
-            )
-        inputs = {
-            key: value.to(self.device) if torch.is_tensor(value) else value
-            for key, value in inputs.items()
-        }
-        prompt_len = inputs["input_ids"].shape[1]
-        with torch.no_grad():
-            generated = self.q25.generate(
-                **inputs, do_sample=False, max_new_tokens=max_new_tokens
-            )
-        answer_ids = generated[0, prompt_len:].tolist()
-        eos = self.q25_processor.tokenizer.eos_token_id
-        if eos in answer_ids:
-            answer_ids = answer_ids[: answer_ids.index(eos)]
-        if not answer_ids:
-            raise RuntimeError("Q2.5 thinker returned an empty answer")
-        self.last_text = self.q25_processor.tokenizer.decode(
-            answer_ids, skip_special_tokens=True
-        ).strip()
-        answer = torch.tensor([answer_ids], device=self.device)
-        embedding = self.q25.get_input_embeddings()
-        if hasattr(embedding, "base_layer"):
-            embedding = embedding.base_layer
-        memory = embedding(answer).to(dtype=torch.float32)
-        return memory
-
-    def _assistant_part(self, start, end, speaker_id, hidden, pad, bos, eos):
-        cfg = self.q3_config
-        talker_cfg = cfg["talker_config"]
-        assistant_hidden = self.talker.text_projection(hidden[:, start:end]).to(self.device)
-        text_hidden = torch.cat(
-            (assistant_hidden[:, :3], pad.expand(-1, 4, -1), bos, assistant_hidden[:, 3:4]),
-            dim=1,
-        )
-        special_ids = torch.tensor(
-            [[
-                talker_cfg["codec_nothink_id"],
-                talker_cfg["codec_think_bos_id"],
-                talker_cfg["codec_think_eos_id"],
-                speaker_id,
-                talker_cfg["codec_pad_id"],
-                talker_cfg["codec_bos_id"],
-            ]],
-            device=self.device,
-        )
-        codec_hidden = torch.cat(
-            (
-                torch.zeros((1, 3, text_hidden.shape[-1]), device=self.device, dtype=self.dtype),
-                self.talker.get_input_embeddings()(special_ids).to(self.device),
-            ),
-            dim=1,
-        )
-        trailing = torch.cat((assistant_hidden[:, 4:], eos), dim=1)
-        input_ids = torch.full(
-            (1, text_hidden.shape[1]),
-            fill_value=cfg["tts_pad_token_id"],
-            dtype=torch.long,
-            device=self.device,
-        )
-        return text_hidden + codec_hidden, input_ids, trailing
-
-    def _speak(self, predicted_ids: list[int], speaker: str, max_audio_tokens: int) -> np.ndarray:
-        cfg = self.q3_config
-        tok = self.q3_processor.tokenizer
-        tts_system = (
-            "You are a high-quality Text-to-Speech (TTS) model. Your task is to convert text "
-            "into natural, fluent, and realistic speech."
-        )
-        tts_user = "Read this Hokkien text aloud naturally and clearly."
-        head = self.q3_processor.apply_chat_template(
-            [{"role": "system", "content": tts_system}, {"role": "user", "content": tts_user}],
-            add_generation_prompt=True,
-            tokenize=False,
-        )
-        head_ids = tok(head, add_special_tokens=False)["input_ids"]
-        full_ids = torch.tensor(
-            [head_ids + predicted_ids + [cfg["im_end_token_id"]]],
-            device=self.device,
-        )
-        hidden = self.q3_embedding(full_ids).to(dtype=self.dtype)
-        special = torch.tensor(
-            [[cfg["tts_bos_token_id"], cfg["tts_eos_token_id"], cfg["tts_pad_token_id"]]],
-            device=self.device,
-        )
-        bos, eos, pad = self.talker.text_projection(self.q3_embedding(special)).chunk(3, dim=1)
-        speaker_id = cfg["talker_config"]["speaker_id"].get(speaker.lower())
-        if speaker_id is None:
-            raise ValueError(f"unknown speaker {speaker!r}")
-
-        starts = torch.nonzero(full_ids[0] == cfg["im_start_token_id"]).view(-1)
-        ends = torch.cat([starts, torch.tensor([full_ids.shape[1]], device=self.device)])
-        parts, id_parts, trailing = [], [], None
-        for i in range(len(ends) - 1):
-            start, end = int(ends[i]), int(ends[i + 1])
-            role = int(full_ids[0, start + 1])
-            if role == cfg["user_token_id"]:
-                parts.append(self.talker.text_projection(hidden[:, start:end]))
-                id_parts.append(full_ids[:, start:end])
-            elif role == cfg["assistant_token_id"] and i == len(ends) - 2:
-                part, ids, trailing = self._assistant_part(
-                    start, end, speaker_id, hidden, pad, bos, eos
-                )
-                parts.append(part)
-                id_parts.append(ids)
-        if trailing is None:
-            raise RuntimeError("failed to construct Q3 assistant Talker prefix")
-
-        talker_cfg = cfg["talker_config"]
-        vocab = talker_cfg["text_config"]["vocab_size"]
-        codec_eos = talker_cfg["codec_eos_token_id"]
-        suppress = [i for i in range(vocab - 1024, vocab) if i != codec_eos]
-        with torch.no_grad():
-            result = self.talker.generate(
-                inputs_embeds=torch.cat(parts, dim=1),
-                trailing_text_hidden=trailing,
-                tts_pad_embed=pad,
-                talker_input_ids=torch.cat(id_parts, dim=1),
-                max_new_tokens=max_audio_tokens,
-                do_sample=True,
-                top_k=50,
-                top_p=1.0,
-                temperature=0.9,
-                repetition_penalty=1.05,
-                eos_token_id=codec_eos,
-                suppress_tokens=suppress,
-                output_hidden_states=True,
-                return_dict_in_generate=True,
-            )
-            codes = torch.stack(
-                [states[-1] for states in result.hidden_states if states[-1] is not None],
-                dim=1,
-            ).transpose(1, 2)
-            wav = self.code2wav.chunked_decode(
-                codes.to(next(self.code2wav.parameters()).device),
-                chunk_size=300,
-                left_context_size=25,
-            )
-        return wav.float().squeeze().cpu().numpy().astype(np.float32)
-
-    def generate(
-        self,
-        *,
-        audio: Optional[Union[str, Path]] = None,
-        text: Optional[str] = None,
-        language: str = "nan",
-        speaker: str = "Ethan",
-        prompt: Optional[str] = None,
-        max_new_tokens: int = 512,
-        max_audio_tokens: int = 4096,
-        seed: Optional[int] = None,
-        return_text: bool = True,
-        return_audio: bool = True,
-    ) -> MynaHokkienOutput:
-        if language != "nan":
-            raise ValueError("this release currently supports language='nan' only")
-        if audio is not None and text is not None:
-            raise ValueError("audio and text are mutually exclusive")
-        if prompt is not None and audio is None:
-            raise ValueError("prompt= is only used with audio=; include instructions directly in text=")
-        if not return_text and not return_audio:
-            raise ValueError("at least one of return_text or return_audio must be True")
-        if seed is not None:
-            torch.manual_seed(seed)
-            if torch.cuda.is_available():
-                torch.cuda.manual_seed_all(seed)
-        memory = self._q25_answer_memory(text, audio, prompt, max_new_tokens)
-        reasoning_text = self.last_text
-        q3_tok = self.q3_processor.tokenizer
-        predicted = self.bridge.generate(
-            memory,
-            self.q3_embedding.weight,
-            eos_id=q3_tok.eos_token_id,
-            max_len=max_new_tokens,
-        ).tolist()
-        if not predicted:
-            raise RuntimeError("Bridge1 returned an empty Q3 token sequence")
-        waveform = self._speak(predicted, speaker, max_audio_tokens) if return_audio else None
-        return MynaHokkienOutput(
-            text=reasoning_text if return_text else None,
-            audio=waveform,
-            sampling_rate=self.sampling_rate,
-        )

processing_mynahokkien.py

@@ -1,30 +0,0 @@
-from __future__ import annotations
-
-from pathlib import Path
-
-from configuration_mynahokkien import MynaHokkienConfig
-
-
-class MynaHokkienProcessor:
-    """Private two-processor implementation behind the unified public API."""
-
-    def __init__(self, reasoning_processor, generation_processor):
-        self.reasoning_processor = reasoning_processor
-        self.generation_processor = generation_processor
-
-    @classmethod
-    def from_pretrained(cls, root: str | Path):
-        from transformers import Qwen2_5OmniProcessor, Qwen3OmniMoeProcessor
-
-        root = Path(root)
-        config = MynaHokkienConfig.from_pretrained(root, local_files_only=True)
-        reasoning = Qwen2_5OmniProcessor.from_pretrained(
-            root / config.reasoning_processor_path,
-            local_files_only=True,
-        )
-        generation = Qwen3OmniMoeProcessor.from_pretrained(
-            root / config.generation_processor_path,
-            local_files_only=True,
-        )
-        return cls(reasoning, generation)
-

processor_config.json

@@ -1,5 +0,0 @@
-{
-  "generation_processor_path": "processors/generation",
-  "processor_class": "MynaHokkienProcessor",
-  "reasoning_processor_path": "processors/reasoning"
-}

processors/generation/config.json

@@ -1,589 +0,0 @@
-{
-  "architectures": [
-    "Qwen3OmniMoeForConditionalGeneration"
-  ],
-  "assistant_token_id": 77091,
-  "code2wav_config": {
-    "attention_bias": false,
-    "attention_dropout": 0.0,
-    "codebook_dim": 512,
-    "codebook_size": 2048,
-    "decoder_dim": 1536,
-    "dtype": "bfloat16",
-    "hidden_act": "silu",
-    "hidden_size": 1024,
-    "initializer_range": 0.02,
-    "intermediate_size": 3072,
-    "layer_scale_initial_scale": 0.01,
-    "max_position_embeddings": 8000,
-    "model_type": "",
-    "num_attention_heads": 16,
-    "num_hidden_layers": 8,
-    "num_key_value_heads": 16,
-    "num_quantizers": 16,
-    "num_semantic_quantizers": 1,
-    "rms_norm_eps": 1e-05,
-    "rope_parameters": {
-      "rope_theta": 10000,
-      "rope_type": "default"
-    },
-    "rope_theta": 10000,
-    "semantic_codebook_size": 4096,
-    "sliding_window": 72,
-    "upsample_rates": [
-      8,
-      5,
-      4,
-      3
-    ],
-    "upsampling_ratios": [
-      2,
-      2
-    ],
-    "vector_quantization_hidden_dimension": 512
-  },
-  "dtype": "bfloat16",
-  "enable_audio_output": true,
-  "im_end_token_id": 151645,
-  "im_start_token_id": 151644,
-  "initializer_range": 0.02,
-  "model_type": "qwen3_omni_moe",
-  "system_token_id": 8948,
-  "talker_config": {
-    "accept_hidden_layer": 24,
-    "audio_end_token_id": 151670,
-    "audio_start_token_id": 151669,
-    "audio_token_id": 151675,
-    "audio_tower_config": {},
-    "code_predictor_config": {
-      "_name_or_path": "",
-      "add_cross_attention": false,
-      "architectures": null,
-      "attention_bias": false,
-      "attention_dropout": 0,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "cross_attention_hidden_size": null,
-      "decoder_start_token_id": null,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "dtype": "bfloat16",
-      "early_stopping": false,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "head_dim": 128,
-      "hidden_act": "silu",
-      "hidden_size": 1024,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "initializer_range": 0.02,
-      "intermediate_size": 3072,
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "layer_types": [
-        "full_attention",
-        "full_attention",
-        "full_attention",
-        "full_attention",
-        "full_attention"
-      ],
-      "length_penalty": 1.0,
-      "max_length": 20,
-      "max_position_embeddings": 32768,
-      "max_window_layers": 28,
-      "min_length": 0,
-      "model_type": "qwen3_omni_moe_talker_code_predictor",
-      "no_repeat_ngram_size": 0,
-      "num_attention_heads": 16,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_code_groups": 16,
-      "num_hidden_layers": 5,
-      "num_key_value_heads": 8,
-      "num_return_sequences": 1,
-      "output_attentions": false,
-      "output_hidden_states": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repetition_penalty": 1.0,
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "rms_norm_eps": 1e-06,
-      "rope_parameters": {
-        "rope_theta": 1000000,
-        "rope_type": "default"
-      },
-      "rope_scaling": null,
-      "rope_theta": 10000,
-      "sep_token_id": null,
-      "sliding_window": null,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": false,
-      "tokenizer_class": null,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "use_bfloat16": false,
-      "use_cache": true,
-      "use_sliding_window": false,
-      "vocab_size": 2048
-    },
-    "codec_bos_id": 2149,
-    "codec_config": {},
-    "codec_eos_token_id": 2150,
-    "codec_nothink_id": 2155,
-    "codec_pad_id": 2148,
-    "codec_think_bos_id": 2156,
-    "codec_think_eos_id": 2157,
-    "dtype": "bfloat16",
-    "image_token_id": 151655,
-    "initializer_range": 0.02,
-    "model_type": "",
-    "num_code_groups": 16,
-    "output_router_logits": false,
-    "position_id_per_seconds": 13,
-    "seconds_per_chunk": 2,
-    "spatial_merge_size": 2,
-    "speaker_id": {
-      "aiden": 2303,
-      "chelsie": 2301,
-      "ethan": 2302
-    },
-    "text_config": {
-      "_name_or_path": "",
-      "add_cross_attention": false,
-      "architectures": null,
-      "attention_bias": false,
-      "attention_dropout": 0,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "cross_attention_hidden_size": null,
-      "decoder_sparse_step": 1,
-      "decoder_start_token_id": null,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "dtype": "bfloat16",
-      "early_stopping": false,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "head_dim": 128,
-      "hidden_act": "silu",
-      "hidden_size": 1024,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "initializer_range": 0.02,
-      "intermediate_size": 2048,
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "length_penalty": 1.0,
-      "max_length": 20,
-      "max_position_embeddings": 65536,
-      "min_length": 0,
-      "mlp_only_layers": [],
-      "model_type": "qwen3_omni_moe_talker_text",
-      "moe_intermediate_size": 384,
-      "no_repeat_ngram_size": 0,
-      "norm_topk_prob": true,
-      "num_attention_heads": 16,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_experts": 128,
-      "num_experts_per_tok": 6,
-      "num_hidden_layers": 20,
-      "num_key_value_heads": 2,
-      "num_local_experts": 128,
-      "num_return_sequences": 1,
-      "output_attentions": false,
-      "output_hidden_states": false,
-      "output_router_logits": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repetition_penalty": 1.0,
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "rms_norm_eps": 1e-06,
-      "rope_parameters": {
-        "interleaved": true,
-        "mrope_section": [
-          24,
-          20,
-          20
-        ],
-        "rope_theta": 1000000,
-        "rope_type": "default",
-        "type": "default"
-      },
-      "rope_scaling": {
-        "interleaved": true,
-        "mrope_section": [
-          24,
-          20,
-          20
-        ],
-        "rope_type": "default",
-        "type": "default"
-      },
-      "rope_theta": 10000,
-      "router_aux_loss_coef": 0.001,
-      "sep_token_id": null,
-      "shared_expert_intermediate_size": 768,
-      "sliding_window": null,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": false,
-      "tokenizer_class": null,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "use_bfloat16": false,
-      "use_cache": true,
-      "use_sliding_window": false,
-      "vocab_size": 3072
-    },
-    "thinker_hidden_size": 2048,
-    "tie_word_embeddings": false,
-    "video_token_id": 151656,
-    "vision_start_token_id": 151652
-  },
-  "thinker_config": {
-    "audio_config": {
-      "_name_or_path": "",
-      "activation_dropout": 0,
-      "activation_function": "gelu",
-      "add_cross_attention": false,
-      "architectures": null,
-      "attention_dropout": 0,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "conv_chunksize": 500,
-      "cross_attention_hidden_size": null,
-      "d_model": 1280,
-      "decoder_start_token_id": null,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "downsample_hidden_size": 480,
-      "dropout": 0,
-      "dtype": "bfloat16",
-      "early_stopping": false,
-      "encoder_attention_heads": 20,
-      "encoder_ffn_dim": 5120,
-      "encoder_layers": 32,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "initializer_range": 0.02,
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "length_penalty": 1.0,
-      "max_length": 20,
-      "max_source_positions": 1500,
-      "min_length": 0,
-      "model_type": "qwen3_omni_moe_audio_encoder",
-      "n_window": 50,
-      "n_window_infer": 800,
-      "no_repeat_ngram_size": 0,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_hidden_layers": 32,
-      "num_mel_bins": 128,
-      "num_return_sequences": 1,
-      "output_attentions": false,
-      "output_dim": 2048,
-      "output_hidden_states": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repetition_penalty": 1.0,
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "scale_embedding": false,
-      "sep_token_id": null,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": true,
-      "tokenizer_class": null,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "use_bfloat16": false
-    },
-    "audio_end_token_id": 151670,
-    "audio_start_token_id": 151669,
-    "audio_token_id": 151675,
-    "dtype": "bfloat16",
-    "image_token_id": 151655,
-    "initializer_range": 0.02,
-    "model_type": "qwen3_omni_moe_thinker",
-    "position_id_per_seconds": 13,
-    "seconds_per_chunk": 2,
-    "text_config": {
-      "_name_or_path": "",
-      "add_cross_attention": false,
-      "architectures": null,
-      "attention_bias": false,
-      "attention_dropout": 0.0,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "cross_attention_hidden_size": null,
-      "decoder_sparse_step": 1,
-      "decoder_start_token_id": null,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "dtype": "bfloat16",
-      "early_stopping": false,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "head_dim": 128,
-      "hidden_act": "silu",
-      "hidden_size": 2048,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "initializer_range": 0.02,
-      "intermediate_size": 768,
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "length_penalty": 1.0,
-      "max_length": 20,
-      "max_position_embeddings": 65536,
-      "min_length": 0,
-      "mlp_only_layers": [],
-      "model_type": "qwen3_omni_moe_text",
-      "moe_intermediate_size": 768,
-      "no_repeat_ngram_size": 0,
-      "norm_topk_prob": true,
-      "num_attention_heads": 32,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_experts": 128,
-      "num_experts_per_tok": 8,
-      "num_hidden_layers": 48,
-      "num_key_value_heads": 4,
-      "num_return_sequences": 1,
-      "output_attentions": false,
-      "output_hidden_states": false,
-      "output_router_logits": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repetition_penalty": 1.0,
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "rms_norm_eps": 1e-06,
-      "rope_parameters": {
-        "interleaved": true,
-        "mrope_interleaved": true,
-        "mrope_section": [
-          24,
-          20,
-          20
-        ],
-        "rope_theta": 1000000,
-        "rope_type": "default",
-        "type": "default"
-      },
-      "rope_scaling": {
-        "interleaved": true,
-        "mrope_interleaved": true,
-        "mrope_section": [
-          24,
-          20,
-          20
-        ],
-        "rope_type": "default",
-        "type": "default"
-      },
-      "rope_theta": 1000000.0,
-      "router_aux_loss_coef": 0.001,
-      "sep_token_id": null,
-      "shared_expert_intermediate_size": 0,
-      "sliding_window": null,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": false,
-      "tokenizer_class": null,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "use_bfloat16": false,
-      "use_cache": true,
-      "use_qk_norm": true,
-      "use_sliding_window": false,
-      "vocab_size": 152064
-    },
-    "tie_word_embeddings": false,
-    "user_token_id": 872,
-    "video_token_id": 151656,
-    "vision_config": {
-      "_name_or_path": "",
-      "add_cross_attention": false,
-      "apply_vit_abs_pos_embed": true,
-      "architectures": null,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "cross_attention_hidden_size": null,
-      "decoder_start_token_id": null,
-      "deepstack_visual_indexes": [
-        8,
-        16,
-        24
-      ],
-      "depth": 27,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "dtype": "bfloat16",
-      "early_stopping": false,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "hidden_act": "gelu_pytorch_tanh",
-      "hidden_size": 1152,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "image_size": 768,
-      "in_channels": 3,
-      "in_chans": 3,
-      "initializer_range": 0.02,
-      "intermediate_size": 4304,
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "length_penalty": 1.0,
-      "max_length": 20,
-      "min_length": 0,
-      "model_type": "qwen3_omni_moe_vision_encoder",
-      "no_repeat_ngram_size": 0,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_heads": 16,
-      "num_position_embeddings": 2304,
-      "num_return_sequences": 1,
-      "out_hidden_size": 2048,
-      "output_attentions": false,
-      "output_hidden_states": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "patch_size": 16,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repetition_penalty": 1.0,
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "sep_token_id": null,
-      "spatial_merge_size": 2,
-      "spatial_patch_size": 16,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "temporal_patch_size": 2,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": true,
-      "tokenizer_class": null,
-      "tokens_per_second": 2,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "use_bfloat16": false
-    },
-    "vision_end_token_id": 151653,
-    "vision_start_token_id": 151652
-  },
-  "transformers_version": "4.57.1",
-  "tts_bos_token_id": 151672,
-  "tts_eos_token_id": 151673,
-  "tts_pad_token_id": 151671,
-  "user_token_id": 872
-}

processors/generation/generation_config.json

@@ -1,8 +0,0 @@
-{
-  "talker_max_new_tokens": 4096,
-  "talker_repetition_penalty": 1.05,
-  "talker_temperature": 0.9,
-  "talker_top_k": 50,
-  "talker_top_p": 1.0,
-  "transformers_version": "4.57.1"
-}

processors/generation/processor_config.json

@@ -1,109 +0,0 @@
-{
-  "feature_extractor": {
-    "chunk_length": 30,
-    "dither": 0.0,
-    "feature_extractor_type": "WhisperFeatureExtractor",
-    "feature_size": 128,
-    "hop_length": 160,
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_processor_type": "Qwen2VLImageProcessor",
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "max_pixels": 12845056,
-    "merge_size": 2,
-    "min_pixels": 3136,
-    "n_fft": 400,
-    "n_samples": 480000,
-    "nb_max_frames": 3000,
-    "padding_side": "right",
-    "padding_value": 0.0,
-    "patch_size": 16,
-    "return_attention_mask": true,
-    "sampling_rate": 16000,
-    "temporal_patch_size": 2
-  },
-  "image_processor": {
-    "dither": 0.0,
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "feature_size": 128,
-    "hop_length": 160,
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_processor_type": "Qwen2VLImageProcessor",
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "merge_size": 2,
-    "n_fft": 400,
-    "n_samples": 4800000,
-    "nb_max_frames": 30000,
-    "padding_side": "right",
-    "padding_value": 0.0,
-    "patch_size": 16,
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "return_attention_mask": true,
-    "sampling_rate": 16000,
-    "size": {
-      "longest_edge": 12845056,
-      "shortest_edge": 3136
-    },
-    "temporal_patch_size": 2
-  },
-  "processor_class": "Qwen3OmniMoeProcessor",
-  "video_processor": {
-    "dither": 0.0,
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "do_sample_frames": false,
-    "feature_size": 128,
-    "hop_length": 160,
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "max_frames": 768,
-    "merge_size": 2,
-    "min_frames": 4,
-    "n_fft": 400,
-    "n_samples": 4800000,
-    "nb_max_frames": 30000,
-    "padding_side": "right",
-    "padding_value": 0.0,
-    "patch_size": 16,
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "return_attention_mask": true,
-    "return_metadata": false,
-    "sampling_rate": 16000,
-    "size": {
-      "longest_edge": 12845056,
-      "shortest_edge": 3136
-    },
-    "temporal_patch_size": 2,
-    "video_processor_type": "Qwen2VLVideoProcessor"
-  }
-}

processors/reasoning/added_tokens.json

@@ -1,24 +0,0 @@
-{
-  "</tool_call>": 151658,
-  "<tool_call>": 151657,
-  "<|AUDIO|>": 151646,
-  "<|IMAGE|>": 151655,
-  "<|VIDEO|>": 151656,
-  "<|audio_bos|>": 151647,
-  "<|audio_eos|>": 151648,
-  "<|box_end|>": 151649,
-  "<|endoftext|>": 151643,
-  "<|file_sep|>": 151664,
-  "<|fim_middle|>": 151660,
-  "<|fim_pad|>": 151662,
-  "<|fim_prefix|>": 151659,
-  "<|fim_suffix|>": 151661,
-  "<|im_end|>": 151645,
-  "<|im_start|>": 151644,
-  "<|quad_end|>": 151651,
-  "<|quad_start|>": 151650,
-  "<|repo_name|>": 151663,
-  "<|vision_bos|>": 151652,
-  "<|vision_eos|>": 151653,
-  "<|vision_pad|>": 151654
-}

processors/reasoning/chat_template.jinja

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

processors/reasoning/chat_template.json

@@ -1,3 +0,0 @@
-{
-  "chat_template": "{% set audio_count = namespace(value=0) %}{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_bos|><|IMAGE|><|vision_eos|>{% elif content['type'] == 'audio' or 'audio' in content or 'audio_url' in content %}{% set audio_count.value = audio_count.value + 1 %}{% if add_audio_id %}Audio {{ audio_count.value }}: {% endif %}<|audio_bos|><|AUDIO|><|audio_eos|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_bos|><|VIDEO|><|vision_eos|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}"
-}

processors/reasoning/config.json

@@ -1,650 +0,0 @@
-{
-  "architectures": [
-    "Qwen2_5OmniForConditionalGeneration"
-  ],
-  "dtype": "bfloat16",
-  "enable_audio_output": false,
-  "enable_talker": true,
-  "model_type": "qwen2_5_omni",
-  "talker_config": {
-    "_attn_implementation_autoset": true,
-    "_name_or_path": "Qwen2.5-Omni-3B/talker",
-    "architectures": [
-      "Qwen2OmniTalkerForConditionalGeneration"
-    ],
-    "attention_dropout": 0.0,
-    "audio_end_token_id": 151648,
-    "audio_start_token_id": 151647,
-    "audio_token_index": 151646,
-    "dtype": "bfloat16",
-    "embedding_size": 2048,
-    "head_dim": 64,
-    "hidden_act": "silu",
-    "hidden_size": 896,
-    "image_token_index": 151655,
-    "init_std": 0.02,
-    "initializer_range": 0.02,
-    "intermediate_size": 4864,
-    "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"
-    ],
-    "max_position_embeddings": 32768,
-    "max_window_layers": 28,
-    "model_type": "qwen2_5_omni_talker",
-    "num_attention_heads": 14,
-    "num_hidden_layers": 24,
-    "num_key_value_heads": 2,
-    "position_id_per_seconds": 25,
-    "rms_norm_eps": 1e-06,
-    "rope_scaling": {
-      "mrope_section": [
-        16,
-        16,
-        0
-      ],
-      "rope_type": "default",
-      "type": "default"
-    },
-    "rope_theta": 1000000.0,
-    "seconds_per_chunk": 2,
-    "sliding_window": null,
-    "spatial_merge_size": 2,
-    "tts_codec_end_token_id": 8294,
-    "tts_codec_mask_token_id": 8296,
-    "tts_codec_pad_token_id": 8292,
-    "tts_codec_start_token_id": 8293,
-    "tts_text_end_token_id": 151861,
-    "tts_text_pad_token_id": 151859,
-    "tts_text_start_token_id": 151860,
-    "use_cache": true,
-    "use_sliding_window": false,
-    "video_token_index": 151656,
-    "vision_end_token_id": 151653,
-    "vision_start_token_id": 151652,
-    "vocab_size": 8448
-  },
-  "thinker_config": {
-    "_attn_implementation_autoset": true,
-    "_name_or_path": "Qwen2.5-Omni-3B/thinker",
-    "architectures": [
-      "Qwen2OmniNaViTThinkerForConditionalGeneration"
-    ],
-    "audio_config": {
-      "_attn_implementation_autoset": true,
-      "_name_or_path": "",
-      "activation_dropout": 0.0,
-      "activation_function": "gelu",
-      "add_cross_attention": false,
-      "architectures": null,
-      "attention_dropout": 0.0,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "cross_attention_hidden_size": null,
-      "d_model": 1280,
-      "decoder_start_token_id": null,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "dropout": 0.0,
-      "dtype": null,
-      "early_stopping": false,
-      "encoder_attention_heads": 20,
-      "encoder_ffn_dim": 5120,
-      "encoder_layerdrop": 0.0,
-      "encoder_layers": 32,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "init_std": 0.02,
-      "initializer_range": 0.02,
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "length_penalty": 1.0,
-      "max_length": 20,
-      "max_source_positions": 1500,
-      "min_length": 0,
-      "model_type": "qwen2_5_omni_audio_encoder",
-      "n_window": 100,
-      "no_repeat_ngram_size": 0,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_hidden_layers": 32,
-      "num_mel_bins": 128,
-      "num_return_sequences": 1,
-      "output_attentions": false,
-      "output_dim": 2048,
-      "output_hidden_states": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repetition_penalty": 1.0,
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "scale_embedding": false,
-      "sep_token_id": null,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": true,
-      "tokenizer_class": null,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "use_bfloat16": false
-    },
-    "audio_end_token_id": 151648,
-    "audio_start_token_id": 151647,
-    "audio_token_index": 151646,
-    "bos_token_id": 151644,
-    "dtype": "bfloat16",
-    "eos_token_id": 151645,
-    "ignore_index": -100,
-    "image_token_index": 151655,
-    "init_std": 0.02,
-    "initializer_range": 0.02,
-    "model_type": "qwen2_5_omni_thinker",
-    "pad_token_id": 151643,
-    "position_id_per_seconds": 25,
-    "seconds_per_chunk": 2,
-    "text_config": {
-      "_name_or_path": "",
-      "add_cross_attention": false,
-      "architectures": null,
-      "attention_dropout": 0.0,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "cross_attention_hidden_size": null,
-      "decoder_start_token_id": null,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "dtype": null,
-      "early_stopping": false,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "hidden_act": "silu",
-      "hidden_size": 2048,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "init_std": 0.02,
-      "initializer_range": 0.02,
-      "intermediate_size": 11008,
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "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"
-      ],
-      "length_penalty": 1.0,
-      "max_length": 20,
-      "max_position_embeddings": 32768,
-      "max_window_layers": 70,
-      "min_length": 0,
-      "model_type": "qwen2_5_omni_text",
-      "no_repeat_ngram_size": 0,
-      "num_attention_heads": 16,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_hidden_layers": 36,
-      "num_key_value_heads": 2,
-      "num_return_sequences": 1,
-      "output_attentions": false,
-      "output_hidden_states": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repetition_penalty": 1.0,
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "rms_norm_eps": 1e-06,
-      "rope_scaling": {
-        "mrope_section": [
-          16,
-          24,
-          24
-        ],
-        "rope_type": "default",
-        "type": "default"
-      },
-      "rope_theta": 1000000.0,
-      "sep_token_id": null,
-      "sliding_window": null,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": false,
-      "tokenizer_class": null,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "use_bfloat16": false,
-      "use_cache": true,
-      "use_sliding_window": false,
-      "vocab_size": 151936,
-      "rope_parameters": {
-        "interleaved": true,
-        "mrope_interleaved": true,
-        "mrope_section": [
-          24,
-          20,
-          20
-        ],
-        "rope_theta": 1000000,
-        "rope_type": "default",
-        "type": "default"
-      }
-    },
-    "tie_word_embeddings": false,
-    "use_cache": true,
-    "user_token_id": 872,
-    "video_token_index": 151656,
-    "vision_config": {
-      "_attn_implementation_autoset": true,
-      "_name_or_path": "",
-      "add_cross_attention": false,
-      "architectures": null,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "cross_attention_hidden_size": null,
-      "decoder_start_token_id": null,
-      "depth": 32,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "dtype": null,
-      "early_stopping": false,
-      "embed_dim": 1280,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "fullatt_block_indexes": [
-        7,
-        15,
-        23,
-        31
-      ],
-      "hidden_act": "silu",
-      "hidden_size": 1280,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "in_channels": 3,
-      "in_chans": 3,
-      "init_std": 0.02,
-      "initializer_range": 0.02,
-      "intermediate_size": 3420,
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "length_penalty": 1.0,
-      "max_length": 20,
-      "min_length": 0,
-      "model_type": "qwen2_5_omni_vision_encoder",
-      "no_repeat_ngram_size": 0,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_heads": 16,
-      "num_return_sequences": 1,
-      "out_hidden_size": 2048,
-      "output_attentions": false,
-      "output_hidden_states": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "patch_size": 14,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repetition_penalty": 1.0,
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "sep_token_id": null,
-      "spatial_merge_size": 2,
-      "spatial_patch_size": 14,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "temporal_patch_size": 2,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": true,
-      "tokenizer_class": null,
-      "tokens_per_second": 25,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "use_bfloat16": false,
-      "window_size": 112
-    },
-    "vision_end_token_id": 151653,
-    "vision_start_token_id": 151652,
-    "vision_token_id": 151654
-  },
-  "token2wav_config": {
-    "_attn_implementation_autoset": true,
-    "bigvgan_config": {
-      "_attn_implementation_autoset": true,
-      "_name_or_path": "",
-      "add_cross_attention": false,
-      "architectures": null,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "cross_attention_hidden_size": null,
-      "decoder_start_token_id": null,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "dtype": null,
-      "early_stopping": false,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "length_penalty": 1.0,
-      "max_length": 20,
-      "mel_dim": 80,
-      "min_length": 0,
-      "model_type": "qwen2_5_omni_bigvgan",
-      "no_repeat_ngram_size": 0,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_return_sequences": 1,
-      "output_attentions": false,
-      "output_hidden_states": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repetition_penalty": 1.0,
-      "resblock_dilation_sizes": [
-        [
-          1,
-          3,
-          5
-        ],
-        [
-          1,
-          3,
-          5
-        ],
-        [
-          1,
-          3,
-          5
-        ]
-      ],
-      "resblock_kernel_sizes": [
-        3,
-        7,
-        11
-      ],
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "sep_token_id": null,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": true,
-      "tokenizer_class": null,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "upsample_initial_channel": 1536,
-      "upsample_kernel_sizes": [
-        11,
-        7,
-        4,
-        4,
-        4,
-        4
-      ],
-      "upsample_rates": [
-        5,
-        3,
-        2,
-        2,
-        2,
-        2
-      ],
-      "use_bfloat16": false,
-      "use_bias_at_final": false
-    },
-    "dit_config": {
-      "_attn_implementation_autoset": true,
-      "_name_or_path": "",
-      "add_cross_attention": false,
-      "architectures": null,
-      "bad_words_ids": null,
-      "begin_suppress_tokens": null,
-      "block_size": 24,
-      "bos_token_id": null,
-      "chunk_size_feed_forward": 0,
-      "cross_attention_hidden_size": null,
-      "decoder_start_token_id": null,
-      "depth": 22,
-      "dim": 1024,
-      "diversity_penalty": 0.0,
-      "do_sample": false,
-      "dropout": 0.1,
-      "dtype": "float32",
-      "early_stopping": false,
-      "emb_dim": 512,
-      "enc_attention_channels": 64,
-      "enc_channels": [
-        256,
-        256,
-        256,
-        256,
-        768
-      ],
-      "enc_dilations": [
-        1,
-        2,
-        3,
-        4,
-        1
-      ],
-      "enc_dim": 128,
-      "enc_emb_dim": 192,
-      "enc_global_context": true,
-      "enc_kernel_sizes": [
-        5,
-        3,
-        3,
-        3,
-        1
-      ],
-      "enc_lin_neurons": 192,
-      "enc_res2net_scale": 2,
-      "enc_se_channels": 64,
-      "encoder_no_repeat_ngram_size": 0,
-      "eos_token_id": null,
-      "exponential_decay_length_penalty": null,
-      "ff_mult": 2,
-      "finetuning_task": null,
-      "forced_bos_token_id": null,
-      "forced_eos_token_id": null,
-      "head_dim": 64,
-      "heads": 16,
-      "hidden_size": 1024,
-      "id2label": {
-        "0": "LABEL_0",
-        "1": "LABEL_1"
-      },
-      "is_decoder": false,
-      "is_encoder_decoder": false,
-      "label2id": {
-        "LABEL_0": 0,
-        "LABEL_1": 1
-      },
-      "length_penalty": 1.0,
-      "look_ahead_layers": [
-        10
-      ],
-      "look_backward_layers": [
-        0,
-        20
-      ],
-      "max_length": 20,
-      "max_position_embeddings": 32768,
-      "mel_dim": 80,
-      "min_length": 0,
-      "model_type": "qwen2_5_omni_dit",
-      "no_repeat_ngram_size": 0,
-      "num_attention_heads": 16,
-      "num_beam_groups": 1,
-      "num_beams": 1,
-      "num_embeds": 8193,
-      "num_hidden_layers": 22,
-      "num_return_sequences": 1,
-      "output_attentions": false,
-      "output_hidden_states": false,
-      "output_scores": false,
-      "pad_token_id": null,
-      "prefix": null,
-      "problem_type": null,
-      "pruned_heads": {},
-      "remove_invalid_values": false,
-      "repeats": 2,
-      "repetition_penalty": 1.0,
-      "return_dict": true,
-      "return_dict_in_generate": false,
-      "rope_theta": 10000.0,
-      "sep_token_id": null,
-      "suppress_tokens": null,
-      "task_specific_params": null,
-      "temperature": 1.0,
-      "tf_legacy_loss": false,
-      "tie_encoder_decoder": false,
-      "tie_word_embeddings": true,
-      "tokenizer_class": null,
-      "top_k": 50,
-      "top_p": 1.0,
-      "torchscript": false,
-      "typical_p": 1.0,
-      "use_bfloat16": false
-    },
-    "dtype": "bfloat16",
-    "model_type": "qwen2_5_omni_token2wav"
-  },
-  "transformers_version": "4.57.1"
-}

processors/reasoning/generation_config.json

@@ -1,4 +0,0 @@
-{
-  "_from_model_config": true,
-  "transformers_version": "4.57.1"
-}

processors/reasoning/preprocessor_config.json

@@ -1,31 +0,0 @@
-{
-  "chunk_length": 300,
-  "dither": 0.0,
-  "feature_extractor_type": "WhisperFeatureExtractor",
-  "feature_size": 128,
-  "hop_length": 160,
-  "image_mean": [
-    0.48145466,
-    0.4578275,
-    0.40821073
-  ],
-  "image_processor_type": "Qwen2VLImageProcessor",
-  "image_std": [
-    0.26862954,
-    0.26130258,
-    0.27577711
-  ],
-  "max_pixels": 12845056,
-  "merge_size": 2,
-  "min_pixels": 3136,
-  "n_fft": 400,
-  "n_samples": 4800000,
-  "nb_max_frames": 30000,
-  "padding_side": "right",
-  "padding_value": 0.0,
-  "patch_size": 14,
-  "processor_class": "Qwen2_5OmniProcessor",
-  "return_attention_mask": true,
-  "sampling_rate": 16000,
-  "temporal_patch_size": 2
-}

processors/reasoning/special_tokens_map.json

@@ -1,38 +0,0 @@
-{
-  "additional_special_tokens": [
-    "<|im_start|>",
-    "<|im_end|>",
-    "<|AUDIO|>",
-    "<|audio_bos|>",
-    "<|audio_eos|>",
-    "<|box_end|>",
-    "<|quad_start|>",
-    "<|quad_end|>",
-    "<|vision_bos|>",
-    "<|vision_eos|>",
-    "<|vision_pad|>",
-    "<|IMAGE|>",
-    "<|VIDEO|>"
-  ],
-  "audio_bos_token": "<|audio_bos|>",
-  "audio_eos_token": "<|audio_eos|>",
-  "audio_token": "<|AUDIO|>",
-  "eos_token": {
-    "content": "<|im_end|>",
-    "lstrip": false,
-    "normalized": false,
-    "rstrip": false,
-    "single_word": false
-  },
-  "image_token": "<|IMAGE|>",
-  "pad_token": {
-    "content": "<|endoftext|>",
-    "lstrip": false,
-    "normalized": false,
-    "rstrip": false,
-    "single_word": false
-  },
-  "video_token": "<|VIDEO|>",
-  "vision_bos_token": "<|vision_bos|>",
-  "vision_eos_token": "<|vision_eos|>"
-}

processors/reasoning/tokenizer.json

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:8441917e39ae0244e06d704b95b3124795cec478e297f9afac39ba670d7e9d99
-size 11421870

processors/reasoning/tokenizer_config.json

@@ -1,222 +0,0 @@
-{
-  "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": "<|AUDIO|>",
-      "lstrip": false,
-      "normalized": false,
-      "rstrip": false,
-      "single_word": false,
-      "special": true
-    },
-    "151647": {
-      "content": "<|audio_bos|>",
-      "lstrip": false,
-      "normalized": false,
-      "rstrip": false,
-      "single_word": false,
-      "special": true
-    },
-    "151648": {
-      "content": "<|audio_eos|>",
-      "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_bos|>",
-      "lstrip": false,
-      "normalized": false,
-      "rstrip": false,
-      "single_word": false,
-      "special": true
-    },
-    "151653": {
-      "content": "<|vision_eos|>",
-      "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|>",
-      "lstrip": false,
-      "normalized": false,
-      "rstrip": false,
-      "single_word": false,
-      "special": true
-    },
-    "151656": {
-      "content": "<|VIDEO|>",
-      "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
-    }
-  },
-  "additional_special_tokens": [
-    "<|im_start|>",
-    "<|im_end|>",
-    "<|AUDIO|>",
-    "<|audio_bos|>",
-    "<|audio_eos|>",
-    "<|box_end|>",
-    "<|quad_start|>",
-    "<|quad_end|>",
-    "<|vision_bos|>",
-    "<|vision_eos|>",
-    "<|vision_pad|>",
-    "<|IMAGE|>",
-    "<|VIDEO|>"
-  ],
-  "audio_bos_token": "<|audio_bos|>",
-  "audio_eos_token": "<|audio_eos|>",
-  "audio_token": "<|AUDIO|>",
-  "bos_token": null,
-  "clean_up_tokenization_spaces": false,
-  "eos_token": "<|im_end|>",
-  "errors": "replace",
-  "extra_special_tokens": {
-    "audio_bos_token": "<|audio_bos|>",
-    "audio_eos_token": "<|audio_eos|>",
-    "audio_token": "<|AUDIO|>",
-    "image_token": "<|IMAGE|>",
-    "video_token": "<|VIDEO|>",
-    "vision_bos_token": "<|vision_bos|>",
-    "vision_eos_token": "<|vision_eos|>"
-  },
-  "image_token": "<|IMAGE|>",
-  "model_max_length": 32768,
-  "pad_token": "<|endoftext|>",
-  "processor_class": "Qwen2_5OmniProcessor",
-  "split_special_tokens": false,
-  "tokenizer_class": "Qwen2Tokenizer",
-  "unk_token": null,
-  "video_token": "<|VIDEO|>",
-  "vision_bos_token": "<|vision_bos|>",
-  "vision_eos_token": "<|vision_eos|>"
-}

processors/reasoning/video_preprocessor_config.json

@@ -1,55 +0,0 @@
-{
-  "chunk_length": 300,
-  "crop_size": null,
-  "data_format": "channels_first",
-  "default_to_square": true,
-  "device": null,
-  "dither": 0.0,
-  "do_center_crop": null,
-  "do_convert_rgb": true,
-  "do_normalize": true,
-  "do_rescale": true,
-  "do_resize": true,
-  "do_sample_frames": false,
-  "feature_extractor_type": "WhisperFeatureExtractor",
-  "feature_size": 128,
-  "fps": null,
-  "hop_length": 160,
-  "image_mean": [
-    0.48145466,
-    0.4578275,
-    0.40821073
-  ],
-  "image_std": [
-    0.26862954,
-    0.26130258,
-    0.27577711
-  ],
-  "input_data_format": null,
-  "max_frames": 768,
-  "max_pixels": 12845056,
-  "merge_size": 2,
-  "min_frames": 4,
-  "min_pixels": 3136,
-  "n_fft": 400,
-  "n_samples": 4800000,
-  "nb_max_frames": 30000,
-  "num_frames": null,
-  "pad_size": null,
-  "padding_side": "right",
-  "padding_value": 0.0,
-  "patch_size": 14,
-  "processor_class": "Qwen2_5OmniProcessor",
-  "resample": 3,
-  "rescale_factor": 0.00392156862745098,
-  "return_attention_mask": true,
-  "return_metadata": false,
-  "sampling_rate": 16000,
-  "size": {
-    "longest_edge": 12845056,
-    "shortest_edge": 3136
-  },
-  "temporal_patch_size": 2,
-  "video_metadata": null,
-  "video_processor_type": "Qwen2VLVideoProcessor"
-}

pyproject.toml

@@ -4,33 +4,23 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "myna-hokkien"
-version = "0.1.0"
-description = "iNLP-Lab Myna-Hokkien inference runtime"
+version = "0.2.0"
+description = "iNLP-Lab Myna-Hokkien native Qwen3-Omni inference runtime"
 requires-python = ">=3.10"
 dependencies = [
   "torch>=2.6",
-  "torchvision",
   "transformers==4.57.1",
-  "accelerate==1.14.0",
-  "peft==0.17.1",
-  "huggingface_hub==0.36.2",
-  "safetensors==0.8.0",
-  "numpy==2.2.6",
-  "librosa==0.11.0",
-  "soundfile==0.14.0",
-  "pillow>=10.0",
+  "accelerate>=1.10,<2",
+  "huggingface_hub>=0.36,<2",
+  "safetensors>=0.4",
+  "numpy>=1.26,<3",
+  "librosa>=0.11,<1",
+  "soundfile>=0.13,<1",
 ]
 
 [project.urls]
 Homepage = "https://huggingface.co/iNLP-Lab/MynaHokkien"
 Repository = "https://huggingface.co/iNLP-Lab/MynaHokkien"
 
-[tool.setuptools]
-py-modules = [
-  "configuration_mynahokkien",
-  "modeling_mynahokkien",
-  "processing_mynahokkien",
-]
-
 [tool.setuptools.packages.find]
 include = ["mynahokkien*"]

release_config.json

@@ -0,0 +1,18 @@
+{
+  "release": "Myna-Hokkien release_v2",
+  "architecture": "native Qwen3-Omni-30B-A3B-Instruct",
+  "thinker": "release_v2 Thinker Stage 2",
+  "talker": "release_v2 Talker Stage 4 epoch 1",
+  "speaker": "Ethan",
+  "language": "nan",
+  "input_sampling_rate": 16000,
+  "output_sampling_rate": 24000,
+  "system_prompt": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory inputs and generating text and speech.",
+  "default_audio_prompt": "Listen to the spoken Hokkien and reply naturally in concise Singapore Hokkien. Always answer in colloquial Singapore Hokkien written in Hanji. Never answer in Mandarin or English. Do not repeat or transcribe the input; respond to it directly.",
+  "default_text_prompt": "Read the text and reply naturally in concise Singapore Hokkien. Always answer in colloquial Singapore Hokkien written in Hanji. Never answer in Mandarin or English. Do not repeat the input; respond to it directly.",
+  "prompt_position": "user turn",
+  "thinker_do_sample": false,
+  "talker_do_sample": false,
+  "thinker_max_new_tokens": 256,
+  "talker_max_new_tokens": 4096
+}

requirements.txt

@@ -1,11 +1,8 @@
 torch>=2.6
-torchvision
 transformers==4.57.1
-accelerate==1.14.0
-peft==0.17.1
-huggingface_hub==0.36.2
-safetensors==0.8.0
-numpy==2.2.6
-librosa==0.11.0
-soundfile==0.14.0
-pillow>=10.0
+accelerate>=1.10,<2
+huggingface_hub>=0.36,<2
+safetensors>=0.4
+numpy>=1.26,<3
+librosa>=0.11,<1
+soundfile>=0.13,<1

processors/generation/tokenizer.json → tokenizer.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b34e316e4a3886855d9af88d0713da9a41dc8e00359c890ac022df6c6a1101a3
-size 11424429
+oid sha256:09267689b8362020b9763b65dd5be7e086b31e28d72e02837a9e781de9a91bc7
+size 11423986

processors/generation/tokenizer_config.json → tokenizer_config.json

@@ -305,12 +305,8 @@
     "vision_eos_token": "<|vision_end|>"
   },
   "image_token": "<|image_pad|>",
-  "max_length": null,
   "model_max_length": 131072,
-  "pad_to_multiple_of": null,
   "pad_token": "<|endoftext|>",
-  "pad_token_type_id": 0,
-  "padding_side": "left",
   "processor_class": "Qwen3OmniMoeProcessor",
   "split_special_tokens": false,
   "tokenizer_class": "Qwen2Tokenizer",