Text-to-Speech
KimiAudio
Safetensors
English
Chinese
audio
audio-language-model
speech-recognition
audio-understanding
audio-generation
chat
custom_code
Instructions to use moonshotai/Kimi-Audio-7B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- KimiAudio
How to use moonshotai/Kimi-Audio-7B-Instruct with KimiAudio:
# Example usage for KimiAudio # pip install git+https://github.com/MoonshotAI/Kimi-Audio.git from kimia_infer.api.kimia import KimiAudio model = KimiAudio(model_path="moonshotai/Kimi-Audio-7B-Instruct", load_detokenizer=True) sampling_params = { "audio_temperature": 0.8, "audio_top_k": 10, "text_temperature": 0.0, "text_top_k": 5, } # For ASR asr_audio = "asr_example.wav" messages_asr = [ {"role": "user", "message_type": "text", "content": "Please transcribe the following audio:"}, {"role": "user", "message_type": "audio", "content": asr_audio} ] _, text = model.generate(messages_asr, **sampling_params, output_type="text") print(text) # For Q&A qa_audio = "qa_example.wav" messages_conv = [{"role": "user", "message_type": "audio", "content": qa_audio}] wav, text = model.generate(messages_conv, **sampling_params, output_type="both") sf.write("output_audio.wav", wav.cpu().view(-1).numpy(), 24000) print(text) - Notebooks
- Google Colab
- Kaggle
Add Kimi-Audio EOS and pad token ids
#20
by tunglinwood - opened
- generation_config.json +8 -2
generation_config.json
CHANGED
|
@@ -1,3 +1,9 @@
|
|
| 1 |
{
|
| 2 |
-
"max_length": 8192
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
{
|
| 2 |
+
"max_length": 8192,
|
| 3 |
+
"eos_token_id": [
|
| 4 |
+
151644,
|
| 5 |
+
151645,
|
| 6 |
+
151667
|
| 7 |
+
],
|
| 8 |
+
"pad_token_id": 152063
|
| 9 |
+
}
|