Update package and model

README.md

@@ -16,6 +16,7 @@ tags:
   - text-to-speech
   - speech-synthesis
 preload_from_hub:
+  - herimor/voxtream2 model.safetensors,config.json,phoneme_to_token.json
   - kyutai/moshiko-pytorch-bf16 tokenizer-e351c8d8-checkpoint125.safetensors
   - sarulab-speech/sidon-v0.1 decoder_cuda.pt
   - sarulab-speech/sidon-v0.1 feature_extractor_cuda.pt

app.py

@@ -74,7 +74,7 @@ def demo_app(config: SpeechGeneratorConfig, demo_examples, synthesize_fn):
     with gr.Blocks(css=CUSTOM_CSS, title="VoXtream2") as demo:
         gr.Markdown("# VoXtream2 TTS demo")
         gr.Markdown(
-            "⚠️ First 3-5 runs may have higher latency due to model loading and warmup."
+            "⚠️ The initial latency can be high due to deployment on ZeroGPU. For faster inference, please try local deployment. For more details, please visit [VoXtream GitHub repo](https://github.com/herimor/voxtream)"
         )
 
         with gr.Row(equal_height=True, elem_id="cols"):

configs/generator.json

@@ -11,7 +11,7 @@
     "topk": 5,
     "top_p": 0.9,
     "max_audio_length_ms": 60000,
-    "model_repo": "herimor/debug",
+    "model_repo": "herimor/voxtream2",
     "model_name": "model.safetensors",
     "model_config_name": "config.json",
     "mimi_sr": 24000,

requirements.txt

@@ -1,5 +1,3 @@
---extra-index-url https://test.pypi.org/simple/
-
-voxtream==0.2.2
+voxtream==0.2.0
 gradio_client==1.3.0
 pydantic==2.10.6