|
|
--- |
|
|
title: Qwen2.5 1.5B Instruct Gkd Demo |
|
|
emoji: 📊 |
|
|
colorFrom: yellow |
|
|
colorTo: red |
|
|
sdk: gradio |
|
|
sdk_version: 5.34.2 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
license: mit |
|
|
short_description: Qwen2.5-1.5B-Instruct-gkd-demo |
|
|
--- |
|
|
|
|
|
# Qwen2.5 1.5B Instruct Gkd Demo |
|
|
|
|
|
A voice-enabled chatbot powered by: |
|
|
- `Qwen2.5-1.5B-Instruct-gkd` for language generation |
|
|
- `SpeechT5` for text-to-speech |
|
|
- `Whisper-small` for speech-to-text |
|
|
|
|
|
Try chatting, listen to the audio response, or upload an audio file for transcription. |
|
|
|
|
|
--- |
|
|
|
|
|
## ✅ What’s Fixed & Improved |
|
|
|
|
|
| Issue | Fix / Enhancement | |
|
|
|------|---------------------| |
|
|
| 🔁 **Duplicate TTS Block** | Removed duplicate code in `generate_response_and_audio` | |
|
|
| ❌ **Incorrect Condition Check** | Replaced unsafe `all([...])` with proper `is not None` checks | |
|
|
| 📏 **Long Text Handling (TTS)** | Added chunking to avoid exceeding 512 token limit | |
|
|
| 🧠 **LLM Generation Safety** | Ensures `generated_text` is always defined | |
|
|
| ⚙️ **Model Loading Optimization** | Moved model loading into the first request (Hugging Face Spaces friendly) | |
|
|
| 🧼 **Code Cleanliness** | Better structure, comments, and readability | |
|
|
|
|
|
--- |
