Spaces:

---

AIencoder

/

Axon

like 1

Running

🚀 V2 Release: CPU Optimization & Stability Fixes

This update brings major stability improvements, specifically targeting the Hugging Face Free CPU Tier.

📋 Changelog:

• ⚡ CPU Optimization: Implemented smart thread limiting (torch.set_num_threads) and switched to float32 to prevent deadlocks on 0% GPU hardware.
• 🧠 Model Stability: successfully running Qwen 2.5 Coder 1.5B on CPU by disabling sampling (do_sample=False) for faster, deterministic outputs.
• 🖥️ New UI Layout: Redesigned the interface with a "Control Center" layout—buttons are now side-by-side with the audio input for better usability.
• 🐛 Bug Fixes:
  • Fixed concurrency_count deprecation error.
  • Fixed Whisper pipeline input format (switching to raw numpy).
  • Resolved the "silent crash" deadlock issue during cold starts.

Enjoy the faster and more stable experience!