"""Debug script to save intermediate outputs from the speaker encoder.""" audio_path = "/root/qwen-3-tts-ggml/clone.wav" model_path = "/root/qwen-3-tts-ggml/models ...
1. DiT (denoising) - if diffusers latents work with our VAE 2. VAE decoder - if diffusers latents also produce grid pattern with our VAE f.write(np.array([len(shape)], dtype=np.uint64).tobytes()) ...
The tiny editor has some big features.