Nov 3 Yesterday

1. 04:51

   Disable VAE tile parallelism in workers + post-inference logging Running

   akhaliq HF Staff
   Verified signature

   99affe4
2. 04:33

   Patch flex_attention -> SDPA (ZeroGPU has no torch.compile), bump GPU duration for cold starts

   akhaliq HF Staff
   Verified signature

   6e13574
3. 04:28

   ZeroGPU shims: trivial device mesh, gpu_memory probe, spaces-first import, drop byt5/Glyph downloads

   akhaliq HF Staff
   Verified signature

   1b4f3b1
4. 04:22

   ZeroGPU: gloo pg init, drop cuda.set_device (emulation wall)

   akhaliq HF Staff
   Verified signature

   39fcfc1
5. 03:47

   Glyph-SDXL-v2 from HF mirror instead of ModelScope

   akhaliq HF Staff
   Verified signature

   b58bd2e
6. 03:30

   Gradio 6 compat: gr.Number precision -> step

   akhaliq HF Staff
   Verified signature

   20536f5
7. 03:27

   ZeroGPU: @spaces.GPU(xlarge) wrappers with dynamic duration

   akhaliq HF Staff
   Verified signature

   a935b80
8. 03:24

   FrameForge demo: in-process wrapper of official infer.py, three tabs

   akhaliq HF Staff
   Verified signature

   19207c7