Hugging Face's logo

Spaces:
WeReCooking
/
FE2E-CPU
Paused

App Files Community
FE2E-CPU

Commit History

  1. Update README.md

    Nekochu
    Verified signature
    3469c86

  1. Remove onnxruntime

    Nekochu
    Verified signature
    2fda2eb

  2. Revert to PyTorch INT8 (ONNX export produces NaN)

    Nekochu
    Verified signature
    563ef1b

  3. Reduce ORT memory: disable prepacking, basic optimization, 1 thread

    Nekochu
    Verified signature
    10d0786

  4. Fix: unsqueeze prompt embeddings

    Nekochu
    Verified signature
    1620b49

  5. Add onnxruntime dependency

    Nekochu
    Verified signature
    ac9fb33

  6. Switch to ONNX INT8 DiT

    Nekochu
    Verified signature
    6ee4bac

  1. fixes: cudagc guard, rm conditioner.py, turbo depth colormap, proper normal viz, compact UI, example

    Nekochu
    d65d5b5

  1. fix: .weight.dtype crashes on INT8 quantized Linear, use .float()

    Nekochu
    2aca4a6

  2. fix: use FP32 dtype on CPU (no bfloat16 autocast with INT8 model)

    Nekochu
    09dfa8a

  3. fix: load full torch.save model directly (no FP32 construction, mmap)

    Nekochu
    9d04602

  4. fix OOM: use mmap loading + assign (no memory copy)

    Nekochu
    b57f6e2

  5. fix: eager model load at startup (lazy load causes SSE timeout)

    Nekochu
    19a7c6b

  6. fix: move demo to module level (Gradio SDK needs it, not inside main())

    Nekochu
    284342e

  7. add example images for lazy-loaded examples

    Nekochu
    214c569

  8. switch to Gradio SDK (no Docker needed for pure Python)

    Nekochu
    2c1c6de

  9. fix: PYTHONUNBUFFERED for Docker log visibility

    Nekochu
    a73932a

  10. use pre-quantized INT8 model (no FP8 casting, no LoRA merge at runtime)

    Nekochu
    a3ba5ed

  11. fix: diffusers dep, layer-by-layer FP8->FP32 cast, LoRA merge in FP32, INT8 quant

    Nekochu
    551acb3

  12. FE2E depth+normal CPU Space: FP8 dynamic INT8, single denoise

    Nekochu
    405d2b1