Commit History

28 days ago

  1. Update app.py Running

    Verified signature
    91ab218

  1. Remove verbose= kwarg from ReduceLROnPlateau: removed in newer PyTorch (we are now on 2.8.0)

    Verified signature
    ab4dd9b
  2. Reset torch default device to cpu after offload.profile(): fixes device mismatches in econdataset/pymaf/smplx code that assumes CPU-default tensor creation, without affecting mmgps explicit hook-based device management

    Verified signature
    181be20
  3. Fix device mismatch: maskrcnn detector lands on cuda under mmgps lingering default device, but torch.from_numpy() input stays on cpu

    Verified signature
    cf35ba7
  4. Bump GPU duration to 600s: memory issue is fixed (attention slicing), task now just needs more time for the full multiview diffusion + mesh reconstruction pipeline

    Verified signature
    cb5743a
  5. Implement attention slicing: chunk SDPA calls along the batch*heads dim since torch 2.8s fused kernels reject this shape on this GPU, forcing an O(seq_len^2 * batch) math fallback that OOMs even a 96GB card

    Verified signature
    63d710a
  6. Diagnose why SDPA auto-select avoids flash/efficient backends: try FLASH_ATTENTION explicitly and log the actual error, plus dtype/contiguity/mask info

    Verified signature
    4df1266
  7. Add memory diagnostics at each attention call to see if VRAM usage is growing unbounded or just spiking once

    Verified signature
    4ba1e6a
  8. Add GPU memory diagnostic print to confirm whether size=xlarge is actually being granted

    Verified signature
    33ea232
  9. Request ZeroGPU xlarge (full 96GB Blackwell card) instead of default large (48GB half-card): the multiview attention activation memory OOMs on the half-card even with mmgp weight offloading

    Verified signature
    8526e94
  10. Try mmgp VerylowRAM_LowVRAM profile: LowRAM_LowVRAM still OOMs during attention

    Verified signature
    432c498
  11. Fix remaining comment typo

    Verified signature
    3a0919b
  12. Fix comment typo (apostrophe escaping) from previous commit

    Verified signature
    97e5396
  13. Exclude image_normalizer from mmgp offload management (its .scale() method bypasses mmgps forward-hook, stranding buffers on CPU); pin it on GPU directly instead

    Verified signature
    a143759
  14. Fix create_mean_pose() to build its numpy array from CPU tensors explicitly, instead of masking it with a global torch default-device reset that broke mmgp's own device handling

    Verified signature
    7f3d882
  15. Reset torch default device to cpu after offload.profile(): mmgp leaves it as cuda, which broke SMPLDataset tensor creation later in the script

    Verified signature
    fc017ad
  16. Bump diffusers to 0.29.0: mmgp (via optimum-quanto) needs PixArtTransformer2DModel, absent from the old 0.26.0 pin; verified all of PSHumans own diffusers imports (incl. LoRACompatibleConv/AdaLayerNorm) still resolve at 0.29.0

    Verified signature
    6ce5fd0
  17. Use MMGP budget-based offloading/quantization instead of a blanket .to('cuda'): the pipeline genuinely OOMs on this ZeroGPU MIG slice's effective VRAM

    Verified signature
    9eacf98
  18. Revert forced EFFICIENT_ATTENTION backend (unsupported for these tensor shapes on this GPU); back to auto-selected SDPA

    Verified signature
    42c2660
  19. Force SDPA's memory-efficient backend for PSHuman's multiview attention: rule out a math-fallback OOM on Blackwell

    Verified signature
    ff5e65e

  1. Disable expandable_segments CUDA allocator: it hits an NVML assertion on MIG-sliced Blackwell GPUs

    Verified signature
    bb80551
  2. Keep PSHuman's custom multiview attention processors but swap their inner xformers.ops.memory_efficient_attention call for torch's native scaled_dot_product_attention (xformers' Hopper kernel crashes on Blackwell/sm_120)

    Verified signature
    2080275
  3. Drop xformers: its Hopper-specific flash-attention kernel crashes on Blackwell (sm_120) with 'CUDA error: invalid argument'. torch 2.8's native SDPA (used automatically by diffusers) replaces it.

    Verified signature
    4725c84
  4. Bump warp-lang to 1.17.0: kaolin 0.18.0 physics subsystem imports warp.fem.linalg.inverse_qr, missing from the old 1.4.2 pin

    Verified signature
    e4d389f
  5. Bump sympy to 1.13.3 and triton to 3.4.0: both hard-required by torch 2.8.0 (sympy>=1.13.3, triton==3.4.0 exactly on linux x86_64)

    Verified signature
    2cc7752
  6. Bump typing_extensions to 4.12.2: torch 2.8.0 requires >=4.10.0, conflicting with the old 4.9.0 pin

    Verified signature
    e7397e9
  7. Upgrade to torch 2.8.0+cu128 (adds Blackwell/sm_12x kernels) with matching kaolin 0.18.0, pytorch3d 0.7.8+pt2.8.0cu128, torch_scatter 2.1.2+pt28cu128, xformers 0.0.32.post2

    Verified signature
    c2448e9
  8. Add nvidia-cuda-runtime + LD_LIBRARY_PATH shim so nvdiffrast's JIT-compiled CUDA extension can find libcudart.so.13 on Blackwell/CUDA-13 ZeroGPU workers

    Verified signature
    f5c6af6
  9. Fix torch pin to 2.1.2 (xformers==0.0.23.post1 requires torch==2.1.2 exactly, conflicting with the 2.1.0 I pinned previously)

    Verified signature
    b908182
  10. Pin torch/torchvision/torchaudio to 2.1.0+cu121: unpinned torch was drifting to latest PyPI (now 2.14.0/cu13x), breaking ABI with the hard-pinned kaolin/pytorch3d/torch_scatter cu121 wheels (libcudart.so.13 not found)

    Verified signature
    f75e574
  11. Pin setuptools<82: pkg_resources was fully removed in setuptools 82, breaking nvdiffrast/torch.utils.cpp_extension import

    Verified signature
    606c14a
  12. Add setuptools so pkg_resources is available for nvdiffrast/torch.utils.cpp_extension import

    Verified signature
    f2c23c5

  1. Update requirements.txt

    Verified signature
    3e0afe2
  2. Update requirements.txt

    Verified signature
    8561d1a

  1. increase ZeroGPU needed time for inference

    Verified signature
    afb1563
  2. includes ZeroGPU deco for main function

    Verified signature
    eb0fbff
  3. import spaces for ZeroGPU

    Verified signature
    bd444a3

  1. set xformers 0.0.23.post1

    Verified signature
    f79ea6f
  2. Update requirements.txt

    Verified signature
    729a66a
  3. Update requirements.txt

    Verified signature
    62b7383
  4. Update requirements.txt

    Verified signature
    51191bd
  5. downgradio 5.8.0

    Verified signature
    67e2a93
  6. Update requirements.txt

    Verified signature
    ea3c3d6
  7. back to working properly config, psguman currently not compatible with Zero

    Verified signature
    e1eb50a
  8. Update requirements.txt

    Verified signature
    9f8cd50
  9. Update requirements.txt

    Verified signature
    934dbb6
  10. Update requirements.txt

    Verified signature
    0121406
  11. upgrade xformers==0.0.29.post1

    Verified signature
    55f4540
  12. back to torch modules 2.1.2 v

    Verified signature
    0dcb8cc
  13. look for the right putorch3d

    Verified signature
    3a9651e