Issues

hecate: --device has no MPS branch, so Apple Silicon silently takes the 5.76x slower CPU path (and bf16 on MPS is slower, so that guard is right)

#1814

· AndreasHad04 opened

on Sep 16, 2026

vc_render_tifxyz: a failed remote chunk fetch aborts with SIGABRT (rc 134) instead of a clean error

#1809

· Bullo27 opened

on Sep 16, 2026

PHercParis4 windings 128-129: both published tracings run through the fused outermost wrap, and the ink model finds no text there

#1777

· rodriguescarson opened

on Sep 14, 2026

normal_grid_level is silently ignored for single-scale normal-grid stores

#1776

· rodriguescarson opened

on Sep 14, 2026

optimized_inference README: the recommended 62-layer window for resnet3d-152-3d-decoder is off-centre on the published 2 um volumes

#1765

· kadenpool opened

on Sep 11, 2026

`koine_machines.inference`: no MPS branch, so on Apple Silicon the published `ink_9um` checkpoints run on the CPU with the GPU idle and nothing is printed

#1764

· SurgeFok opened

on Sep 11, 2026

Proposal: a header-only OME-Zarr pyramid integrity check at publish time — the whole class of defects it catches is invisible to every reader by construction

#1760

· sgsllc-jr opened

on Sep 10, 2026

`community-uploads/bruniss`: one undecodable store (48k chunks, no `.zarray`), one pyramid whose levels 1–5 are header-only, and a stray partial copy of `s5/surfaces/090.zarr` at the parent level

#1759

· sgsllc-jr opened

on Sep 10, 2026

`s5-fiber-directions.zarr` has no root `.zgroup`/`.zattrs`, so the `.zarr` path cannot be opened as a group even though `horizontal/` and `vertical/` inside it are valid

#1758

· sgsllc-jr opened

on Sep 10, 2026

Scroll 5 volume `20241024131838.zarr`: levels 1–5 are stored uncompressed (`compressor: null`, 389 GB) while level 0 is `blosc/zstd` — the same L0-compressed / L1+-null signature appears in 8 community pyramids

#1757

· sgsllc-jr opened

on Sep 10, 2026

`other/dev/inked_zarrs`: `3336_predictions.zarr` has headers at all 6 levels but zero chunks; `3336_predictions{9,10,11}.zarr` declare `scale` vectors that contradict their own shapes

#1756

· sgsllc-jr opened

on Sep 10, 2026

Frag3 88 keV volumes declare a 6-level pyramid but only level 0 exists (both `volumes_zarr` and `volumes_standardized`)

#1755

· sgsllc-jr opened

on Sep 10, 2026

Navigation Menu

Uh oh!

All issues

hecate: --device has no MPS branch, so Apple Silicon silently takes the 5.76x slower CPU path (and bf16 on MPS is slower, so that guard is right)

vc_render_tifxyz: a failed remote chunk fetch aborts with SIGABRT (rc 134) instead of a clean error

PHercParis4 windings 128-129: both published tracings run through the fused outermost wrap, and the ink model finds no text there

normal_grid_level is silently ignored for single-scale normal-grid stores

optimized_inference README: the recommended 62-layer window for resnet3d-152-3d-decoder is off-centre on the published 2 um volumes

`koine_machines.inference`: no MPS branch, so on Apple Silicon the published `ink_9um` checkpoints run on the CPU with the GPU idle and nothing is printed

Proposal: a header-only OME-Zarr pyramid integrity check at publish time — the whole class of defects it catches is invisible to every reader by construction

`community-uploads/bruniss`: one undecodable store (48k chunks, no `.zarray`), one pyramid whose levels 1–5 are header-only, and a stray partial copy of `s5/surfaces/090.zarr` at the parent level

`s5-fiber-directions.zarr` has no root `.zgroup`/`.zattrs`, so the `.zarr` path cannot be opened as a group even though `horizontal/` and `vertical/` inside it are valid

Scroll 5 volume `20241024131838.zarr`: levels 1–5 are stored uncompressed (`compressor: null`, 389 GB) while level 0 is `blosc/zstd` — the same L0-compressed / L1+-null signature appears in 8 community pyramids

`other/dev/inked_zarrs`: `3336_predictions.zarr` has headers at all 6 levels but zero chunks; `3336_predictions{9,10,11}.zarr` declare `scale` vectors that contradict their own shapes

Frag3 88 keV volumes declare a 6-level pyramid but only level 0 exists (both `volumes_zarr` and `volumes_standardized`)

Uh oh!

Search results