sci-ml/torchcodec
Decode/encode video and audio into PyTorch tensors via FFmpeg
Screenshots
-
torchcodec-0.17.0~amd64 ~arm64cuda python_single_target_python3_12 python_single_target_python3_13 python_single_target_python3_14 debug
View 
Download 
Browse License: BSD-2 Overlay: stuff -
torchcodec-0.17.0~amd64cuda +gif heif +jpeg +png +webp python_single_target_python3_12 python_single_target_python3_13 python_single_target_python3_14 debug
View Download Browse License: BSD gif? ( MIT ) Overlay: bentoo -
torchcodec-0.16.0-r2~amd64+avif +gif +heic +jpeg +png +webp python_single_target_python3_12 python_single_target_python3_13 python_single_target_python3_14 debug +amdgpu_targets_gfx908 +amdgpu_targets_gfx90a +amdgpu_targets_gfx942 +amdgpu_targets_gfx950 +amdgpu_targets_gfx1030 +amdgpu_targets_gfx1100 +amdgpu_targets_gfx1101 +amdgpu_targets_gfx1200 +amdgpu_targets_gfx1201 amdgpu_targets_gfx803 amdgpu_targets_gfx900 amdgpu_targets_gfx906 amdgpu_targets_gfx940 amdgpu_targets_gfx941 amdgpu_targets_gfx1010 amdgpu_targets_gfx1011 amdgpu_targets_gfx1012 amdgpu_targets_gfx1031 amdgpu_targets_gfx1102 amdgpu_targets_gfx1103 amdgpu_targets_gfx1150 amdgpu_targets_gfx1151 cuda rocm
View Download Browse License: BSD Overlay: local-ai -
torchcodec-0.16.0-r1~amd64+avif +gif +heic +jpeg +png +webp python_single_target_python3_12 python_single_target_python3_13 python_single_target_python3_14 debug +amdgpu_targets_gfx908 +amdgpu_targets_gfx90a +amdgpu_targets_gfx942 +amdgpu_targets_gfx950 +amdgpu_targets_gfx1030 +amdgpu_targets_gfx1100 +amdgpu_targets_gfx1101 +amdgpu_targets_gfx1200 +amdgpu_targets_gfx1201 amdgpu_targets_gfx803 amdgpu_targets_gfx900 amdgpu_targets_gfx906 amdgpu_targets_gfx940 amdgpu_targets_gfx941 amdgpu_targets_gfx1010 amdgpu_targets_gfx1011 amdgpu_targets_gfx1012 amdgpu_targets_gfx1031 amdgpu_targets_gfx1102 amdgpu_targets_gfx1103 amdgpu_targets_gfx1150 amdgpu_targets_gfx1151 cuda rocm
View Download Browse License: BSD Overlay: local-ai -
torchcodec-0.16.0~amd64+avif +gif +heic +jpeg +png +webp python_single_target_python3_12 python_single_target_python3_13 python_single_target_python3_14 debug +amdgpu_targets_gfx908 +amdgpu_targets_gfx90a +amdgpu_targets_gfx942 +amdgpu_targets_gfx950 +amdgpu_targets_gfx1030 +amdgpu_targets_gfx1100 +amdgpu_targets_gfx1101 +amdgpu_targets_gfx1200 +amdgpu_targets_gfx1201 amdgpu_targets_gfx803 amdgpu_targets_gfx900 amdgpu_targets_gfx906 amdgpu_targets_gfx940 amdgpu_targets_gfx941 amdgpu_targets_gfx1010 amdgpu_targets_gfx1011 amdgpu_targets_gfx1012 amdgpu_targets_gfx1031 amdgpu_targets_gfx1102 amdgpu_targets_gfx1103 amdgpu_targets_gfx1150 amdgpu_targets_gfx1151 cuda rocm
View Download Browse License: BSD Overlay: local-ai
ChangeLog
commit 020dbb7bc66042253674f9bbcced226b0d8d43b0
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Oct 8 18:13:19 2026 +0200
sci-ml/torchcodec: take the CUDA host compiler from cuda_gccdir
The ebuild exported CUDAHOSTCXX=/usr/bin/g++-15 without depending on
gcc 15, so configure failed where that file was missing. It did so
unconditionally, because Torch's CMake enables the CUDA language
whenever the toolkit is installed, CPU builds included.
cuda_gccdir returns the newest installed gcc the toolkit accepts: 15
or older before 13.4, up to 16 from 13.4. It needs the toolkit, so
call it only for USE=cuda or when the toolkit is on the build host.
Where it has no answer (non-GNU toolchain, CHOST other than
*-pc-linux-gnu) nvcc keeps its default.
No change to what an existing install provides; no revision bump.
commit 1e6588795a291b4bd69af55933dd1960076f112f
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Oct 1 08:51:52 2026 +0200
sci-ml/torchcodec: add 0.17.0
The release adds audio conversion sources, but the build knobs the
ebuild sets (TORCHCODEC_BUILD_IMAGE, ENABLE_CUDA, BUILD_VERSION and the
license guard) are unchanged, and the image decoders still compile as
stubs that raise at runtime, linking no image libraries.
Against a ROCm build of caffe2 2.14.0, configure fails for this and
for 0.16.0 alike: Torch's installed LoadHIP.cmake looks for clang++
under an unslotted LLVM path. That is a caffe2 bug, so it is fixed
there rather than worked around in each consumer.
commit 597dba530d4bb9da27a56c52a396f6909e5f5680
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Sep 16 12:38:19 2026 +0200
sci-ml/torchcodec: default the CUDA architecture list
With USE=cuda and no TORCH_CUDA_ARCH_LIST, Torch's CMake emits gencode
flags for every architecture it knows, starting at compute_50. CUDA 13
dropped Maxwell, so nvcc rejects that target and the build fails before
compiling anything (issue #292).
Default the list to compute capability 7.5 with PTX, and warn that it may
not suit the GPU in use, pointing at package.env for a better value. The
PTX matters: this default is what a machine with no visible GPU falls
back to, and a 7.5 cubin on its own will not load on a newer card, while
PTX lets the driver JIT for it. An explicit setting is still honoured, so
anyone who already targets their own GPU sees no change.
commit 2eaac282f30722575b5fb2b1915f908f78232419
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 05:01:29 2026 +0200
sci-ml/torchcodec: trim 0.15.0 comment prose
commit 42a1a0653abe86d4559aba7473c55598dfff6ce8
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 04:02:08 2026 +0200
sci-ml/torchcodec: trim 0.16.0 comment prose
commit 1b0c1489c13afc483fa7dd9664347b158c756aa9
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 15:05:53 2026 +0200
sci-ml/torchcodec: drop 0.14.0
commit f34813b3b7fe83c14f1a002bdb50b1d2d2549094
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 15:05:50 2026 +0200
sci-ml/torchcodec: add 0.16.0
commit db901676e150bb0eb1db885312a1ebefd959dd15
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jul 16 10:25:36 2026 +0200
sci-ml/torchcodec: drop 0.13.0
commit 9261025447dd225a8273274301d53d28865afa96
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jul 16 10:25:36 2026 +0200
sci-ml/torchcodec: add 0.15.0
commit d173d0a84d3c08b8bd915d963db82d472f43fb32
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Jun 20 15:49:49 2026 +0200
sci-ml/torchcodec: Keyword 0.13.0 for ~arm64
commit 2a96510d24f3c7bc8d0b5bddf08fa1cd529ec138
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jun 17 22:58:09 2026 +0200
sci-ml/torchcodec: Keyword 0.14.0 for ~arm64
Torch C++/ffmpeg media-decoding extension; builds against sci-ml/pytorch
on aarch64 and imports cleanly. C++ (CMake); aarch64-portable.
commit b0b5ad4e0010966feb643ee1f3b88fbf85e15304
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jun 15 23:53:50 2026 +0200
sci-ml/torchcodec: generalize CUDA host-compiler comment
Drop the maintainer-private notes reference and host-specific wording
from the CUDAHOSTCXX rationale comment; the always-pin reason for the
unconditional caffe2 enable_language(CUDA) path is kept.
commit 6bc16c69bf040c094873257d0f0ea206939a701a
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jun 4 01:55:20 2026 +0200
sci-ml/torchcodec: drop 0.12.0
commit c8754f3464a3f65be63f276a89385482e4d0d792
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jun 4 01:55:01 2026 +0200
sci-ml/torchcodec: add 0.14.0
commit 5e74c9bc230193b302a220c743e12e883d573336
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 21 21:25:08 2026 +0200
sci-ml/torchcodec: bump to 0.13.0
commit c0669f2fe6e737feaf4f33e1ba078b5c16584b10
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon May 18 19:46:22 2026 +0200
sci-ml/torchcodec: follow upstream github move to meta-pytorch org
pytorch/torchcodec permanently redirects to meta-pytorch/torchcodec.
Update HOMEPAGE, SRC_URI, and metadata.xml remote-id + bugs-to.
Same tarball content at the new URL (Manifest unchanged).
commit f5dd46925d8564c9d7c6c800fbb3d57f2dabc200
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat May 16 10:42:01 2026 +0200
sci-ml/torchcodec: new package, add 0.12.0
Decode/encode video and audio into PyTorch tensors via FFmpeg.
Required runtime dep of sci-ml/pyannote-audio (input pipelines).
C++ extension; setup.py drives a custom CMake invocation against
system media-video/ffmpeg (the libtorchcodec_*<MAJOR>.so libs
are built per FFmpeg major present and dlopen'd at runtime). Host
ffmpeg-8.1.1 produces just the FFmpeg-8 lib variant.
CUDAHOSTCXX is pinned to g++-15 unconditionally because Caffe2's
cmake config (pulled in via find_package(Torch)) calls
enable_language(CUDA) whenever it finds /opt/cuda, even on
USE=-cuda builds; nvcc would otherwise reject the active gcc-16.
See the internal CUDA-13 host-compiler note.
Upstream's BUILD_AGAINST_ALL_FFMPEG_FROM_S3 path vendors the FFmpeg
binaries to dodge GPL questions on redistributed wheels; we link
against system media-video/ffmpeg directly via the
I_CONFIRM_THIS_IS_NOT_A_LICENSE_VIOLATION opt-out, which only
matters for redistributed binary wheels (not source builds).
CPU-only build verified; CUDA path gated behind USE=cuda for now,
not yet exercised in this overlay.
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Oct 8 18:13:19 2026 +0200
sci-ml/torchcodec: take the CUDA host compiler from cuda_gccdir
The ebuild exported CUDAHOSTCXX=/usr/bin/g++-15 without depending on
gcc 15, so configure failed where that file was missing. It did so
unconditionally, because Torch's CMake enables the CUDA language
whenever the toolkit is installed, CPU builds included.
cuda_gccdir returns the newest installed gcc the toolkit accepts: 15
or older before 13.4, up to 16 from 13.4. It needs the toolkit, so
call it only for USE=cuda or when the toolkit is on the build host.
Where it has no answer (non-GNU toolchain, CHOST other than
*-pc-linux-gnu) nvcc keeps its default.
No change to what an existing install provides; no revision bump.
commit 1e6588795a291b4bd69af55933dd1960076f112f
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Oct 1 08:51:52 2026 +0200
sci-ml/torchcodec: add 0.17.0
The release adds audio conversion sources, but the build knobs the
ebuild sets (TORCHCODEC_BUILD_IMAGE, ENABLE_CUDA, BUILD_VERSION and the
license guard) are unchanged, and the image decoders still compile as
stubs that raise at runtime, linking no image libraries.
Against a ROCm build of caffe2 2.14.0, configure fails for this and
for 0.16.0 alike: Torch's installed LoadHIP.cmake looks for clang++
under an unslotted LLVM path. That is a caffe2 bug, so it is fixed
there rather than worked around in each consumer.
commit 597dba530d4bb9da27a56c52a396f6909e5f5680
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Sep 16 12:38:19 2026 +0200
sci-ml/torchcodec: default the CUDA architecture list
With USE=cuda and no TORCH_CUDA_ARCH_LIST, Torch's CMake emits gencode
flags for every architecture it knows, starting at compute_50. CUDA 13
dropped Maxwell, so nvcc rejects that target and the build fails before
compiling anything (issue #292).
Default the list to compute capability 7.5 with PTX, and warn that it may
not suit the GPU in use, pointing at package.env for a better value. The
PTX matters: this default is what a machine with no visible GPU falls
back to, and a 7.5 cubin on its own will not load on a newer card, while
PTX lets the driver JIT for it. An explicit setting is still honoured, so
anyone who already targets their own GPU sees no change.
commit 2eaac282f30722575b5fb2b1915f908f78232419
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 05:01:29 2026 +0200
sci-ml/torchcodec: trim 0.15.0 comment prose
commit 42a1a0653abe86d4559aba7473c55598dfff6ce8
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 04:02:08 2026 +0200
sci-ml/torchcodec: trim 0.16.0 comment prose
commit 1b0c1489c13afc483fa7dd9664347b158c756aa9
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 15:05:53 2026 +0200
sci-ml/torchcodec: drop 0.14.0
commit f34813b3b7fe83c14f1a002bdb50b1d2d2549094
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 15:05:50 2026 +0200
sci-ml/torchcodec: add 0.16.0
commit db901676e150bb0eb1db885312a1ebefd959dd15
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jul 16 10:25:36 2026 +0200
sci-ml/torchcodec: drop 0.13.0
commit 9261025447dd225a8273274301d53d28865afa96
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jul 16 10:25:36 2026 +0200
sci-ml/torchcodec: add 0.15.0
commit d173d0a84d3c08b8bd915d963db82d472f43fb32
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Jun 20 15:49:49 2026 +0200
sci-ml/torchcodec: Keyword 0.13.0 for ~arm64
commit 2a96510d24f3c7bc8d0b5bddf08fa1cd529ec138
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jun 17 22:58:09 2026 +0200
sci-ml/torchcodec: Keyword 0.14.0 for ~arm64
Torch C++/ffmpeg media-decoding extension; builds against sci-ml/pytorch
on aarch64 and imports cleanly. C++ (CMake); aarch64-portable.
commit b0b5ad4e0010966feb643ee1f3b88fbf85e15304
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jun 15 23:53:50 2026 +0200
sci-ml/torchcodec: generalize CUDA host-compiler comment
Drop the maintainer-private notes reference and host-specific wording
from the CUDAHOSTCXX rationale comment; the always-pin reason for the
unconditional caffe2 enable_language(CUDA) path is kept.
commit 6bc16c69bf040c094873257d0f0ea206939a701a
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jun 4 01:55:20 2026 +0200
sci-ml/torchcodec: drop 0.12.0
commit c8754f3464a3f65be63f276a89385482e4d0d792
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jun 4 01:55:01 2026 +0200
sci-ml/torchcodec: add 0.14.0
commit 5e74c9bc230193b302a220c743e12e883d573336
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 21 21:25:08 2026 +0200
sci-ml/torchcodec: bump to 0.13.0
commit c0669f2fe6e737feaf4f33e1ba078b5c16584b10
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon May 18 19:46:22 2026 +0200
sci-ml/torchcodec: follow upstream github move to meta-pytorch org
pytorch/torchcodec permanently redirects to meta-pytorch/torchcodec.
Update HOMEPAGE, SRC_URI, and metadata.xml remote-id + bugs-to.
Same tarball content at the new URL (Manifest unchanged).
commit f5dd46925d8564c9d7c6c800fbb3d57f2dabc200
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat May 16 10:42:01 2026 +0200
sci-ml/torchcodec: new package, add 0.12.0
Decode/encode video and audio into PyTorch tensors via FFmpeg.
Required runtime dep of sci-ml/pyannote-audio (input pipelines).
C++ extension; setup.py drives a custom CMake invocation against
system media-video/ffmpeg (the libtorchcodec_*<MAJOR>.so libs
are built per FFmpeg major present and dlopen'd at runtime). Host
ffmpeg-8.1.1 produces just the FFmpeg-8 lib variant.
CUDAHOSTCXX is pinned to g++-15 unconditionally because Caffe2's
cmake config (pulled in via find_package(Torch)) calls
enable_language(CUDA) whenever it finds /opt/cuda, even on
USE=-cuda builds; nvcc would otherwise reject the active gcc-16.
See the internal CUDA-13 host-compiler note.
Upstream's BUILD_AGAINST_ALL_FFMPEG_FROM_S3 path vendors the FFmpeg
binaries to dodge GPL questions on redistributed wheels; we link
against system media-video/ffmpeg directly via the
I_CONFIRM_THIS_IS_NOT_A_LICENSE_VIOLATION opt-out, which only
matters for redistributed binary wheels (not source builds).
CPU-only build verified; CUDA path gated behind USE=cuda for now,
not yet exercised in this overlay.