Bug#1038161: ITP: xnvme -- Cross-platform libraries and tools for efficient I/O and low-level control.
Package: wnpp Severity: wishlist Owner: "Simon A. F. Lund" X-Debbugs-Cc: debian-devel@lists.debian.org, o...@safl.dk * Package name: xnvme Version : 0.7.0 Upstream Author : Simon A. F. Lund * URL : https://xnvme.io/ * License : BSD Programming Lang: C Description : Cross-platform libraries and tools for efficient I/O and low-level control. xNVMe provides a library to program storage devices efficiently from user space and tools to interact with them. xNVMe is strongly motivated by the emergence of storage devices providing I/O commands beyond those of read/write. This is the "NVMe" part of the name. The data plane or I/O layer is a minimal-cost abstraction on top of sync-io with thread pools, POSIX aio, Linux libaio, io_uring, io_uring_cmd, to name the most popular interfaces available via xNVMe on Linux. A control plane or Admin layer provides flexibility and low-level control via APIs for admin commands and device management. On Linux, these are implemented using interfaces such as ioctl(), io_uring_cmd, and vfio-pci. This is one key value point of xNVMe, providing a unified I/O and Admin API interface for storage devices on top of the myriad of interfaces available on todays operating systems. On top of these are command-line utilities, including "zoned" and "kvs". These provide NVMe-command-set interaction available at the fingertips of the command-line. The command-line utilities are related to those of nvme-cli. In this area we are actively working on combining the efforts on the command-line interface. On the library-side, then xNVMe is related to the before-mentioned I/O libraries, which it encapsulates. However, it goes beyond serving an abstraction on top, such that applications implemented using xNVMe can run on platforms other than Linux. Afaik. Then xNVMe is the only library providing a cross-platform storage programming interface. Supporting "traditional" storage and optimized for NVMe. This is the "x" part of the name for "cross" platform. I plan to maintain the package as part of the release process of xNVMe itself, thus making it an integral part of the CI to build, test, and verify the package in "lockstep" with the development of xNVMe. I am seeking help/guidance, possibly from a sponsor / co-maintainer.
Bug#1038205: ITP: basis-universal -- Basis Universal GPU Texture Codec
Package: wnpp Severity: wishlist Owner: Gürkan Myczko X-Debbugs-Cc: debian-devel@lists.debian.org * Package name: basis-universal Version : 1.16.4 Upstream Authors: Binomial LLC URL : https://github.com/BinomialLLC/basis_universal * License : Apache-2.0 Description : Basis Universal GPU Texture Codec This is a "supercompressed" GPU texture data interchange system that supports two highly compressed intermediate file formats (.basis or the .KTX2 open standard from the Khronos Group) that can be quickly transcoded to a very wide variety of GPU compressed and uncompressed pixel formats: ASTC 4x4 L/LA/RGB/RGBA, PVRTC1 4bpp RGB/RGBA, PVRTC2 RGB/RGBA, BC7 mode 6 RGB, BC7 mode 5 RGB/RGBA, BC1-5 RGB/RGBA/X/XY, ETC1 RGB, ETC2 RGBA, ATC RGB/RGBA, ETC2 EAC R11 and RG11, FXT1 RGB, and uncompressed raster image formats /565/.
Bug#1038206: ITP: jpeg-compressor-cpp -- jpeg compression library
Package: wnpp Severity: wishlist Owner: Matthias Geiger X-Debbugs-Cc: debian-devel@lists.debian.org, t...@debian.org, matthias.geiger1...@tutanota.de -BEGIN PGP SIGNED MESSAGE- Hash: SHA512 * Package name: jpeg-compressor-cpp Version : 104 Upstream Contact: Rich Geldreich * URL : https://github.com/richgel999/jpeg-compressor * License : Public Domain or Apache 2.0 Programming Lang: C/C++ Description : jpeg compression library I intend to package jpeg-compressor. It's needed for ppsspp where it's embedded as 3rdparty library. The package just consists of four headers, so it's fairly minimal. It wil be maintained under the collaborative debian/ space on salsa. tar has kindly agreed to sponsor the initial upload. thanks, werdahias -BEGIN PGP SIGNATURE- iQJUBAEBCgA+FiEEwuGmy/3s5RGopBdtGL0QaztsVHUFAmSMak0gHG1hdHRoaWFz LmdlaWdlcjEwMjRAdHV0YW5vdGEuZGUACgkQGL0QaztsVHV67g/7BwffKtkOtv1B J5jMu9egRsFpi8MbqywvibJVVMFSBHIaEA/RO6OI1bk+nm6E4+8wNMC7Cr+fy6FL rBaXIgP0/gg4bVRqJ+H6boxzimPcMPoymmIaNqmPV9VBzNL0dDwRndLy3T0ihivk QBNt3umhWcgKz/2MjA8+fqUPIHQanIQgnqEqAeLzJvMMYHOgFeijkAbwLCDYC1qL WjS28ODf5qP52Wn6Gql34Obf6iV3BYb9XKlCkeUN/ZPGtHIs6WjVLUVDIHrua1gW yGXD1bHV4B9hyT9z4W8H/85G+lwalEVamfm59hHM6JXG9hN0Zp3xG6uHlov8ivUD vUWB/DLY/0fZq4N5o7b8xMzDXSNP5mPgA9J9a5aWGdANSGXCOrb8N/m/ZKXS+LlL Rpqg0oxqLrFZTmvIzZnNbN/ooviKvLrHrHVQq9eY+VpAOoU/IvFUUgNqXq6y9nq5 i6HFd8c4ZN3HsvJfdupap/BYLr6QrtiZ47mppne7pkxadhIGyn7kWokG7ckAZZC8 SAD/IbnRko7J/Jn++AR8DbCGLJbbzshxnXjXtd7gcQ1aByZUzY8D5Loyy5TBDqwM W6cXt8knxGKPUP5FLN2YAE6x9yFqmo0dSjCV6cNa9EH1jQsOfCKeeWJqfDnZvvyi txp1K1n9y3DHu2DCp+XTRxLvJQmZuok= =+I09 -END PGP SIGNATURE-
Bug#1038326: ITP: transformers -- State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow (it ships LLMs)
Package: wnpp Severity: wishlist Owner: Mo Zhou X-Debbugs-Cc: debian-devel@lists.debian.org, debian...@lists.debian.org * Package name: transformers Upstream Contact: HuggingFace * URL : https://github.com/huggingface/transformers * License : Apache-2.0 Description : State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow I've been using this for a while. This package provides a convenient way for people to download and run an LLM locally. Basically, if you want to run an instruct fine-tuned large language model with 7B parameters, you will need at least 16GB of CUDA memory for inference in half/bfloat16 precision. I have not tried to run any LLM with > 3B parameters with CPU ... that can be slow. LLaMa.cpp is a good choice for running LLM on CPU, but that library supports less models than this one. Meanwhile, the cpp library only supports inference. I don't know how many dependencies are still missing, but that should not be too much. Jax and TensorFlow are optional dependencies so they can be missing from our archive. But anyway, I think running a large language model locally with Debian packages will be interesting. The CUDA version of PyTorch is already in the NEW queue. That said, this is actually a very comprehensive library, which provides far more functionalities than running LLMs. Thank you for using reportbug
