Skip to main content

Show HN: Zipslicer, a library for loading LLM checkpoints on consumer hardware https://ift.tt/7idfN5M

Show HN: Zipslicer, a library for loading LLM checkpoints on consumer hardware This is a low-level opensource library I developed for my own use and decided to share, as it makes it possible to process large checkpoints of neural networks without renting high-RAM instances, on a regular PC. It replaces torch.load() with a custom function that produces a dictionary that materializes tensors on the fly. Compared to other solutions it doesn't require sharding or re-encoding checkpoints and uses them completely as-is. It is a foundation to make it possible to run inference and compress language models and other large models one layer at a time - in principle, even one tensor at a time. I describe the rationale and technical details of the library's design in the blogpost: https://ift.tt/tydo8Uh https://ift.tt/OPafXBl March 4, 2023 at 01:29AM

Comments

Popular posts from this blog

Women Pioneers at Muni: Adeline Svendsen and Muni’s First Newsletter

Women Pioneers at Muni: Adeline Svendsen and Muni’s First Newsletter By Jeremy Menzies To close out Women’s History Month, here’s a look back at one woman whose work to bring Muni staff together in the late 1940s created a legacy that lives on to this day. Adeline “Addy” Svendsen was founding editor of Muni’s first internal newsletter, “ Trolley Topics .” Adeline Svendsen sits at her desk in the Geneva Carhouse office building in this 1949 shot. Trolley Topics was a new venture when it started in February 1946. As Svendsen wrote in the first issue it was created, “to bring a little fun, a little news, and a lot of good will to all our fellow employees in the Railway.” Just two years prior in 1944, Muni merged with the Market Street Railway Company, expanding the small municipal operation into the largest transit provider in the city with hundreds of employees, vehicles of every shape and size, and dozens of facilities scattered across town. The newsletter was meant to help unite ...

Show HN: StreetComplete, an OpenStreetMap Editor for Humans https://ift.tt/2J8IL02

Show HN: StreetComplete, an OpenStreetMap Editor for Humans StreetComplete is an OpenStreetMap[0] editor directed at people who want to contribute and want to do this using their smartphone, without learning how to edit things[1]. It is available as an Android application. It is intended to be used as one walks, with quests appearing as markers on the map. Selecting a marker allows one to answer a simple question. The answer will be added to the OpenStreetMap database, with app handling selecting objects for editing, transforming answer into OSM tags and making edits. OpenStreetMap account is needed to apply edits, but it is possible to start without it, make some edits and login/register later. Note: I am not the main author, but I am one of the active contributors. Github page is at https://ift.tt/2g8lasH and https://ift.tt/3nR9PzS shows what was recently released. [0]OpenStreetMap is a Wikipedia of maps, available on the open licence. This dataset is already used for many interestin...

Show HN: Launch VM workloads securely and instantaneously, without VMs https://ift.tt/2QwJ1Kd

Show HN: Launch VM workloads securely and instantaneously, without VMs Hello HN! We've been working on a new hypervisor https://kwarantine.xyz that can run strongly isolated containers. This is still a WIP, but we wanted to give the community an idea about our approach, its benefits, and various use cases it unlocks. Today, VMs are used to host containers, and make up for the lack of strong security as well as kernel isolation in containers. This work adds this missing security piece in containers. We plan on launching a free private beta soon. Meanwhile, we'd deeply appreciate any feedback, and happy to answer any questions here or on our slack channel. Thanks! April 29, 2021 at 07:50AM