Friday, March 3, 2023
Show HN: Zipslicer, a library for loading LLM checkpoints on consumer hardware https://ift.tt/DsnxjRz
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/7PRke1O https://ift.tt/7lOXZvE March 3, 2023 at 11:29PM
Thursday, March 2, 2023
Show HN: Vaulty – No more passwords in emails https://ift.tt/CFIyXEP
Show HN: Vaulty – No more passwords in emails https://ift.tt/YZeXKBo March 2, 2023 at 07:43PM
Show HN: Sort Any Awesome List by GitHub Stars https://ift.tt/So5uYpA
Show HN: Sort Any Awesome List by GitHub Stars As a CS undergrad, I self-taught many topics not covered by the school. Typically awesome list is where I started with. I spent my second half in college researching computer vision and machine learning. CVPR accepts over 2000 papers every year, not to mention arXiv. The awesome lists curated by the community usually serve as checklists when I was doing literature surveys. But since the number of the list items was overwhelming, I had to prioritize which papers to read first. Many of them had been uploaded to arXiv in the last few months, so I had no clue how many citations they would get. When learning web development, I also had difficulty finding the best package for the tasks like data validation and client-side routing. To make a decision, I had to browse through GitHub repositories or read the documentation. I wish I could have my awesome lists sorted, desirably by popularity. This led to the creation of my project: a web app that rearranges any awesome list with hints from GitHub stars. It may not work well when the list items do not contain links to GitHub repositories or pages, but it truly shines when most of them do. I hope this could help anyone else. P.S. I'm aware of a similar project ( https://ift.tt/2KjTzJM ), a CLI application with more features. For example, it checks if any GitHub repository link exists on the linked website. https://ift.tt/ERtkQnT March 2, 2023 at 10:51PM
Subscribe to:
Posts (Atom)