Show HN: Vanity, Recognition and Fighting Perfectionism – Buildlog for Git Vain https://ift.tt/1N2CpXm

Show HN: Vanity, Recognition and Fighting Perfectionism – Buildlog for Git Vain https://ift.tt/yRInxqX July 22, 2023 at 02:19PM

Show HN: I trained a 65B LLM on my texts to talk to myself (details inside) https://ift.tt/j3Fxlic

Show HN: I trained a 65B LLM on my texts to talk to myself (details inside) I trained the 65b model on my texts so I can talk to myself. It's pretty useless as an assistant, and will only do stuff you convince it to, but I guess it's technically uncensored? I'll leave it up for a bit if you want to chat with it. I posted this to Reddit and had several hundred people talking to it. Salient points from that discussion: LLAMA 1 65b Rank 128 5 epochs Batch size 1, 256 cutoff Trained in the Oobabooga suite using bitsandbytes 4-bit quantization for the lora Loss around 1.5 seems to give the most coherent results Trained on raw text dumps that is then parsed by a crappy Blazor Server app I threw together in a few hours. Text format is just "Sender:The Message\n" Trained on 2x 3090 Training took about 16 hours at a 90% power cap on the 3090's Trained on ~30k texts (I talk a lot, that was just 2 years) There's nothing telling it that it's a robot, though it sometimes seems to know It was largely inspired by the Unreal Engine lora tutorial I generated a list of fake names and addresses, pulled a list of my contacts, and then scripted out swapping the names and addresses for fictitious PII. I don't really send other sensitive data through text and my account is so thoroughly associated with my real name/location that the data leakage risk is manageable for the short period of time I'll have this available. It tends to halucinate fake PII as well which I think is partially a side effect of the data scrubbing. You'll notice it says things like that I live at 420 Ligma. I'll need to mix in some actual assistant tasks to the dataset before it will actually be useful as an assistant. Right now it's largely just for idle conversation. It's pretty ADHD and will randomly go off on its own tangents. I don't think it's the model. I think I just talk like that. Let me know if you have any questions or comments. I built it for myself, but figured I'll let the communities that have taught and entertained me so much play with it a little, too. Note: it says some pretty unhinged stuff. There's absolutely no guardrails. It also tends to talk like you're already friends with history. https://ift.tt/eP7rbO9 July 21, 2023 at 09:31PM

Show HN: Guiding LLM outputs using Zod https://ift.tt/FHS1KvI

Show HN: Guiding LLM outputs using Zod https://ift.tt/nV8qMsp July 22, 2023 at 02:32AM

Show HN: Datalake for Computer Vision Projects https://ift.tt/6uKTr1Y

Show HN: Datalake for Computer Vision Projects Buddhika, Kelum, and Chong Han here. We are building a self-hosted data infrastructure platform for computer vision. Our community page is https://ift.tt/swfbVh8 In the past, we worked on a couple of high-scale computer vision projects in retail, farming, and hospitals in various capacities. These projects involved 2D object sections, 3D object tracking, and more advanced 3D perception. Like other CV Engineers, we observed a common factor during these projects: one needs a large volume of high-quality data to build a production-deployable CV system. Our biggest challenge was not having a robust data infrastructure to handle large volumes of data. Our S3 buckets were like a data swamp; we had so much raw image and video in storage buckets without tracking. Instead of working on CV, we had to develop tools for data operations. We understand that many of us have our own custom scripts and stitch them together to make things happen in the CV pipeline. However, it is brittle and cumbersome to maintain. We wanted to build a system on top of the cloud buckets such as S3 that store all file indexes, labels, metadata attributes, inference outputs, model training outcomes, and literally anything related to machine learning/computer vision. This makes it possible for us to search for anything and consume efficiently. This behaves as a DataLake (by the way, "DataLake" is an overused term). All other downstream processes in the CV pipeline can access data more efficiently via SDK and can also return data back to the Lake (e.g., training/inference outcomes). The reason we made it self-hosted is to address data security and privacy concerns. Since data is fundamental to AI, we believe that companies and organizations should have complete control over it. Currently, we support AWS, GCP, and Azure cloud buckets; soon, we will support local storage. We ship this as a Docker container so you can just install it on any VM or local server. The installation script will do all the configuration automatically. The Python SDK and documentation are available but not perfect yet. We’ve launched this under MIT and Elastic licenses so any developer can use it. Our goal is not to charge individual developers. We make money by charging a license fee for things like multiple users, multiple buckets, scalability with K8, and providing support. Give it a try: https://ift.tt/swfbVh8 Let us know what you think. July 22, 2023 at 04:45AM

Show HN: TextToSample – Desktop App to generate audio with MusicGen locally https://ift.tt/kcN906P

Show HN: TextToSample – Desktop App to generate audio with MusicGen locally https://ift.tt/CxkM6rG July 21, 2023 at 09:23PM

Show HN: CopilotKit – a hackable OSS copilot for any react app https://ift.tt/CTMymx2

Show HN: CopilotKit – a hackable OSS copilot for any react app CopilotKit is a typescript library for adding a hackable copilot to any react app. You can let the copilot interact with your app via plain typescript closures , and give it (explicit) read access to app data. An example user interaction could look like: - "Which of these travel destinations has a rich architecture history?" - [Copilot answers] - "Great. Add these to my august trip folder, except the ones where it's typically rainy then" - [Added] Recursive agent integration (via Langchain) is in the works - if you have ideas I'd love to hear them here or on the discord. https://ift.tt/AnxHXCw July 22, 2023 at 12:29AM

Show HN: 3.0 Version of Invoke AI – open-source SD UI and Node-based Back end [video] https://ift.tt/qBrva7m

Show HN: 3.0 Version of Invoke AI – open-source SD UI and Node-based Back end [video] Hey all - Invoke started as one of the earliest Stable Diffusion UIs (you may remember it as “lstein”), and has evolved significantly into a full fledged react/typescript web app. We’ve been hard at work building a professional-grade backend to support our commercial move to serving businesses and enterprise with a hosted offering (invoke.ai), while keeping Invoke one of the best ways to self-host and create content as an open-core project. As of 3.0, all of the developments we’ve been working on and tweaking for our hosted environment are available to install and use locally, including an API and graph-based execution architecture - And, to demonstrate our commitment to free and open-source software, we’ve updated our license to the most explicitly permissive license available - Apache 2.0. — New SD Support in our 3.0 Version: - SDXL Support - We’ve implemented support the impending SDXL model architecture (And the current 0.9 model), and we’ll follow-up with streamlined SDXL support in the core UI interfaces once the 1.0 model is released. - ControlNet - Integrated support for the most popular ControlNet models, directly in the UI, with a simple processor preview and UI/UX. - Boards - Expanded gallery support to better organize and manage large scale images. Multi-select, drag & drop to anywhere in the UI, and backed by a local database to provide performant operations even when you’re thousands of images deep. - Expanded Schedulers - Rather than list all of them here… All the schedulers/samplers you know and love, with the ability to set favorites and disable those you’ll never use. - Model Flexibility - Swap your VAE on demand. Mix and match models as needed. Clip Skip. With the flexibility of the experimental Node Editor, you can even swap models mid-generation. - LoRA Enhancements - Full LoRA support (for all the Lo’s and Ly’s you can name), we’ve also added a mechanism which directly patches the model UNet on loading a LoRA. Test for yourself. - UI/UX Updates - Across the board, we’ve worked to clean up the UI, optimize the options panel for the most commonly used features, and left a number of the tiny little microinteractions across the app that make using Invoke easy for your workflow. It’s our best UI yet, hands down. And more is coming. - Node Editor (Experimental) - The main reason 3.0 took as long as it did (5 months!) is because we disassembled the entire backend of the application, and put it back together one “node” at a time. It’s clean, streamlined, and scalable. This sets us up to drive powerful advanced experiences for power users and gives an easy way for contributors to extend the capabilities of Invoke. The Node Editor that exposes all of the available functionality in the background is in an “experimental” status, for explorers and developers - Mainly because we have a lot of UI/UX polish we want to apply to it, and provide better ways to help less experienced users be successful with it! — *Up Next* More is coming. We have plans to extend on many of the core UI/UX experiences in the application, add more community resources for sharing new plug-ins/nodes, and will release the full version of the Node Editor in 3.1. Stay tuned! –- Whether you're a dev looking to build on or contribute to the project, a professional looking for pro-grade tools to incorporate into your workflow, or just looking for a great open-source SD experience, we're looking forward to you joining our community. You can get the latest version on GitHub https://ift.tt/xeAqhcj https://www.youtube.com/watch?v=A7uipq4lhrk July 21, 2023 at 10:33PM