Article URL: https://engineering.drawthings.ai/p/making-apple-neural-engine-work-in Comments URL: https://news.ycombinator.com/item?id=47798261 Points: 1 # Comments: 0
Hacker News, Training custom wake words like "Hey Alexa" is often a resource-intensive task, demanding powerful hardware and complex manual tuning. NanoWakeWord is an open-source framework designed to solve this. It features an intelligent engine that automates the ML pipeline, making it possible to build high-performance, production-ready wake word models with minimal effort. What makes it different: Train Anywhere, On Anything: The core architecture is built for extreme efficiency. You can train on massive, terabyte-scale datasets using a standard laptop or even a low-spec machine, all without needing a GPU. This is achieved through memory-mapped files that stream data directly from disk, eliminating RAM limitations. Intelligent Automation: The framework analyzes your data to automatically configure an optimal model architecture, learning schedule, and training parameters. It removes the guesswork from building a robust model. Total Flexibility and Control: While it automates everything, it also offers deep customization. You can choose from 11+ built-in architectures (from lightweight DNNs to SOTA Conformers) or easily extend the framework to add your own custom architecture. Every single parameter generated by the engine can be manually overridden for full control. Smarter Data Processing: It moves beyond generic negatives. The system performs phonetic analysis on your wake word to synthesize acoustically confusing counter-examples, which drastically reduces the false positive rate in real-world use. Ready for the Edge: Models are exported to the standard ONNX format. The framework also includes a lightweight, stateful streaming inference engine designed for low-latency performance on devices like the Raspberry Pi. Try It in Your Browser (No Install Needed): This single Google Colab notebook is a playground to train your first model. Inside, you can select and experiment with any of the available architectures with just a few clicks. Launch the Training Notebook: https://colab.research.google.com/github/arcosoph/nanowakewo... The goal is to produce models with an extremely low false positive rate (tests show less than one false activation every 16-28 hours on average). The project is actively developed by Arcosoph, and all feedback or questions are highly welcome! Key Links: GitHub Repo: https://github.com/arcosoph/nanowakeword PyPI Package: https://pypi.org/project/nanowakeword/ Pre-trained Models: https://huggingface.co/arcosoph/nanowakeword-models#pre-trai... Discord Community: https://discord.gg/rYfShVvacB Comments URL: https://news.ycombinator.com/item?id=47794771 Points: 1 # Comments: 0
Hi HN, I'm Alon, and I'm building Alien ( https://alien.dev ), an open-source platform for deploying your software into your customers' cloud accounts - AWS, GCP, or Azure — and keeping it fully managed. In my previous startup, I heard the same question from every single enterprise customer over and over again: "My data is sensitive. Can I deploy your product to my own cloud account?" Every founder I talk to who's building anything in AI or security hits the same wall. To solve this, many teams create a self-hosted version of their product. They send a Docker image or an Helm chart to the customer and let them install the entire product on their side. While self-hosting is great (and will continue to be important!), it has 2 problems: 1. Enterprise customers are forced to operate third-party software and own deployments, upgrades, and security risks. In most cases they don't want that. They prefer a managed experience, with no data leaving their environment. 2. Even with self-hosting, vendors are still accountable when things break, but they have little to no visibility. When something breaks - and it always does - you're on a 2am Zoom call screen share debugging blind because you have no access. No auto-updates, no logs, every customer is on a different version. That's why many successful SaaS companies that deal with sensitive data like Databricks, Wiz, and others spent years building internal infrastructure to automatically deploy, update, and monitor their software across AWS, GCP, and Azure. It's a win-win: no sensitive data leaves the customer’s environment, and the software is still fully managed by the vendor. Alien manages deployments across every customer's cloud through cloud APIs — no network connection to their environment needed. The mental model is like sharing a Google Drive folder: the customer grants least-privilege IAM access to an isolated area in their cloud, you manage what's inside, they can revoke it anytime. The whole thing is written in Rust and works across AWS, GCP, Azure, and locally from a single codebase. You can get started here: https://alien.dev/docs/quickstart Here's how it works: https://alien.dev/docs/how-alien-works GitHub: https://github.com/alienplatform/alien Excited to share Alien with everyone here – let me know what you think! Comments URL: https://news.ycombinator.com/item?id=47791745 Points: 2 # Comments: 0
We built a custom (experimental) file transformation pipeline into our agent because we don't want to be at the mercy of LLM providers for giving models native access to files. Essentially, we extend the agent to expose files to the LLMs context through a standardized pipeline, allowing to convert any file into a file type specific best representation that only consist out of text, image (and audio/video) parts. Comments URL: https://news.ycombinator.com/item?id=47784889 Points: 1 # Comments: 0
Article URL: https://github.com/matryer/hatch Comments URL: https://news.ycombinator.com/item?id=47782591 Points: 2 # Comments: 0
I got frustrated with Gmail's lack of customization, and built an extension for myself. It doesn't do much other than modify how Gmail is displayed in Chrome. Does it as minimally as possible (that I can engineer). Done locally, zero remote processing. Currently built the following settings (want to add more soon): Conversations: * Show newest message first * Show every message in a thread (sometimes they're collapsed into numbered bubbles) * Remove AI overviews (retches) Layout/UI: * Minimize the search & top bar * Minimize the far left "rail" (Mail/Chat/Meet) * Hide a bunch of extraneous icons/buttons/items * Adjust width of Folders/Labels column (still in-process) Mini-Automations: * Automatically open to inbox on page start/refresh * Automatically select first message if the reading pane is open (unless unread, or previous message already selected) * Add a recurring reminder (if wanted) to use Gmail's new Manage Subscriptions folder/feature Bunch more features I'd like to add. Let me know if you think I should change anything. Or if you have any code/bug feedback... :) [email protected] Comments URL: https://news.ycombinator.com/item?id=47757667 Points: 1 # Comments: 1
我现在在ccswitch的配置 model_provider = “custom” model = “gpt-5.4” service_tier = “fast” network_access = “enabled” windows_wsl_setup_acknowledged = true model_reasoning_effort = “xhigh” disable_response_storage = true model_context_window = 1000000 model_auto_compact_token_limit = 900000 [model_providers] [model_providers.custom] name = “custom” wire_api = “responses” requires_openai_auth = true base_url = “ https://ai.spring.top ” 通过接口直接调用是可以的 sub2api里测试也是可以的 2 个帖子 - 2 位参与者 阅读完整话题
github.com GitHub - Aethersailor/Custom_OpenClash_Rules: 分流完善的 OpenClash 订阅转换模板，搭配保姆级 OpenClash... 分流完善的 OpenClash 订阅转换模板，搭配保姆级 OpenClash 设置教程，无需套娃其他插件即可实现完美分流、DNS无污染无泄漏，且快速的国内外上网体验，配套自动化域名规则提交机器人。 有其他的规则推荐吗 5 个帖子 - 5 位参与者 阅读完整话题