AEUPH commited on
Commit
3e7cad9
·
verified ·
1 Parent(s): f2abc27

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -1
README.md CHANGED
@@ -9,4 +9,8 @@ license: mit
9
  short_description: A Neural Operating System running entirely on CPU.
10
  ---
11
 
12
- Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
 
 
 
 
9
  short_description: A Neural Operating System running entirely on CPU.
10
  ---
11
 
12
+ 🧠 NeuralOS: Monolith EditionHigh-Fidelity Generative Desktop Environment"Where latent space meets user space."NeuralOS is an experimental operating system kernel where the Video RAM (VRAM) is replaced by a generative latent space. It renders a functional Windows XP-style desktop entirely through Stable Diffusion 1.5, allowing for real-time UI generation, infinite customization, and neural-native application logic.⚡ Core ArchitectureThe system operates on a novel Latent-First architecture, bypassing traditional rasterization pipelines in favor of diffusion-based rendering.🧬 Neural DNA Driver StackTensor-Based UI Primitives: UI components (buttons, chrome, icons) are stored not as bitmaps, but as 1x4xHxW latent tensors.Dynamic DNA Injection: The kernel injects "perfect" programmatic frames into the latent space before diffusion, ensuring crisp edges and readable text while allowing AI to hallucinate the content.Resolution Scaling: Internally processes at 128x128 latent blocks, decoding to 1024x1024 high-fidelity output via the Tiny AutoEncoder (TAE).🚀 Optimization PipelineLCM-LoRA Integration: Utilizes Latent Consistency Models (LCM) to achieve 1-step inference, reducing generation time from seconds to milliseconds.Tiny AutoEncoder (TAE): Replaces the standard VAE with a distilled version, offering a 200x speedup in decoding latent tensors to pixel space.Spatial Masking: Implements "dirty rectangle" rendering in latent space—only regenerating regions of the screen that have changed state.🛠 System Modules🖥️ The Desktop EnvironmentA fully interactive React-based frontend embedded directly into the Python monolith.Real-Time Interaction:[x] Click-to-Spawn: Clicking the desktop instantly generates new application instances via the neural pipeline.[x] Live Taskbar: Functional taskbar with active process management and window switching.[x] Start Menu: Dynamic launcher for registered neural applications.Supported Applications:📝 Notepad: Generates crisp, white-paper text editors with context-aware content.🎨 Paint: Hallucinates creative canvases and tool palettes.💻 Command Prompt: Renders matrix-style terminal interfaces.📁 Explorer: Simulates file directory traversals visually.🧠 The Kernel (Monolith)Single-file architecture (app.py) combining the backend, frontend, and AI inference engine.Process Management: Custom OSKernel class managing PIDs, Z-ordering, and window focus states.Hybrid Rendering: Combines deterministic PIL drawing for structural integrity with probabilistic Diffusers generation for aesthetic richness.WebSocket Bus: Low-latency, bi-directional communication between the React frontend and the Python kernel.📦 Deployment (Docker)The system is packaged as a monolithic Docker image, optimized for Hugging Face Spaces.Dockerfile# Standard Deployment
13
+ FROM python:3.10-slim
14
+ COPY --chown=user . /app
15
+ CMD ["python", "app.py"]
16
+ 🔧 Environment VariablesVariableDescriptionDefaultUSE_LCMEnable 1-step diffusionTrueUSE_TAEEnable Tiny AutoEncoderTrueSTEPSInference steps per frame1🔮 Future RoadmapLatent Assembly (LASM): A bytecode language for manipulating latent tensors directly (e.g., LADD, LMUL).Persistent VFS: Saving generated windows and states to a persistent latent file system.LLM Integration: Connecting a local LLM (e.g., Qwen/Llama) to drive the OS logic and "thought" processes.