Update README.md

README.md

@@ -1,20 +1,196 @@
+---
+language:
+  - pt
+  - en
+  - es
+  - fr
+  - ar
+tags:
+  - llama.cpp
+  - vulkan
+  - local-ai
+  - gguf
+  - amd
+  - rx580
+  - stable-diffusion
+  - ollama
+  - openwebui
+  - hardware-revival
+  - offline
+  - inference
+license: mit
+library_name: llama-cpp-python
+pipeline_tag: text-generation
+---
+
 # AIVisionsLab — RX580 Vulkan Stack
 
-Modular Local AI Infrastructure optimized for legacy hardware. 
-*“O hardware não morre, ele se transforma.”*
+> *"O hardware não morre, ele se transforma."*
+
+Complete local AI infrastructure for **AMD RX 580 8GB (Polaris/GCN4)** using **Vulkan** as the compute backend — no CUDA, no ROCm, no cloud, no new hardware.
+
+---
+
+## What This Is
+
+This is not a fine-tuned model. This is a **documented stack** — configuration files, build scripts, benchmarks, and guides — for running local AI inference on a 2017-era AMD GPU that mainstream frameworks consider unsupported.
+
+AMD dropped official ROCm support for GCN4/Polaris on Windows. This project proves that `llama.cpp` + Vulkan fills that gap completely.
 
 ---
 
-## 🛠 Acesso Rápido
-- **[Documentação Master](https://setup-ia-local-rx580-vulkan.web.app/)** — O portal do projeto.
-- **[Leia o Manifesto do Laboratório](MANIFESTO.md)** — A filosofia por trás do código.
+## Stack
+
+| Component | Tool | Notes |
+|-----------|------|-------|
+| LLM Inference | `llama.cpp` (Vulkan build) | Compiled with `-DLLAMA_VULKAN=ON` |
+| Model serving | `Ollama` | Auto-detects Vulkan on AMD |
+| Chat interface | `OpenWebUI` | Docker, runs at localhost:3000 |
+| Image generation | `stable-diffusion.cpp` | Vulkan backend, SD 1.5 native |
+| Image interface | `ComfyUI` | WSL2, CPU stable mode |
+| Advanced images | `Flux.1 Schnell` | CPU+GPU hybrid mode |
+
+---
+
+## Hardware Tested
+
+```
+GPU     : AMD RX 580 8GB (Polaris 20 / GCN4)
+CPU     : Intel Xeon E5-2670 v3 (12c/24t @ 2.3GHz)
+RAM     : 32GB DDR4 ECC
+Storage : NVMe SSD (models)
+OS      : Windows 11 + WSL2 (Ubuntu 22.04)
+Driver  : AMD Adrenalin (latest)
+```
 
 ---
 
-## Hardware Configuration
+## Recommended Models
+
+### LLMs (GGUF Q4_K_M)
+
+| Model | Size | Speed on RX 580 | Use Case |
+|-------|------|-----------------|----------|
+| Llama 3.2 3B | ~2GB | ~18 tok/s | Fast general use |
+| Mistral 7B | ~4GB | ~9 tok/s | Best quality/speed |
+| Qwen2.5 7B | ~4GB | ~8 tok/s | Portuguese / multilingual |
+| Phi-3 Mini | ~2GB | ~20 tok/s | Low RAM machines |
+| CodeLlama 7B | ~4GB | ~9 tok/s | Code generation |
+
+### Image Generation
+
+| Model | Backend | Speed | Notes |
+|-------|---------|-------|-------|
+| SD 1.5 512x512 | Vulkan GPU | ~8s/img | 20 steps, native |
+| SDXL | Vulkan GPU | ~45s/img | Possible but slow |
+| Flux.1 Schnell | CPU+GPU | ~3min/img | High quality |
+
+---
+
+## Quick Start
+
+### 1. Install Ollama
+```bash
+# Download from https://ollama.com
+ollama pull llama3.2
+ollama run llama3.2
+```
+
+### 2. Run OpenWebUI
+```bash
+docker run -d \
+  -p 3000:8080 \
+  --add-host=host.docker.internal:host-gateway \
+  -v open-webui:/app/backend/data \
+  --name open-webui \
+  --restart always \
+  ghcr.io/open-webui/open-webui:main
+```
+Open `http://localhost:3000` — connect Ollama at `http://host.docker.internal:11434`
+
+### 3. Build llama.cpp with Vulkan
+```bash
+git clone https://github.com/ggerganov/llama.cpp
+cd llama.cpp
+cmake -B build -DLLAMA_VULKAN=ON
+cmake --build build --config Release -j8
+
+./build/bin/llama-server \
+  --model ./models/mistral-7b-q4_k_m.gguf \
+  --n-gpu-layers 35 \
+  --port 8080
+```
+
+### 4. Build stable-diffusion.cpp with Vulkan
 ```bash
-GPU        : RX 580 8GB
-Backend    : Vulkan
-Inference  : llama.cpp
-Generation : Flux GGUF
-UI         : OpenWebUI
+git clone https://github.com/leejet/stable-diffusion.cpp
+cd stable-diffusion.cpp
+cmake -B build -DSD_VULKAN=ON
+cmake --build build --config Release
+
+./build/sd-server \
+  --model ./models/v1-5-pruned-emaonly.safetensors \
+  --port 7860
+```
+
+---
+
+## Windows Firewall Fix
+
+Docker can't reach Ollama by default. Run as Administrator:
+
+```powershell
+New-NetFirewallRule `
+  -DisplayName "Allow Docker to Ollama" `
+  -Direction Inbound `
+  -Action Allow `
+  -Protocol TCP `
+  -LocalPort 11434 `
+  -RemoteAddress 172.16.0.0/12
+```
+
+---
+
+## Startup Script
+
+```batch
+@echo off
+:: Start Ollama if not running
+tasklist /FI "IMAGENAME eq ollama.exe" 2>NUL | find /I "ollama.exe" >NUL
+if errorlevel 1 start "" "%LOCALAPPDATA%\Programs\Ollama\ollama.exe" serve
+
+:: Start OpenWebUI
+timeout /t 3 >NUL
+docker start open-webui
+
+echo AI Stack ready at http://localhost:3000
+start http://localhost:3000
+```
+
+---
+
+## Why Vulkan?
+
+| | CUDA | ROCm | **Vulkan** |
+|---|---|---|---|
+| NVIDIA support | ✅ | ❌ | ✅ |
+| AMD support (modern) | ❌ | ✅ | ✅ |
+| AMD RX 580 Windows | ❌ | ❌ | **✅** |
+| Open standard | ❌ | ✅ | ✅ |
+| Setup complexity | Medium | High | **Low** |
+
+---
+
+## Links
+
+- 📖 [Full Documentation Portal](https://setup-ia-local-rx580-vulkan.web.app/)
+- 💻 [GitHub Repository](https://github.com/aivisionslab-studios/rx580-local-ai-guide)
+- 🎥 [YouTube Channel](https://youtube.com/@aivisionslab-hub)
+
+---
+
+## License
+
+MIT — use it, fork it, share it.
+
+*Built in São Paulo, Brazil. Tested on hardware from 2014–2017.*