Skip to content

Releases: securenrn-del/Repetere-AI_Beta

Release list

Beta .0.5.1 - Audio Pipeline fixes

Beta .0.5.1 - Audio Pipeline fixes Pre-release
Pre-release

Choose a tag to compare

View all tags
@securenrn-del securenrn-del released this 11 Oct 18:24
Beta-Release-0.5.1
1f79471
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode. 
# Repetere AI (v0.5.1-beta)

**Author:** Nathaniel Richardson Jr.  
**Repository:** [https://github.com/securenrn-del/repetereai](https://github.com/securenrn-del/repetereai)  
**Binary Distribution:** Standalone 1-Click Setup (`RepetereAI-Setup-v0.5.1-beta.exe`) via [GitHub Releases](https://github.com/securenrn-del/repetereai/releases)[cite: 2, 6]

> ⚠️ **BETA NOTICE:** This is an active preview workstation engineered strictly for local execution on NVIDIA CUDA hardware. Local generation requires dedicated processing time—especially for neural audio decoding and weight streaming. Image diffusion is currently an interface layout preview (backend rendering lands in v0.6.0)[cite: 2]. If you encounter hardware-specific edge cases, please submit an issue on GitHub.

**Repetere AI** is a native Windows C++ multimodal desktop workstation engineered to restore computing sovereignty[cite: 6]. Built directly on low-level Win32, Dear ImGui, GLFW, and OpenGL 3 runtimes, it eliminates cloud subscriptions, telemetry, tracking, and webview wrapper memory bloatware[cite: 6]. It delivers private, uncensored intelligence directly on consumer NVIDIA GeForce RTX hardware[cite: 6].

---

## ⚡ System Requirements

| Component | Minimum Specification | Recommended Specification |
| :--- | :--- | :--- |
| **Operating System** | Windows 10 / 11 (64-bit) | Windows 11 (64-bit) |
| **GPU Acceleration** | NVIDIA RTX 30-Series (8 GB VRAM) | NVIDIA RTX 40/50-Series (12 GB+ VRAM)[cite: 6] |
| **Display Drivers** | NVIDIA Game Ready / Studio 528+ (CUDA 12+) | Latest NVIDIA Studio Drivers |
| **System Memory** | 16 GB Physical RAM[cite: 6] | 32 GB Physical RAM (Required for Hybrid Mode) |
| **Target Storage** | 20 GB free space on SSD | 50 GB+ free space on NVMe SSD |

---

## 🚀 Quick Start: Installation & First Launch

1. Download **`RepetereAI-Setup-v0.5.1-beta.exe`** from [GitHub Releases](https://github.com/securenrn-del/repetereai/releases)[cite: 2, 6].
2. Run the installer[cite: 5]. It installs cleanly into `%LOCALAPPDATA%\Programs\Repetere AI\` without requiring administrator privileges[cite: 5].
3. Launch **Repetere AI** from your Desktop shortcut or the Start Menu[cite: 1, 5].
4. Click anywhere on the startup splash screen to enter[cite: 2, 5].
5. The **Hardware Configuration Diagnostic** modal will profile your GPU VRAM and system memory[cite: 1, 6]. Click **Confirm Hardware** to enter the main workstation[cite: 4].

---

## 📦 How to Download & Manage AI Models

AI models are the local weights that power each workspace. They must be present on disk before generation can proceed[cite: 2, 5].

### In-App Downloader (Recommended):
1. In the top menu bar, click **Settings ➔ Repetere Models**[cite: 2, 4].
2. The catalog overlay lists verified models tuned for your hardware[cite: 1, 4]:
   * **Coding Workspace:** *Qwen 2.5 Coder 7B Instruct* (or *14B*)[cite: 1]
   * **General Chat Workspace:** *Meta Llama 3.1 8B Instruct*[cite: 1, 2]
   * **Creative Writing Workspace:** *Hermes 3 Llama 3.1 8B*[cite: 1, 2]
   * **Audio / Music Studio (Both Required):**
     * *YuE2 3B Music & Vocal Engine* (`yue2-3b-q4_0.gguf`, ~2.48 GB)[cite: 1, 4]
     * *YuE2 Neural Audio VAE Decoder* (`yue2-vae-f16.gguf`, ~265 MB)[cite: 1, 4]
3. Click **[ Download ]**[cite: 1, 4]. The in-app manager streams chunks via CPR with real-time transfer speeds and validates the 4-byte GGUF magic header (`0x46554747`)[cite: 6, 8, 10].
4. When complete, the badge turns green (**Installed**). Click **[ Return to Workspace ]**[cite: 4, 8, 18].

### Manual Weight Placement (Advanced Users):
You can place `.gguf` weight files directly into your local installation directory:
```text
%LOCALAPPDATA%\Programs\Repetere AI\models\

(Ensure filenames match the catalog schema exactly so internal CLI arguments resolve cleanly).

🧭 Navigating the Workspaces

Repetere AI partitions workflows across five isolated environments:

1. CODING

  • Engine Target: Qwen 2.5 Coder 7B / 14B / 32B Instruct (Q4_K_M)

  • Syntax Presentation: Automatically extracts triple-backtick language blocks (```) into horizontal-scrolling containers with snippet-level Copy buttons.

  • Quick Response Copy: Single-click [ Copy Full Response ] action directly beneath assistant outputs for rapid clipboard export.

  • Context Preservation: Multi-turn message serialization preserves conversational history across iterative debugging sessions.

2. CHAT

  • Engine Target: Meta Llama 3.1 8B Instruct (Q4_K_M)

  • General Intelligence: Tuned for logical reasoning, document summarization, and interactive assistance.

  • Memory Reset: The red Clear Chat button archives transcripts to disk while resetting active VRAM context.

3. CREATIVE WRITING

  • Engine Target: Hermes 3 Llama 3.1 8B / Narrative checkpoints

  • Uncensored World-Building: Tailored for complex narrative progression, scene development, scriptwriting, and dialogue without cloud filtering.

4. AUDIO (MUSIC) Production Studio

  • Engine Target: YuE2 Neural Audio Synthesis Engine (yue2-3b-q4_0.gguf / yue2-3b-q8_0.gguf + yue2-vae-f16.gguf)

  • Full-Song Vocal Synthesis: Generates full vocal and instrumental tracks based on style tags (e.g., Alternative Rock, 110 bpm, driving drums) and structured lyrics ([Verse], [Chorus], [Bridge]) with generation durations configurable up to 300 seconds.

  • Expanded Lyrics Buffer: 8192-byte input memory buffer (Types.h) preventing text truncation on long multi-verse compositions.

  • Input Sanitization: Automatically sanitizes quotes and flattens multi-line breaks to prevent Windows cmd.exe subshell argument chopping.

  • Dynamic Model Discovery: Embedded engine scanner detects any installed YuE2 quantization (prioritizing q8_0 > q5_k_m > q4_k_m > q4_0) alongside yue2-vae-f16.gguf without hardcoded path dependencies.

  • Smart File Serialization: Tracks are saved as [StyleTag]_[Word1]_[Word2]_[Timestamp].wav in output/music/.

  • Master Transport: Play output tracks immediately via Win32 system shell integration to your default media player.

5. IMAGES (DIFFUSION) Studio (Preview)

  • Status: ⚠️ Visual Layout Preview (Live Diffusion Engine scheduled for v0.6.0).

  • Current Features: Dual positive/negative prompt buffers, step/CFG sliders, resolution presets, and native file dialogs (OpenNativeFileDialog) for base reference images.

⚙️ Hardware Compute Modes

  • GPU Mode (Active Default): Confines execution strictly to dedicated GPU VRAM for maximum token throughput and minimum latency.

  • HYBRID Mode (Interface Preview): Top-bar toggle with confirmation dialogs. Spans weights across system RAM and VRAM for running 14B–32B models on lower-tier cards (full dynamic layer allocation unlocks in v0.6.0).

🔒 Conversation History & Persistence

  • Local JSON Persistence: Transcripts serialize to chat_history.json on context clearing or application shutdown, supporting dynamic model tags across sessions.

  • Transcript Viewer: Accessible via History ➔ Open History with foreground z-index prioritization (ImGui::SetNextWindowFocus()).

  • Context Restoration: Single-click [ Load Conversation into Active Tab ] restores archived discussions directly into active working memory.

  • History Wipe Guard: Deletion actions trigger a modal confirmation dialog to prevent accidental data loss.

🛠️ What's New & Fixed in v0.5.1-beta

  • Dynamic YuE2 Model Discovery: The audio engine scanner (FindBestYueModel) recursively discovers any installed YuE2 quantization (including yue2-3b-q4_0.gguf, yue2-3b-q8_0.gguf, etc.) alongside yue2-vae-f16.gguf instead of crashing on a hardcoded path.

  • Robust Subshell Argument Sanitization: Hardened CLI string formatting for audiocpp_cli.exe. Quotes (", ') and raw newlines are stripped and escaped to eliminate cmd.exe subshell truncations and exit code failures.

  • Expanded Lyrics Memory Buffer: Increased lyricsBuffer in Types.h to 8192 bytes, allowing complete multi-verse arrangements, tags, and bridges to be processed without buffer overflow.

  • Multi-Turn Context Serialization: InferenceClient serializes prior conversational message turns into request payloads, preserving iterative context memory across turns.

  • Universal Model History Persistence: Updated HistoryManager JSON serialization routines to support dynamic model tags (such as Meta Llama 3.1 and Hermes 3) rather than hardcoded identifiers.

  • Safe Music Generation Guard: Added UI validation in RenderMusicStudioWorkspace that disables synthesis buttons and alerts the user if required YuE2 weights or VAE files are missing.

  • Foreground History Modal Snap: Corrected z-indexing and window focus flags (ImGui::SetNextWindowFocus()) so the transcript overlay consistently opens in the foreground.

  • Linker & Object Collision Fixes: Resolved duplicate symbol linker errors (LNK2005) between FREEAI.obj and Theme.obj by isolating stb_image implementation definitions.

  • High-DPI 1600p+ Viewport Optimization: Preserved scaled 24pt/28pt Segoe UI font metrics, auto-maximizing window commands, and 1.15× responsive layout styling (ImGuiStyle::ScaleAllSizes).

  • Optimized Inno Setup Packaging: Configured parallelized LZMA2 solid compression (LZMANumBlockThreads=8) excluding legacy internal runtime directories to deliver a fast compile and an installer size (~1.29 GB) within release thresholds.

📜 License & Acknowledgments

Authored by Nathaniel Richardson Jr. Distributed under the MIT License. Special thanks to the open-source communities powering [[llama.c...

Read more
Loading

Repetere AI - Beta .05

Repetere AI - Beta .05 Pre-release
Pre-release

Choose a tag to compare

View all tags
@securenrn-del securenrn-del released this 10 Oct 03:13
Beta-Release
2bd6645
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode. 
# Repetere AI (v0.5-beta)

**Author:** Nathaniel Richardson Jr.  
**Repository:** [https://github.com/securenrn-del/repetereai](https://github.com/securenrn-del/repetereai)  
**Binary Distribution:** Standalone 1-Click Setup (`RepetereAI-Setup-v0.5-beta.exe`) via [GitHub Releases](https://github.com/securenrn-del/repetereai/releases)
**THIS IS THE FIRST BETA! It might not run on your MACHINE. Remember that LOAL AI isn't instant it needs time to generate your music and download / load your models! Submit an issue if you run into one! Image generation is not currently working.  You must download BOTH music Files. This is currently for CUDA only.

**Repetere AI** is a completely private, native Windows desktop workstation that runs AI directly on your computer's graphics card (GPU). It eliminates cloud subscriptions, accounts, telemetry, and webview bloatware, running 100% offline and uncensored.

---

## ⚡ System Requirements

| Component | Minimum Specification | Recommended Specification |
| :--- | :--- | :--- |
| **Operating System** | Windows 10 or 11 (64-bit) | Windows 11 (64-bit) |
| **Graphics Card (GPU)** | NVIDIA RTX 30-Series (8 GB VRAM) | NVIDIA RTX 40/50-Series (12 GB+ VRAM) |
| **Display Drivers** | NVIDIA Driver 528+ (CUDA 12+) | Latest NVIDIA Studio or Game Ready Driver |
| **System Memory (RAM)** | 16 GB Physical RAM | 32 GB Physical RAM |
| **Storage** | 20 GB free space on SSD | 50 GB+ free space on NVMe SSD |

---

## 🚀 Quick Start: Installation & First Launch

1. Go to the [Releases](https://github.com/securenrn-del/repetereai/releases) page and download **`RepetereAI-Setup-v0.5-beta.exe`**.
2. Run the installer. It will install directly into `%LOCALAPPDATA%\Programs\Repetere AI\` without requiring administrator privileges.
3. Launch **Repetere AI** from your Desktop shortcut or Start Menu.
4. When the startup splash screen appears, **click anywhere** or press any key to enter.
5. A **Hardware Configuration Diagnostic** window will display your detected graphics card and memory tier. Click **Confirm Hardware** to open the main workstation.

---

## 📦 How to Download & Manage AI Models

AI models are the local "brains" that power Repetere AI. They must be downloaded before you can generate text, code, or music.

### Downloading Inside the App (Easiest Method):
1. In the top menu bar, click **Settings ➔ Repetere Models**.
2. A full-screen manager will open showing:
   * **Installed Models:** AI files already detected on your machine.
   * **Curated Catalog:** Models tested and verified for your hardware.
3. Click the **[ Download ]** button next to the model you want:
   * **Coding Workspace:** *Qwen 2.5 Coder 7B Instruct* (or *14B*)
   * **General Chat Workspace:** *Meta Llama 3.1 8B Instruct*
   * **Creative Writing Workspace:** *Hermes 3 Llama 3.1 8B*
   * **Audio / Music Studio:** *YuE2 Full Song Engine*
4. A live progress bar will show the download status. When complete, the button changes to a green **Installed** badge.
5. Click **[ Return to Workspace ]** to start using the model.

### Manual Model Placement (Advanced Users):
You can place `.gguf` model files directly into your local installation directory:
```text
%LOCALAPPDATA%\Programs\Repetere AI\models\

🧭 Navigating the Workspaces

Repetere AI features 5 dedicated tabs across the top of the interface:

1. CODING

  • Purpose: Software engineering, syntax questions, scripting, and code review.
  • Features:
  • Automatically formats code inside horizontal-scrolling boxes with syntax styling.
  • Click the Copy button in the top corner of any code snippet to send it straight to your clipboard.
  • Click [ Copy Full Response ] below any message for quick export.

2. CHAT

  • Purpose: Everyday conversation, reasoning, summaries, and answering general questions.
  • Features: Full conversation history and context tracking across back-and-forth messages.

3. CREATIVE WRITING

  • Purpose: Long-form prose, dialogue, scriptwriting, and uncensored world-building.
  • Features: Free-flowing narrative output without cloud moderation filters.

4. AUDIO (MUSIC) Studio

  • Purpose: Generating complete songs with vocals and accompaniment.
  • Prerequisite Files: Make sure both yue2-3b-q4_0.gguf (~2.48 GB) and yue2-vae-f16.gguf (~265 MB) are installed in your models/ directory.
  • How to Generate Music:
  1. Open the AUDIO (MUSIC) tab.
  2. In Musical Style & Instrumentation, enter your genre and instruments (e.g., Alternative Rock, driving drums, distorted guitar, 110 bpm).
  3. In the lyrics area, enter your structural lyrics with bracket tags (e.g., [Verse], [Chorus]).
  4. Adjust the Track Length slider (try 15–30 seconds for initial tests).
  5. Click [ Generate Music Track ] (or [ Generate Full Song (With Vocals) ]).
  6. When the status banner reports completion, your .wav file is saved in output\music\. Click the playback button in the transport panel to listen immediately.

5. IMAGES (Early Preview / Experimental)

  • Status: ⚠️ Work in progress for v0.6.
  • Details: The visual studio layout—including prompt buffers, resolution selectors, steps/CFG sliders, and reference image upload—is accessible for preview. The live diffusion rendering engine is actively being integrated and will be functional in the next update.

⚙️ Hardware Compute Modes: GPU vs. HYBRID

At the top right of the window, you will see a Mode selector:

  • GPU Mode (Active Default): Runs models entirely inside your graphics card's dedicated VRAM for maximum speed and lowest response latency.
  • HYBRID Mode (Interface Preview): Shows the upcoming system for running massive 32B+ parameter models across both system RAM and VRAM. Dynamic runtime offloading will be unlocked in the v0.6 release.

📜 Conversation Transcripts & History

  • All conversations are automatically saved to chat_history.json when clearing a session or exiting the application.
  • Click History ➔ Open History in the top menu bar to view past sessions.
  • Select an archived session from the right-hand list to read the full log on the left.
  • To reset your history safely, select Manage ➔ Clear All History Records and confirm the prompt.

❓ Troubleshooting & FAQs

  • Red "MODEL NOT INSTALLED" banner appears on a tab:
    Click [ Open Repetere Models to Download ] and hit Download for that workspace's assigned model.
  • Audio generation fails or exits instantly:
    Verify that both yue2-3b-q4_0.gguf and yue2-vae-f16.gguf exist inside %LOCALAPPDATA%\Programs\Repetere AI\models\.
  • Clearing a chat:
    Click the red Clear Chat button at the top right of any chat tab. This archives your current conversation safely to disk and resets GPU memory for a fresh session.
  • Application closes immediately on launch:
    Ensure your NVIDIA drivers are version 528 or newer, and verify your GPU supports CUDA 12.
Loading