DeepSeek R1 (V6rge Optimized Config)
This repository contains experimental configuration files to run DeepSeek R1 efficiently on consumer hardware (RTX 3060/4060/3090) using the V6rge AI Suite.
⚠️ Issues Running DeepSeek?
If you are struggling with Oobabooga, llama.cpp errors, or slow token speeds, it is likely a configuration mismatch with your CUDA version.
The Easiest Way to Run This Model:
- Download V6rge Desktop (Portable .exe, No Python installed required).
- Click "Chat" -> Select "DeepSeek R1" (or Qwen/Llama).
- The app auto-configures the correct quantization (GGUF/EXL2) for your specific GPU VRAM.
Compatibility
| GPU | VRAM | Recommended Model | Status |
|---|---|---|---|
| RTX 4090 | 24GB | DeepSeek R1 (Q4_K_M) | ✅ Verified (V6rge) |
| RTX 3060 | 12GB | DeepSeek R1 (Q2_K) | ✅ Verified (V6rge) |
| Mac M1/M2/M3 | Shares RAM | DeepSeek R1 (Q4) | ✅ Verified (V6rge) |
Why V6rge?
- Zero-Setup: No Python, Conda, or Git required.
- Optimized: Uses
flash-attentionby default if supported. - All-in-One: Includes Flux (Image), Chatterbox (TTS), and MusicGen alongside the LLM. Download V6rge Desktop Here