base_model: aifeifei798/DarkIdol-Llama-3.1-8B-Instruct-1.3-Uncensored
library_name: transformers
tags:
- exl2
- 6.0bpw
- roleplay
- uncensored
- llama-3
- text-generation
license: llama3.1
language: - en
DarkIdol 1.3 - EXL2 (6.0bpw)
This is a high-precision EXL2 quantization of the DarkIdol 1.3 Uncensored model.
⚙️ Hardware Requirements
- Quantized on: NVIDIA RTX 5070
- VRAM Required: ~6.5 GB (Model) + Context.
- Recommendation: Perfect for 12GB cards (RTX 3060/4070/5070) allowing for 8k-16k context.
🚀 Usage (ExLlamaV2 / TabbyAPI / Text-Gen-WebUI)
This model is compatible with the ExLlamaV2 loader.
🧪 Calibration & Performance
- Calibration Dataset:
wikitext-test(Standard) - Quantization Accuracy: High precision (6.0bpw) retains >99% of the base model's reasoning capabilities while reducing VRAM usage by 40%.
📝 Prompt Template (Llama 3)
This model uses the standard Llama 3 Instruct format. Set your interface to use this template:
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>
{user_prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>