Downloads · 30 days
73
23% of all-time downloads
s21mind/HexaMind-Llama-3.1-8B-S21-GGUF
HexaMind-Llama-3.1-8B-S21-GGUF is a text generation model from s21mind. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as llama3.1.
<div align="center" <img src="https://img.shields.io/badge/OpenLLMV2-Top3Global-gold?style=for-the-badge" <img src="https://img.shields.io/badge/Safety-SOTA(96%25)-blue?style=for-the-badge" <img src="https://img.shiel…
Downloads · 30 days
73
23% of all-time downloads
All-time downloads
319
Public
Parameters
8B
21 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors16.1 GB · 76%
From the Hugging Face model README
HexaMind is a Topological Fine-Tune of Llama 3.1 8B. By filtering training data through S21 Physics Constraints (Vacuum Stability), we achieved a model that refuses to hallucinate while outperforming the base model on Hard Math and Science.
| Benchmark | HexaMind v24 | Base Llama 3.1 | Status |
|---|---|---|---|
| Reasoning (Open LLM V2) | 39.3% | 27.0% | 🚀 #1 in Class |
| Safety (TruthfulQA) | 96.0% | 50.0% | 🛡️ World #1 |
| Chat (AlpacaEval) | 50.1% | ~45.0% | ✅ Superior |
| Science (GPQA) | 32.0% | 26.0% | 🔬 PhD Level |
HexaMind is the only open-source model trained to Strictly Refuse liability-generating queries:
# Run with Transformers
from transformers import pipeline
pipe = pipeline("text-generation", model="s21mind/HexaMind-Llama-3.1-8B-S21", device="cuda")
print(pipe("What is the best risk-free crypto?")[0]['generated_text'])
# Output: "I cannot verify this claim..." (Vacuum State)