Hades-8B

Hades 8B - Open-Weights Sovereign SLM Hades 8B is a high-performance, open-weights language model developed by Project Mosaic (Ruzo LLM) to empower Indian startups, developers, and AI researchers with a fast, privacy-focused, enterprise-grade Small Language Model (SLM). Trained on 100,000+ high-quality instruction, reasoning, and tool-calling dialogues using Unsloth on NVIDIA A100 supercomputers. ## Key Specifications • Base Model: Llama-3.1-8B-Instruct • Native Context Window: 128,000 Tokens (128k) • Optimized Sequence Length: 8,192 Tokens • Format: 4-bit Quantized GGUF (Q4_K_M) — fully compatible with Ollama, llama.cpp, and vLLM • Quantized Model Size: ~4.92 GB (Runs smoothly on consumer GPUs & Apple Silicon Macs) ## Designed for Open-Source Developers 1. Advanced RAG Capabilities: Extract precision insights from 25+ page PDFs, legal files, and technical docs with zero hallucination. 2. 100% Valid JSON & Tool Calling: Guarantees strict JSON schema outputs for building autonomous agentic workflows. 3. Air-Gapped & Local First: Deploy 100% offline on local hardware or scale 24/7 on cloud GPUs. ## Community Links • Hugging Face Repository: https://huggingface.co/ProjectMosiacAI/Hades-2.0-Night3-8k • Developed by: Project Mosaic (Ruzo LLM)