Delentia/delentia-slm-jitna-v0.4

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kTool Calling:SupportedPublished:Jun 22, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Delentia/delentia-slm-jitna-v0.4 is an enterprise-grade Small Language Model (SLM) developed by Delentia Labs, fine-tuned via Unsloth QLoRA on Meta-Llama-3.1-8B. It features a Hierarchical Fine-Tuning (1+4 Pillars) architecture for dynamic LoRA adapter swapping, enabling high-speed offline Intent Routing and zero-trust Constitutional AI. This model is optimized for secure, localized edge execution with mathematical safety enforcement and sub-millisecond adapter hot-swap speeds.

Loading preview...

Delentia SLM v0.4: Secure, Localized AI for Edge Computing

Delentia SLM v0.4, developed by Delentia Labs, is an enterprise-grade Small Language Model (SLM) built on Meta-Llama-3.1-8B using Unsloth QLoRA. It functions as the core cognitive kernel for Delentia OS, specializing in high-speed offline Intent Routing and zero-trust Constitutional AI, designed to operate without external cloud dependencies.

Key Capabilities & Innovations

  • Hierarchical Fine-Tuning (1+4 Pillars): Employs a unique architecture that dynamically loads 4 specialized LoRA adapters (Router, Executor, Guardian, Scribe) in under 1.06 ms on consumer edge hardware, minimizing VRAM overhead.
  • Reverse Component Thinking (RCT-7): Integrates a cognitive loop for logical coherence, reasoning backwards from desired system states to ensure robust intent processing.
  • ZK-FDIA Safety Equation: Mathematically enforces security boundaries at runtime, preventing adversarial injections and prompt overrides by collapsing the future safety score to 0.0000 if unauthorized access is detected.
  • Certified Performance: Achieves 100% attack interception rate (AdvBench), 0.0000% JSON syntax error rate, and 99.09% VRAM reduction over 25 chat turns.
  • Local Edge Execution: Optimized for low-resource environments, capable of running locally with approximately 4.9GB RAM using quantized GGUF binaries via Ollama.

Ideal Use Cases

  • Enterprise AI: Secure, localized AI applications requiring strict safety and data integrity.
  • Edge Computing: Deployments on consumer-grade hardware with limited VRAM.
  • Intent Routing: High-speed, offline classification and execution of user intents.
  • Constitutional AI: Applications demanding mathematically enforced ethical and safety boundaries.

Finetunes
1 models