NGARiAI/ngari-ft-distilled

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 3, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

NGARiAI/ngari-ft-distilled is a 1.5 billion parameter QA model developed by NGARiAI, distilled from a larger teacher model. Built on Qwen2.5-1.5B-Instruct, it is fine-tuned on NGARi's sovereign-agent domain data. This model is specifically optimized for real-time content safety judging and QA tasks on resource-constrained aarch64 edge hardware with 8GB RAM, operating with zero cloud dependency. Its primary strength lies in delivering advanced QA capabilities in an air-gapped, on-device environment.

Loading preview...

NGARi FT Distilled: Edge-Optimized QA and Safety

NGARiAI/ngari-ft-distilled is a compact 1.5 billion parameter model designed for efficient, on-device execution. Developed by NGARiAI, this model serves as a critical component within the NGARi Sovereign Business Operating System (NS-BOS), functioning as a real-time content-safety judge and QA model for agent responses.

Key Capabilities and Design

  • Distilled Performance: This model is a "student" in NGARi's teacher-student architecture, where a larger 27B-class teacher model generates reasoning traces and synthetic data. The ngari-ft-distilled model is then fine-tuned on these outputs, compressing advanced capabilities into a small footprint.
  • Edge Hardware Optimization: Specifically engineered to run on aarch64 edge hardware like the NVIDIA Jetson AGX Orin with just 8GB RAM, ensuring zero cloud dependency and air-gapped operation.
  • Core Function: Its primary role is QA and content safety judging, acting as a guardrail for agent interactions.
  • Base Model: Built upon Qwen/Qwen2.5-1.5B-Instruct (Apache 2.0 license).
  • Secondary Capability: While its main focus is QA, it also exhibits some tool detection capabilities (tool detection rate of 0.6), though for high-accuracy tool calling, the dedicated NGARiAI/ngari-tool model is recommended.

Ideal Use Cases

  • On-device AI: Deploying AI models directly on user-owned hardware without cloud reliance.
  • Content Moderation: Real-time judging of generated content for safety and compliance.
  • Sovereign AI Applications: Building systems that require complete data control and air-gapped operation.
  • Resource-Constrained Environments: When high-performance QA is needed on devices with limited memory and processing power.

Finetunes
1 models