NovatasticRoScript/Atomight-V2.1-0.5B-Inference

TEXT GENERATIONConcurrency Cost:1Model Size:0.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:May 27, 2026License:mitArchitecture:Transformer0.0K Open Weights Cold

Atomight-V2.1-0.5B-Inference by NovatasticRoScript is an ultra-compact, 494 million parameter causal language model built on a Qwen-derived foundation. It is specifically optimized for reasoning-oriented tasks and structured text outputs, utilizing GRPO reinforcement tuning for high-signal reasoning vectors. This model is designed for highly efficient edge-device inference, lightweight coding assistance, and rapid deployment under severe compute constraints, loading into approximately 1GB VRAM at FP16.

Loading preview...

Atomight-V2.1-0.5B-Inference: Ultra-Compact Reasoning Model

Atomight-V2.1-0.5B-Inference, developed by NovatasticRoScript under the Atomight Ecosystem, is a 494 million parameter causal language model. It is built on a Qwen-derived foundation and refined using GRPO (Group Relative Policy Optimization) reinforcement tuning, which focuses on high-signal reasoning vectors rather than brute-force dataset scale.

Key Capabilities & Features

  • Ultra-Compact Footprint: Approximately 494M parameters, loading into ~1GB VRAM at FP16, making it suitable for resource-constrained environments.
  • Reasoning-Oriented: Specifically designed for reasoning tasks, structured text outputs, and lightweight coding assistance.
  • Edge-Optimized: Engineered for low-overhead mobile, local, and browser-based inference loops, including Google Colab and Kaggle native workflows.
  • Competitive Benchmarks: Outperforms Meta's larger Llama-3.2-1B-Instruct on localized logic-retrieval metrics, achieving 59.34% on ARC-Easy and 33.79% on ARC-Challenge. It also demonstrates higher raw mathematical accuracy on GSM8K (Flexible Extraction) compared to Qwen2.5-0.5B-Instruct and Llama-3.2-1B-Instruct.

When to Use This Model

This model is ideal for developers requiring a highly efficient and compact language model for:

  • Edge-device deployment: Mobile, local, or browser-based applications where compute resources are limited.
  • Reasoning tasks: Scenarios demanding strong logical inference and problem-solving capabilities, particularly in scientific fact retrieval and commonsense reasoning.
  • Structured output generation: Use cases where precise, formatted text outputs are crucial, even if the model might bypass rigid formatting for dense thinking traces in mathematical problems.
  • Lightweight coding assistance: Providing support for coding tasks without the overhead of larger models.