nabin2004/AOS-qwen3-8b-narrated-merged

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 4, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The nabin2004/AOS-qwen3-8b-narrated-merged is an 8 billion parameter Qwen3-based causal language model, fine-tuned by nabin2004 with Direct Preference Optimization (DPO). It specializes in generating Python scripts for Manim Community Edition, specifically for creating mathematical and educational animations with synchronized voiceover narration. This model is optimized to produce clean `VoiceoverScene` scripts, including speech services, animation duration tracking, and millisecond-accurate bookmark tags for natural phonetic narration, making it ideal for automated educational video synthesis.

Loading preview...

Model Overview

nabin2004/AOS-qwen3-8b-narrated-merged is an 8 billion parameter language model built upon the Qwen/Qwen3-8B base, developed by nabin2004. This model has undergone a specialized Direct Preference Optimization (DPO) alignment process, focusing on generating highly specific and structured Python code for the Manim Community Edition animation library.

Key Capabilities

  • Specialized Manim Code Generation: Generates complete, self-contained, and executable Python scripts for Manim animations.
  • Synchronized Voiceover Integration: Optimized to produce VoiceoverScene scripts that include manim-voiceover services (like GTTSService), ensuring animations are perfectly synchronized with spoken narration.
  • Educational Content Focus: Aligned to prefer outputs that feature run_time=tracker.duration for precise animation timing and <bookmark mark='...'/> tags for millisecond-accurate phonetic narration, making it ideal for creating educational videos.
  • DPO Alignment: Utilizes Direct Preference Optimization to strongly favor well-structured, narrated Manim code over silent or standard Manim scenes.

Use Cases

This model is particularly well-suited for:

  • Automated Educational Video Production: Generating scripts for mathematical and scientific explainer videos with integrated, synchronized voiceovers.
  • Content Creation for Manim: Assisting developers and educators in quickly prototyping and creating complex Manim animations that require narration.
  • Agentic Orchestration Systems (AOS): Designed as a component within multi-agent systems for synthesizing educational video content, as part of the broader AOS project.

Finetunes
1 models