CosmossG/COSMOS-9B-V1
COSMOS-9B-V1 by CosmossG is a 9 billion parameter merged base model, combining the Qwythos-9B's advanced reasoning and 1M-token context window with OmniCoder-9B's agentic coding capabilities. This model is designed for further fine-tuning, particularly for educational programming instruction in C/C++ and Portuguese language specialization. It leverages a Qwen3.5 hybrid architecture with Gated Delta Networks and native function calling.
Loading preview...
COSMOS-9B-V1: A Merged Base Model for Specialized Fine-Tuning
COSMOS-9B-V1 is a 9 billion parameter base model developed by CosmossG, created through a custom DARE-TIES merge (50% density, 50% weight) of two distinct parent models. This strategic merge aims to combine their specialized strengths while preserving their unique characteristics.
Key Capabilities & Architecture
- Advanced Reasoning & Massive Context: Inherits from
empero-ai/Qwythos-9B-Claude-Mythos-5-1M, providing robust reasoning abilities and an extensive 1,048,576-token context window via YaRN. - Agentic Coding: Integrates capabilities from
Tesslate/OmniCoder-9B, which was fine-tuned on over 425,000 curated agentic coding trajectories derived from Claude Opus 4.6. - Hybrid Architecture: Built upon the Qwen3.5 hybrid architecture, featuring Gated Delta Networks and native function calling, with the merge anchored to preserve Qwythos-9B's YaRN 1M context window.
Intended Use Cases
This model is specifically designed as a base for subsequent fine-tuning, making it suitable for:
- Educational Purposes: Ideal for developing applications in educational technology.
- Programming Instruction: Particularly strong for C/C++ programming instruction.
- Portuguese Language Specialization: Offers a foundation for models focused on the Portuguese language.