yimn-Aghosh/zegrate-ai

TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 5, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Zegrate AI by yimn-Aghosh is a 14.8 billion parameter conversational AI model, fine-tuned from Qwen2.5-14B-Instruct with a 32k token context length. It is specifically optimized for humor-first conversational interactions, developed using 4-bit QLoRA + DoRA on curated multi-shard data. This model is designed for open-ended dialogue where a humorous and engaging personality is desired.

Loading preview...

Zegrate AI: Humor-First Conversational Model

Zegrate AI, developed by yimn-Aghosh, is a 14.8 billion parameter open conversational AI model fine-tuned from Qwen2.5-14B-Instruct. It features a distinct humor-first personality, making it suitable for engaging and entertaining dialogue. The model leverages a 32k token context length, allowing for extended and coherent conversations.

Key Capabilities

  • Humor-First Personality: Deliberately trained to exhibit a humorous tone in interactions.
  • Conversational AI: Optimized for open-ended dialogue and assistant-like functions.
  • Efficient Fine-tuning: Utilizes 4-bit QLoRA + DoRA for efficient training on a single GPU.
  • Extended Context: Supports a 32k token context window for longer, more complex exchanges.

Good for

  • Applications requiring an AI with a distinct, humorous personality.
  • Interactive chatbots and virtual assistants where engagement and entertainment are key.
  • Developers looking for a Qwen2.5-14B-Instruct based model with specialized conversational tuning.

Note: The model is currently undergoing Phase 2 training, meaning weights are in-progress and may exhibit "rough edges" until completion. It is primarily English-focused and, as a 14B class model, functions as a strong assistant rather than a frontier-scale model.

Finetunes
1 models