suryatmodulus/fable-traces

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 3, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

fable-traces is a 4 billion parameter instruction-tuned language model developed by suryatmodulus, built upon the Qwen3-4B-Instruct-2507 architecture. Optimized for short, conversational responses, this model is designed to run efficiently on a single mid-range GPU. It leverages bfloat16 precision and uses the ChatML prompt format, inheriting its context length from the base Qwen3 model.

Loading preview...

Overview

fable-traces is a compact, instruction-tuned language model developed by suryatmodulus, based on the Qwen3-4B-Instruct-2507 architecture. With approximately 4 billion parameters, it is specifically optimized for generating concise, conversational replies.

Key Capabilities

  • Efficient Operation: Designed to run comfortably on a single mid-range GPU, making it accessible for various deployment scenarios.
  • Instruction-Tuned: Fine-tuned for following instructions and generating appropriate responses in a conversational context.
  • Precision: Utilizes bfloat16 precision for efficient computation and memory usage.
  • Prompt Format: Employs the ChatML prompt format, ensuring compatibility with standard chat templates.

Good For

  • Applications requiring short, direct conversational interactions.
  • Deployment on hardware with limited GPU resources.
  • Use cases where a balance between model size and conversational capability is crucial.