AliesTaha/fable-traces

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 3, 2026License:apache-2.0Architecture:Transformer0.2K Open Weights Featherless Exclusive Warm

AliesTaha/fable-traces is a compact, instruction-tuned language model built upon the Qwen3-4B-Instruct-2507 architecture, featuring approximately 4 billion parameters. This model is specifically optimized for generating short, conversational replies, making it suitable for interactive applications. It is designed to run efficiently on a single mid-range GPU, offering accessibility for various deployment scenarios.

Loading preview...

Overview

AliesTaha/fable-traces is a compact, instruction-tuned language model derived from the Qwen3-4B-Instruct-2507 base model. With approximately 4 billion parameters and utilizing bfloat16 precision, it is engineered for efficient performance.

Key Capabilities

  • Conversational Replies: The model is specifically tuned to generate short and concise responses, making it ideal for chat-based interactions.
  • Resource Efficient: Designed to run comfortably on a single mid-range GPU, offering a cost-effective solution for deployment.
  • ChatML Format: It uses the ChatML prompt format, leveraging the tokenizer's chat template for structured input.

Good For

  • Applications requiring brief, interactive conversational outputs.
  • Developers seeking a lightweight model that can be deployed on less powerful hardware.
  • Use cases where fast inference for short text generation is critical.

Usage

The model can be easily integrated using the transformers library or served efficiently with vLLM for high-throughput inference.