GMatherne/qwen3-8b-human-sft

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 13, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

GMatherne/qwen3-8b-human-sft is a QLoRA fine-tune of Qwen3-8B, specifically trained to generate human-sounding, conversational prose for educational and tutoring questions. This 8 billion parameter model focuses on producing text that AI-text detectors identify as human-written, rather than AI-generated. Its primary strength lies in its unique 'voice' for educational content, though this comes at a cost to factual accuracy compared to the base model. It is intended for research into data-driven behavior control and generating educational text where precise factual authority is not critical.

Loading preview...

Model Overview

GMatherne/qwen3-8b-human-sft is a QLoRA fine-tune of the Qwen3-8B model, developed by GMatherne. Its core purpose is to generate natural, human-sounding prose for educational and tutoring inquiries, specifically aiming to bypass AI-text detection. This model demonstrates "behavior from data" by adopting a conversational, direct, and first-person voice, similar to a knowledgeable human answering questions in a forum.

Key Characteristics

  • Human-Sounding Voice: Optimized to produce text that AI detectors (like Pangram) classify as human-written, avoiding typical AI-generated patterns.
  • Educational Focus: Designed to explain concepts, assist with essays, and correct mistakes in a conversational style.
  • Training Data: Fine-tuned on approximately 1,800 curated human answers from platforms like StackExchange and Reddit, heavily cleaned to remove forum scaffolding and non-English text.
  • Limitations: This fine-tuning for human voice lowers factual accuracy and increases fabrication rates compared to the untuned base model. It is not suitable for authoritative facts, mathematics, code generation, or tasks requiring strict format adherence.

Intended Use Cases

  • Research: Ideal for studying data-driven behavior control in LLMs.
  • Educational Content Generation: Useful for creating educational prose where a human-like voice is prioritized over absolute factual precision.

Usage Notes

  • Uses the ChatML template with a specific system prompt to guide its conversational style.
  • Recommended sampling parameters include temperature=0.7, top_p=0.8, top_k=20, and repetition_penalty=1.15.

Finetunes
1 models