LRieser/steam-ai-mention-qwen3-8b-merged

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:May 5, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

LRieser/steam-ai-mention-qwen3-8b-merged is an 8 billion parameter Qwen3-based model developed by LRieser, fine-tuned to classify Steam game reviews. It identifies whether a review discusses generative AI used in game content production (art, voice, music, writing, code, translation) and outputs a structured JSON response. This multilingual model achieves a Macro-F1 score of 0.953 on generative AI mention detection in Steam reviews, making it suitable for large-scale content analysis.

Loading preview...

Steam AI-Mention Classifier (Qwen3-8B)

This model, developed by LRieser, is a fine-tuned version of Qwen/Qwen3-8B specifically designed to analyze Steam game reviews. Its primary function is to detect mentions of generative AI being used in the production of game content, such as art, voice, music, writing, code, or translation. The model outputs a structured JSON object indicating whether AI is mentioned, a confidence level, and a reasoning phrase.

Key Capabilities

  • Generative AI Detection: Accurately identifies explicit terms, tool names, developer disclosures, and speculative claims related to AI content generation in game reviews.
  • Multilingual Support: Trained on reviews in 20 languages, including English, Simplified Chinese, Russian, Italian, and German.
  • Structured Output: Provides a JSON output with mentions_genai (boolean), confidence (low, medium, high), and reasoning (decisive phrase).
  • High Performance: Achieves a Macro-F1 score of 0.953 on a held-out test set for keyword-matched reviews, with 0 false positives on a random sample.
  • Efficient Inference: Optimized for corpus-scale inference, capable of processing approximately 11 reviews per second on an RTX 4090 using vLLM.

Training Details

The model was fine-tuned using QLoRA with DoRA on 19,152 Steam reviews, labeled by DeepSeek V4-Pro. The training data included both keyword-matched reviews and a uniform random sample to ensure robust detection and minimize false positives. The full bf16 weights are merged, eliminating the need for a separate base model download.

Use Cases

This model is ideal for developers, researchers, or platforms needing to:

  • Automatically identify discussions about generative AI in user-generated content.
  • Monitor public sentiment and discourse around AI in game development.
  • Analyze trends in AI adoption or perception within the gaming community.

Finetunes
1 models