cyberandy/gemma3-1b-feliSEO-2run

Hugging Face
TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kPublished:Apr 23, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

The cyberandy/gemma3-1b-feliSEO-2run model is an experimental 1 billion parameter fine-tune of unsloth/gemma-3-1b-it, developed by cyberandy (WordLift Lab) using Group Policy Optimization (GRPO). It is specifically designed to perform basic SEO-related reasoning tasks and structure outputs with and XML-like tags, guided by the SEOntology project. This model, with a 32768 token context length, is an early iteration for research into structured SEO reasoning and is not suitable for production use due to its small training dataset and potential for hallucinations.

Loading preview...

Overview

This model, cyberandy/gemma3-1b-feliSEO-2run, is an experimental fine-tune of the unsloth/gemma-3-1b-it base model, developed by cyberandy within the WordLift Lab. It utilizes Group Policy Optimization (GRPO) with trl and Unsloth optimizations to teach basic SEO-related reasoning. The model is designed to structure its output using specific XML-like tags: <reasoning> for step-by-step thinking and <answer> for the final response, drawing concepts from the SEOntology project.

Key Capabilities

  • Structured SEO Reasoning: Attempts to provide logical steps and answers for SEO tasks within <reasoning> and <answer> tags.
  • Format Adherence: Shows good adherence to the requested output format.
  • Experimental GRPO Application: Demonstrates the application of GRPO for fine-tuning on a small, custom dataset.

Good for

  • Research and Experimentation: Ideal for researchers interested in applying GRPO to teach structured reasoning, particularly in the SEO domain.
  • Understanding Fine-tuning Processes: Useful for studying the fine-tuning methodology with trl and Unsloth optimizations.
  • Early-stage Concept Prototyping: Can be used to explore how LLMs can be guided by ontologies like SEOntology for specific task outputs.

Note: This is an early iteration trained on only 100 synthetic examples. It is not suitable for production use due to limited generalization, potential for hallucinations, and unverified factual accuracy.

Finetunes
1 models