icecubetr/GRMR-V3-G4B
GRMR-V3-G4B by qingy2024 is a 4.3 billion parameter Gemma 3 model, fine-tuned from unsloth/gemma-3-4b-pt, specifically optimized for English grammar correction tasks. It excels at fixing grammatical errors, punctuation, and spelling, utilizing a specialized chat template for structured input and output. The model has a context length of 32768 tokens and is designed to improve text quality by addressing various language issues.
Loading preview...
GRMR-V3-G4B: Grammar Correction Model
GRMR-V3-G4B is a 4.3 billion parameter Gemma 3 model developed by qingy2024, fine-tuned from unsloth/gemma-3-4b-pt. Its primary purpose is grammar correction, making it a specialized tool for improving text quality.
Key Capabilities
- Grammar Error Correction: Identifies and fixes grammatical mistakes.
- Punctuation Correction: Addresses incorrect or missing punctuation.
- Spelling Correction: Rectifies spelling errors.
- Sentence Structure Improvement: Enhances clarity and structure of sentences.
Training and Features
The model was fine-tuned using full parameter fine-tuning on the qingy2024/grmr-v4-60k dataset, which contains 60,000 examples of original and corrected text. It utilizes a custom chat template with <|text_start|>/<|text_end|> for user inputs and <|corrected_start|>/<|corrected_end|> for model outputs, ensuring clear distinction between original and corrected content. Optimal performance is achieved with specific sampler settings, including a temperature of 0.7.
Limitations
While effective for general grammar correction, the model may face challenges with highly technical or domain-specific content, struggle with context-dependent rules, and its performance can vary for non-standard English or text with numerous errors. It focuses on correctness rather than stylistic enhancements.