Overview

The farffadet/syllogym-judge-qwen3-4b-grpo-v2 is a 4 billion parameter language model based on the Qwen3 architecture. Developed by farffadet, this model was fine-tuned from the unsloth/Qwen3-4B-unsloth-bnb-4bit base model.

Key Characteristics

• Base Model: Qwen3-4B
• Parameter Count: 4 billion parameters
• Context Length: 32768 tokens
• Training Method: Utilizes Unsloth and Huggingface's TRL library for accelerated fine-tuning, resulting in 2x faster training compared to standard methods.
• License: Apache-2.0

Intended Use

This model is suitable for a variety of general language understanding and generation tasks, benefiting from its Qwen3 foundation and efficient fine-tuning process. Its 32K context window allows for processing longer inputs and generating more coherent, extended outputs.