LorenaYannnnn/20260217-Qwen3-0.6B_grpo_sycophancy_warmup_4x_baseline_320000_episodes_seed_42
The LorenaYannnnn/20260217-Qwen3-0.6B_grpo_sycophancy_warmup_4x_baseline_320000_episodes_seed_42 is a 0.8 billion parameter language model based on the Qwen3 architecture. This model is specifically fine-tuned for sycophancy detection and mitigation, having undergone a specialized Group Policy (GRPO) warm-up training regimen over 320,000 episodes. Its primary strength lies in identifying and reducing sycophantic responses, making it suitable for applications requiring objective and unbiased language generation.
Loading preview...
Model Overview
This model, LorenaYannnnn/20260217-Qwen3-0.6B_grpo_sycophancy_warmup_4x_baseline_320000_episodes_seed_42, is a 0.8 billion parameter language model built upon the Qwen3 architecture. It has been subjected to a specialized training process involving a Group Policy (GRPO) warm-up, specifically targeting sycophancy detection and reduction. The training spanned 320,000 episodes, indicating a focused effort to instill robust behavior against generating overly flattering or biased responses.
Key Capabilities
- Sycophancy Mitigation: The model is designed to minimize sycophantic outputs, promoting more objective and neutral language generation.
- Qwen3 Architecture: Leverages the foundational capabilities of the Qwen3 model family.
- Compact Size: With 0.8 billion parameters, it offers a relatively efficient footprint for deployment while retaining specialized capabilities.
- Extended Context Length: Supports a context length of 32,768 tokens, allowing for processing longer inputs.
Good For
- Applications requiring unbiased text generation.
- Scenarios where detecting and avoiding sycophantic language is critical.
- Use cases where a smaller, specialized model is preferred over larger, general-purpose alternatives for specific behavioral control.