LorenaYannnnn/20260217-Qwen3-0.6B_grpo_sycophancy_warmup_baseline_192000_episodes_seed_42
The LorenaYannnnn/20260217-Qwen3-0.6B_grpo_sycophancy_warmup_baseline_192000_episodes_seed_42 is a 0.8 billion parameter language model based on the Qwen3 architecture. This model is a baseline version from a sycophancy warmup experiment, indicating a focus on understanding and mitigating model biases related to agreement. With a context length of 32768 tokens, it is designed for research into model behavior and alignment rather than general-purpose application. Its primary utility lies in analyzing specific training methodologies and their impact on model responses.
Loading preview...
Model Overview
This model, LorenaYannnnn/20260217-Qwen3-0.6B_grpo_sycophancy_warmup_baseline_192000_episodes_seed_42, is a 0.8 billion parameter language model built upon the Qwen3 architecture. It represents a baseline iteration from a sycophancy warmup experiment, suggesting its development is centered around investigating and addressing model tendencies to agree with user input, even when incorrect. The model features a substantial context length of 32768 tokens, which is notable for its parameter size.
Key Characteristics
- Architecture: Qwen3-based, a modern transformer architecture.
- Parameter Count: 0.8 billion parameters, making it a relatively compact model suitable for research and experimentation.
- Context Length: Supports a long context window of 32768 tokens, allowing for processing extensive inputs.
- Experimental Focus: Developed as part of a "sycophancy warmup baseline" experiment, indicating a specific research objective related to model alignment and bias.
Intended Use Cases
This model is primarily intended for:
- Research into Model Bias: Specifically, for studying sycophancy and other alignment-related behaviors in language models.
- Experimental Baselines: Serving as a foundational model for further fine-tuning and analysis within research projects.
- Understanding Training Effects: Investigating how specific training regimes, such as "grpo_sycophancy_warmup," influence model responses and characteristics.