3l3ktr4/donorsim-qwen3-8b-modeAB-step75
The 3l3ktr4/donorsim-qwen3-8b-modeAB-step75 is an 8 billion parameter Qwen3 model fine-tuned using GRPO on the iterated Donor's Game, specifically Stage 2 with mixed Mode A/B scenarios. This model is specialized for simulating and analyzing complex social interactions and decision-making within group dynamics, particularly in game theory contexts. It is designed for research and development in AI behavior modeling and social simulation, offering insights into strategic interactions.
Loading preview...
Model Overview
The 3l3ktr4/donorsim-qwen3-8b-modeAB-step75 is an 8 billion parameter Qwen3 model that has undergone specialized fine-tuning. It utilizes GRPO (verl 0.7.1, LoRA r16/alpha32 merged into bf16 weights) to optimize its performance.
Key Specialization
This model is specifically fine-tuned on the iterated Donor's Game, Stage 2. This involves complex group games with varying group sizes (K in {2,4,6}) and within-group partner rotation. A notable aspect of its training is the mixed Mode A/B setup, where 50% of scenarios included real groupmate discussion, enhancing its ability to model nuanced social interactions.
Training Details
- Training Method: Fine-tuned with GRPO (verl 0.7.1, LoRA r16/alpha32 merged).
- Training Data: Iterated Donor's Game, Stage 2, with mixed Mode A/B scenarios.
- Lineage: The model's training lineage traces back through
donorsim-qwen3-8b-modeAB-step6, with subsequent steps on increasing GPU resources (4 GPUs for steps 7-11, 8 GPUs for steps 12-75). - Final Step: Represents the 75th and final training step of its run.
Usage
The model's full weights are merged, allowing it to be loaded directly using transformers or vLLM without requiring an adapter. This makes it readily deployable for applications in social simulation and game theory research.
Ideal Use Cases
- Social Simulation: Modeling and predicting behavior in iterated social dilemmas.
- Game Theory Research: Investigating strategic decision-making in multi-agent environments.
- AI Behavior Analysis: Studying the impact of communication and group dynamics on AI agents.