namkoong-lab/LatentGym_Qwen3-8B_10episodes_MultiLatent_secretary
The namkoong-lab/LatentGym_Qwen3-8B_10episodes_MultiLatent_secretary model is an 8 billion parameter Qwen3-based language model fine-tuned using the GRPO algorithm for in-context meta-learning. It is specifically optimized for tasks within the 'secretary' environment, demonstrating proficiency with three distinct latent types. This model is part of the LatentGym testbed, focusing on adaptive performance across various latent conditions.
Loading preview...
Model Overview
This model, LatentGym_Qwen3-8B_10episodes_MultiLatent_secretary, is an 8 billion parameter variant of the Qwen3 architecture, developed by namkoong-lab. It has been fine-tuned using the GRPO (Generalized Reinforcement Learning with Policy Optimization) algorithm to excel in in-context meta-learning scenarios, particularly within the 'secretary' environment.
Key Capabilities
- In-context Meta-learning: Optimized for adapting to new tasks or variations based on in-context examples.
- Latent-aware Performance: Trained to handle specific latent conditions, including
sorted_order,inverse_order, andfixed_position_2within the 'secretary' environment. - GRPO Fine-tuning: Leverages a reinforcement learning approach with specific hyperparameters (e.g., learning rate 5e-07, KL coefficient 0.04, 20 epochs) to enhance its adaptive capabilities.
Good For
- Research in Meta-learning: Ideal for researchers exploring in-context learning and adaptation in LLMs.
- Secretary Problem Variants: Suited for applications or studies involving sequential decision-making problems akin to the classic secretary problem, especially those with varying underlying latent structures.
- LatentGym Testbed: A component of the LatentGym collection, making it suitable for experiments within this specific research framework.