namkoong-lab/LatentGym_Qwen3-8B_1episode_4Envs_full

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 13, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The namkoong-lab/LatentGym_Qwen3-8B_1episode_4Envs_full 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 adapting to new tasks within four distinct environments: hangman, wordladder, secretary, and number_guessing. This model demonstrates meta-learning capabilities by leveraging latent representations learned during training across these diverse environments.

Loading preview...

Model Overview

namkoong-lab/LatentGym_Qwen3-8B_1episode_4Envs_full is an 8 billion parameter language model built upon the Qwen/Qwen3-8B architecture. It has been fine-tuned using the GRPO (Generalized Reinforcement Learning with Policy Optimization) algorithm, specifically designed for in-context meta-learning across a set of diverse environments.

Key Capabilities & Training

This model's primary strength lies in its ability to meta-learn within specific task environments. It was trained with a focus on adapting to new instances of tasks within:

  • Hangman: Utilizing latents like vowel_count_4 and ending_ABLE.
  • Wordladder: Incorporating hub_word_3letter, hub_word_4letter, and order_outside_in latents.
  • Secretary: Trained with inverse_order and fixed_position_2 latents.
  • Number Guessing: Using set_of_3 and range_100 latents.

The training involved a single episode per trajectory (N=1) and utilized an AdamW optimizer with a constant learning rate schedule and a KL coefficient β of 0.04. The model was trained for 20 epochs with a max generation length of 64 tokens, aiming for cumulative reward (Σᵢ Gᵢ).

Good For

  • Research and experimentation in in-context meta-learning.
  • Developing agents that need to quickly adapt to new tasks within the specified environments.
  • Exploring the effectiveness of GRPO for fine-tuning large language models on meta-learning benchmarks.

Finetunes
1 models