SelectiveDOPD/JustRL-Qwen3-1p7b-Selective-Top15pct

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 11, 2026Architecture:Transformer Featherless Exclusive Cold

JustRL-Qwen3-1p7b-Selective-Top15pct is a 2 billion parameter causal language model developed by SelectiveDOPD, based on the Qwen3 architecture. This model was uploaded from the BiDirect-OPD experiments, specifically from the `justrl_qwen3_1p7b_js_ladder_85_100_kl` run. It is characterized by its origin in reinforcement learning experiments, suggesting optimization for specific task performance through iterative training.

Loading preview...

JustRL-Qwen3-1p7b-Selective-Top15pct Overview

This model, developed by SelectiveDOPD, is a 2 billion parameter causal language model built upon the Qwen3 architecture. It originates from the BiDirect-OPD experiments, specifically from the justrl_qwen3_1p7b_js_ladder_85_100_kl run, indicating its development through reinforcement learning methodologies.

Key Characteristics

  • Architecture: Based on the Qwen3 family of models.
  • Parameter Count: Features approximately 2 billion parameters, making it a compact yet capable model.
  • Development Origin: Derived from the BiDirect-OPD experiments, suggesting a focus on performance optimization through reinforcement learning.
  • Checkpoints: Multiple global_step checkpoints are available, ranging from global_step_20 to global_step_300, allowing for exploration of different training stages.

Good for

  • Research in Reinforcement Learning: Ideal for researchers studying the effects of reinforcement learning on language model performance, particularly within the BiDirect-OPD framework.
  • Experimentation with Checkpoints: Users can experiment with various training stages to understand model evolution and performance at different global_step intervals.
  • Applications requiring a compact Qwen3-based model: Suitable for tasks where a smaller, Qwen3-based model with specific RL-driven optimizations is beneficial.

Finetunes
1 models