SelectiveDOPD/JustRL-Qwen3-4b-DirectOPD

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 11, 2026Architecture:Transformer Featherless Exclusive Cold

JustRL-Qwen3-4b-DirectOPD is a 4 billion parameter language model based on the Qwen3 architecture, developed by SelectiveDOPD as part of the BiDirect-OPD experiments. This model is specifically designed for applications requiring direct output policy distillation (DirectOPD) within reinforcement learning contexts. It offers a 32768 token context length, making it suitable for tasks that benefit from processing extensive input sequences.

Loading preview...

Model Overview

JustRL-Qwen3-4b-DirectOPD is a 4 billion parameter language model developed by SelectiveDOPD. It is derived from the Qwen3 architecture and was created as part of the BiDirect-OPD experimental series, focusing on direct output policy distillation within reinforcement learning.

Key Characteristics

  • Architecture: Based on the Qwen3 model family.
  • Parameter Count: 4 billion parameters, offering a balance between performance and computational efficiency.
  • Context Length: Supports a substantial context window of 32768 tokens, enabling the processing of longer inputs and more complex interactions.
  • Development Context: Originated from the justrl_Direct_OPD_qwen3_4b project within the BiDirect-OPD experiments, indicating a specialized focus on reinforcement learning applications with direct policy output.

Available Checkpoints

The main branch hosts the global_step_300 checkpoint. Additionally, several earlier checkpoints are available as separate branches, allowing users to experiment with different stages of the model's training progression:

  • global_step_20
  • global_step_40
  • global_step_60
  • global_step_80
  • global_step_100
  • global_step_120
  • global_step_140
  • global_step_160
  • global_step_180
  • global_step_200
  • global_step_220
  • global_step_240
  • global_step_260
  • global_step_280

Intended Use Cases

This model is particularly suited for research and development in reinforcement learning, especially where direct output policy distillation is a core requirement. Its large context window also makes it valuable for tasks needing to process extensive sequential data.

Finetunes
1 models