Stage-org/4b-diversity-LH-300-nyshot-27b-z-epoch3

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 20, 2026Architecture:Transformer Featherless Exclusive Cold

Stage-org/4b-diversity-LH-300-nyshot-27b-z-epoch3 is a 4.5 billion parameter language model based on the Qwen/Qwen3.5-4B architecture, fine-tuned using a reinforcement learning (RL) method. This model was trained for 3 epochs on a diverse dataset, focusing on enhancing its capabilities through an advanced RL setup. It features a substantial context length of 32768 tokens, making it suitable for tasks requiring extensive contextual understanding and generation.

Loading preview...

Model Overview

Stage-org/4b-diversity-LH-300-nyshot-27b-z-epoch3 is a 4.5 billion parameter language model derived from the Qwen/Qwen3.5-4B base architecture. It has undergone reinforcement learning (RL) fine-tuning over 3 epochs, utilizing a specific dataset named Stage-org/4b-diversity-LH-300-nyshot-27b-z.

Key Training Details

  • Base Model: Qwen/Qwen3.5-4B
  • Training Method: Reinforcement Learning (RL)
  • Epochs: 3
  • Batch Size: 128
  • Sequence Length: 300,000 (during training, though the model's context length is 32768 tokens)
  • Optimization: AdamW optimizer with a learning rate of 1e-06 and Flash Attention 2 for efficiency.
  • Generation Parameters: Features a temperature of 0.9 and a maximum token generation of 4096, with enable_thinking set to true, suggesting enhanced reasoning capabilities during generation.
  • Judge Model: Utilizes an external gpt-5.6-luna model for open-ended judging during the RL process, indicating a focus on quality and nuanced evaluation of generated outputs.

Potential Use Cases

This model is particularly well-suited for applications that benefit from:

  • Advanced Reasoning: The RL setup with an external judge and enable_thinking suggests improved reasoning and response quality.
  • Long Context Processing: With a 32768 token context length, it can handle complex tasks requiring extensive input understanding.
  • Generative Tasks: Its fine-tuning and generation parameters indicate a focus on producing high-quality, coherent, and contextually relevant text.