Stage-org/4b-strat-300-LH-27b-z-e1-iter3-epoch3

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 23, 2026Architecture:Transformer Featherless Exclusive Cold

The Stage-org/4b-strat-300-LH-27b-z-e1-iter3-epoch3 model is a 4.5 billion parameter language model with a 32768 token context length. It was developed by Stage-org and is the result of an iterative reinforcement learning (RL) training process. This model is specifically fine-tuned using a custom dataset and RL methods, indicating an optimization for strategic reasoning or complex task execution. Its primary use case is likely in environments requiring advanced decision-making or interactive problem-solving, leveraging its RL-trained capabilities.

Loading preview...

Overview

Stage-org/4b-strat-300-LH-27b-z-e1-iter3-epoch3 is a 4.5 billion parameter language model developed by Stage-org, featuring a substantial context length of 32768 tokens. This model has undergone an iterative reinforcement learning (RL) training process, building upon a base model (Stage-org/4b-strat-300-LH-27b-z-iter2-epoch3-agent-rl-epoch1-low_lr). The training involved 10,000 learner steps over 3 epochs, utilizing a custom dataset and a batch size of 128 with a sequence length of 300,000.

Key Capabilities

  • Reinforcement Learning (RL) Fine-tuning: Optimized through an RL method, suggesting enhanced capabilities in strategic decision-making and complex task execution.
  • Extended Context Window: Supports a 32768 token context length, enabling processing of longer inputs and maintaining coherence over extended interactions.
  • Advanced Inference Configuration: Utilizes flash_attention_2 for efficient attention mechanisms and vllm_extra with qwen3 reasoning and tool call parsers, indicating potential for sophisticated reasoning and tool-use integration.
  • Iterative Training: Benefits from an iterative training approach, likely leading to refined performance in its target domain.

Good For

  • Strategic Reasoning Tasks: Ideal for applications requiring models to make sequential decisions or engage in strategic planning due to its RL-based training.
  • Complex Problem Solving: Suitable for scenarios where understanding and generating responses based on extensive context is crucial.
  • Interactive AI Systems: Can be leveraged in environments that benefit from models capable of learning from interaction and feedback, such as agents or simulators.
  • Research in RL for LLMs: Provides a foundation for further experimentation and development in applying reinforcement learning to large language models.

Finetunes
5 models