Phantomcloak19/qwen2.5-3b-dpo

TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 29, 2026Architecture:Transformer Cold

Phantomcloak19/qwen2.5-3b-dpo is a 3.1 billion parameter causal language model based on the Qwen2.5-3B-Instruct architecture. This model represents the DPO (Direct Preference Optimization) phase of the HorusLLM sequential training pipeline, following an initial Supervised Fine-Tuning (SFT) stage. It is specifically optimized through DPO, indicating a focus on aligning its outputs with human preferences and instructions, making it suitable for applications requiring nuanced and preferred responses.

Loading preview...

Overview

Phantomcloak19/qwen2.5-3b-dpo is a 3.1 billion parameter language model derived from the Qwen/Qwen2.5-3B-Instruct base model. It is a key component of the HorusLLM sequential training pipeline, specifically representing the DPO (Direct Preference Optimization) phase. This DPO stage follows an initial Supervised Fine-Tuning (SFT) phase and precedes a Safety-GRPO phase, indicating a structured approach to model development.

Key Characteristics

  • Base Model: Built upon the robust Qwen/Qwen2.5-3B-Instruct architecture.
  • Training Phase: Optimized using Direct Preference Optimization (DPO), which aims to align the model's responses more closely with human preferences and instructions.
  • Pipeline Integration: Part of a multi-stage training process (SFT → DPO → Safety-GRPO) designed for comprehensive model refinement.

Good For

  • Applications requiring models fine-tuned for human preference alignment.
  • Use cases where instruction following and generating preferred responses are critical.
  • Further experimentation or integration into pipelines that benefit from a DPO-optimized base.