SaudF/qwen35-4b-dpo-v6-identity
SaudF/qwen35-4b-dpo-v6-identity is a 4.5 billion parameter language model based on the Qwen3.5 architecture, developed by SaudF. This model is a fully merged version of unsloth/Qwen3.5-4B with aziz9788/qwen35-saudidraft-full-final-adapter, designed to be loaded as a standalone Transformers or vLLM model without requiring an adapter. It is specifically noted that Identity SFT has not been applied, indicating a focus on its base merged capabilities.
Loading preview...
Model Overview
SaudF/qwen35-4b-dpo-v6-identity is a 4.5 billion parameter language model built upon the Qwen3.5 architecture. This model represents a direct merge of the unsloth/Qwen3.5-4B base weights with the aziz9788/qwen35-saudidraft-full-final-adapter, utilizing a lora_alpha=22 configuration. A key characteristic is that it is provided as a fully merged model, meaning it can be loaded and used directly with standard Transformers or vLLM libraries without the need for separate adapter loading.
Key Characteristics
- Architecture: Based on the Qwen3.5 model family.
- Parameter Count: 4.5 billion parameters.
- Integration: Fully merged weights, allowing for standalone use without adapters.
- Development: Result of merging
unsloth/Qwen3.5-4Bwithaziz9788/qwen35-saudidraft-full-final-adapter. - Identity SFT: Explicitly states that Identity SFT has not been applied, which differentiates its training and potential behavior from models that have undergone such fine-tuning.
Intended Use
This model is suitable for developers and researchers looking for a Qwen3.5-based model with specific merged adaptations, ready for direct deployment in Transformers or vLLM environments. Its standalone nature simplifies integration, making it a convenient option for applications where the specific saudidraft adaptations are desired without additional adapter management.