groxaxo/Qwevolution-V0-27B
groxaxo/Qwevolution-V0-27B is a 27 billion parameter vision-language model, a weight-space blend of ConicCat/Qwen3.5-27B-Writer-V2 and a coding LoRA adapter. This model integrates enhanced coding capabilities while retaining the original vision processing of its base model. It is designed for tasks requiring both visual understanding and strong code generation or comprehension, leveraging its 32K token context length.
Loading preview...
Overview
groxaxo/Qwevolution-V0-27B is a 27 billion parameter model that combines the visual and writing capabilities of the ConicCat/Qwen3.5-27B-Writer-V2 base model with specialized coding enhancements from the groxaxo/Qwen3.5-27B-Writer-V2-Heretic-Coding-LoRA adapter. This model was created through a deliberate weight-space merge, ensuring that the coding improvements are integrated while preserving the original vision tower's functionality.
Key Composition and Features
- Base Model:
ConicCat/Qwen3.5-27B-Writer-V2, aQwen3_5ForConditionalGenerationarchitecture with a 32K context length, providing strong general language and vision capabilities. - LoRA Adapter:
groxaxo/Qwen3.5-27B-Writer-V2-Heretic-Coding-LoRA, which specifically targets and enhances coding performance. - Vision Capabilities: The vision tower remains identical to the base model, as the coding LoRA adapter did not modify these modules, ensuring consistent visual understanding.
- Merging Process: The merge was performed on CPU in bfloat16 with
safe_merge=True, and integrity was verified through reloading and forward pass checks.
When to Use This Model
- Combined Vision and Code Tasks: Ideal for applications that require understanding visual input alongside generating or analyzing code.
- Enhanced Coding: Benefits from the integrated coding LoRA, making it suitable for programming-related tasks.
- Writer-V2 Lineage: Maintains the characteristics of the Qwen3.5-27B-Writer-V2 family, offering robust language generation.