groxaxo/Code-Writer-V2-Obliterated-BF16
Code-Writer-V2-Obliterated-BF16 by groxaxo is a 27 billion parameter Qwen3.5-based model, preserving full BF16 precision. It is a merged model combining a Qwen3.5-27B-Writer-V2-uncensored-heretic base with a purpose-trained coding LoRA, featuring a long context of up to 200,000 tokens and a vision tower. This model excels at both creative writing and code generation, offering high fidelity for complex tasks.
Loading preview...
Overview
Code-Writer-V2-Obliterated-BF16 is a 27 billion parameter model developed by groxaxo, built upon the Qwen3.5 architecture. This version is a full-fidelity BF16 checkpoint, ensuring maximum precision without quantization. It merges the llmfan46/Qwen3.5-27B-Writer-V2-uncensored-heretic base with a specialized coding_mix_8k LoRA, making it highly capable in both creative writing and programming tasks.
Key Capabilities
- Full BF16 Precision: Offers the highest fidelity for advanced use cases, serving as a reference master or a base for further quantization.
- Extended Context Length: Supports an impressive context window of up to 200,000 tokens, ideal for processing entire codebases or lengthy manuscripts.
- Hybrid Architecture: Utilizes Qwen3.5's unique design with 64 decoder layers, including 16 full attention layers and GDN linear attention for long memory.
- Vision-Capable: Includes a vision tower, though its image processing capabilities are noted as present but untested.
- Uncensored & Free: Designed to be unrestricted in its output, providing direct responses without filtering or lecturing.
Good For
- Writing: Excels in generating fiction, screenplays, copy, and other forms of prose.
- Code Generation: Optimized for producing functional code, benefiting from its dedicated coding LoRA.
- Long-form Content: Its extensive context window makes it suitable for handling large documents, complex conversations, and entire projects.