ByteShape 发布 Qwen3.8-27B GGUF 量化模型
原标题:byteshape/Qwen3.8-27B-GGUF
AI 摘要
ByteShape 在 Hugging Face 发布 Qwen3.8-27B 的 GGUF 量化版本,使用其 ShapeLearn 方法为每个张量学习最优数据类型,以在极低位宽下保持质量。该版本取代此前的 ShapeLearn-Lite 量化,包含 5 个 GPU 优化模型(2.56–3.84 bpw,8.8–13.1 GB),并在数学、编程、指令遵循和智能体工具使用等基准上对照 BF16 基线评测。模型支持视觉输入,内置 MTP 头以支持投机解码,需 llama.cpp b10658 及以上版本。
正文节选
--- library_name: transformers license: apache-2.0 license_link: https://huggingface.co/Qwen/Qwen3.8-27B/blob/main/LICENSE pipeline_tag: image-text-to-text base_model: - Qwen/Qwen3.8-27B tags: - qwen3.8 - byteshape - shapelearn - mtp - dflash --- # Qwen3.8-27B GGUF (ShapeLearn Quantized) This is a GGUF-quantized version of Qwen3.8-27B produced with **ByteShape's ShapeLearn**, which learns the optimal datatype per tensor to maintain high quality even at very low bitlengths. This release replac