Multiverse Computing presents Quantization-Aware Healing research

更新摘要

Multiverse Computing describes Quantization-Aware Healing, which distills a compressed and quantized model directly from the original full-precision teacher. In an experiment compressing GPT-OSS 120B to 60B and quantizing it to MXFP4, the model exceeded the corresponding bfloat16 checkpoint on seven of nine benchmarks.

本页仅索引并客观转述公开信息，产品能力、价格与可用范围以官方原文为准。