Autoround
An open-source weight quantization algorithm developed by Huawei Noah's Ark Lab. It uses an adaptive rounding strategy to maintain strong model quality under low-bit quantization, with particularly notable accuracy at 4-bit quantization.
Timeline (last 90 days)
When running Qwen3-Next-Flash (Autoround W4A16 quantization), the prefill speed was approximately 1.3k tokens/s, and the generation speed reached 70 tokens/s (code) / 60 tokens/s (prose), with support for 128k+ long context.
Autoround is an open-source weight quantization algorithm from Huawei Noah's Ark Lab that maintains better model quality at low bit-widths through an adaptive rounding strategy.
All Facts (2)
When running Qwen3-Next-Flash (Autoround W4A16 quantization), the prefill speed was approximately 1.3k tokens/s, and the generation speed reached 70 tokens/s (code) / 60 tokens/s (prose), with support for 128k+ long context.
50%UnverifiedAutoround is an open-source weight quantization algorithm from Huawei Noah's Ark Lab that maintains better model quality at low bit-widths through an adaptive rounding strategy.
50%