Skip to content

Commit

Permalink
Update README.md
Browse files Browse the repository at this point in the history
  • Loading branch information
billlight authored Mar 29, 2024
1 parent ee540e3 commit 4432d97
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -81,7 +81,7 @@ ONNX分支:
## 模型量化


小米SD采用训练后量化(PTQ)将Float32的模型转换为INT8数据类型的量化模型,将模型大小缩减到了约${1} \over {4}$的大小,模型大小约为1.1GB。
小米SD采用训练后量化(PTQ)将Float32的模型转换为INT8数据类型的量化模型,将模型大小缩减到了约$`\frac{1}{4}`$的大小,模型大小约为1.1GB。

通过收集高精度校准数据和优化量化方法方法对模型进行量化:
* 校准数据集收集:收集多领域、高质量prompt数据集,作为Calibration数据集。
Expand Down

0 comments on commit 4432d97

Please sign in to comment.