onnxruntime
dcd072b1
- Prevent FP16 QDQ bias-scale underflow with FP32 dequantization
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
8 days ago
Prevent FP16 QDQ bias-scale underflow with FP32 dequantization Co-authored-by: tianleiwu <30328909+tianleiwu@users.noreply.github.com>
References
#33239 - [WIP] Fix float16 weights quantization to avoid zeroing out tensors
Author
Copilot
Parents
3e51806b
Loading