-
I just installed Rawtherapee on a Windows 10 laptop, which comes with a decent trackpad and a handful of useful gestures. I'm most interested in the zoom and swipe to scroll between photos.
Looking…
-
@giorgiodaneri and myself are working on a university project about mapping various transformer architectures on different CiM implementations.
It would be nice to have more details about the workloa…
-
Hi,
I'm having some trouble to get any tested device to lock on simulated location, using Adalm-Pluto:
Linux pluto 4.14.0-41915-gc2041af #279 SMP PREEMPT Mon Jan 14 13:13:47 CET 2019 armv7l GNU/…
-
**Is your feature request related to a problem? Please describe.**
Need to reduce model size of YOLOv10 while maintaining performance.
**Describe the solution you'd like**
Sparse and Quantizatio…
-
Model can be in fp16 instead of fp32 when quantizing to int8/uint8.
For example, brevitas quantizes such fp16 models to int8. In such cases,
ONNX models have "cast"/"convert" nodes before and …
-
## 🐛 Bug: Quantization - we need a better solution for tracking quantization backend settings in a model
Currently, there are various points of confusion:
1. a target backend (qnnpack / fbgemm) is…
vkuzo updated
11 months ago
-
### System Info
transformers==4.42.3
torch==2.3.0
### Who can help?
_No response_
### Information
- [X] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially…
-
I noticed that the default value for the number of position bits when quantizing positions with gltfpack is 14 bits, while the component type is uint16, which means that 2 bits of data is cropped unle…
-
Good day everyone, I am trying to run llama agentic system on RTX4090 with FP8 Quantization for the inference model and meta-llama/Llama-Guard-3-8B-INT8 for the Guard. WIth sufficiently small max_seq_…
anret updated
2 months ago
-
## ❔Question
Did someone tried model **post training dynamic quantization**?
When I quantize, model size is increasing twice and inferencing time is same with FP32 model.
Based on pytorch tutor…