-
`Flux diffusion model implementation using quantized fp8 matmul & remaining layers use faster half precision accumulate, which is ~2x faster on consumer devices.`
Hello there!
Thanks for sharing you…
-
The following IR will crash `llc` on (at least) all releases between (and including) 16 and 18:
```
define void @drjit_641b8f0d1521b37df5f00905a3202c37(i64 %start, i64 %end, i32 %thread_id, ptr no…
-
This issue is mainly for tracking the upstream counterpart (https://github.com/numpy/numpy/issues/14753) for compatibility purposes. However, there are unique challenges and opportunities in CUDA that…
-
我想问下现在模型支持半精度吗
-
Hey,
just wanted to ask whether it is safe to run DINOv2 in half precision for inference. Is there any degradation in the quality of the features?
Thanks!
-
As of now, usage of half precision is not straightforward. Not only for extension libraries as mentioned in issue #266 , but also for the generation of standard kernels. E.g., using `half` (from the …
-
phold version: v0.2.0
Python version: 3.11
Operating System: Linux
### Description
Running phold predict on multiple .gbk pharokka outputs
The process is stuck at the stage pasted below a…
-
BFMap fails without good debugging information when trying to use half precision floats.
Minimal example:
```from bifrost import ndarray as BFArray
from bifrost import map as BFMap
import numpy a…
-
CUDA supports efficient computation with half-precision (16-bit) floats. This is probably enough precision for the pixel data in our problem. We might do this with CUDA's half2 type, but this would …
-
Hi Jacob,
I tried to estimate the size of my model that uses half precision without any luck.
I changed line 34 to HalfTensor, but got the following message:
RuntimeError: "unfolded3d_copy_cpu…