-
I've been reading about how converting to half-precision/fp16 can potentially affect the quality in the model. Is there a way to disable this feature? That way, I can compare the two versions?
Than…
-
Thank you for bringing us this great audio restoration project.
I'm training a vocal stem enhancement model on a single RTX 3090 in bfloat16 precision (which reduced the vram usage in half).
In th…
-
while running the object detection using video_demo_half.py script (16-bit half precision floats), I am getting the following error
_Traceback (most recent call last):
File "video_demo_half.py",…
-
From Matt's comment on [756](https://github.com/amazon-ion/ion-rust/pull/756#discussion_r1592608371).
> Can we create an issue to track this? It would be great for users (and our own benchmarks) if…
-
As stated in the title, we may want to find a polynom based on arbitrary, lower than 32 bits, precision.
Looks like there is a lol/half but it's a very precise f16 implementation.
-
### Description
`jax.nn.dot_product_attention` does the first dot product with `preferred_element_type=jnp.float32` (see [here](https://github.com/jax-ml/jax/blob/7f655972c47658768b6ecce752fa29c3a…
-
### 🐛 Describe the bug
When I was using `leaky_relu` in PyTorch with the `half` type, I noticed there was a significant difference between PyTorch's output and NumPy/TensorFlow's output.
To furthe…
-
```
variant='diffusion_transformer_384p' # For low resolution
model_path = "/content/Pyramid-Flow" # The downloaded checkpoint dir
model_dtype = 'bf16'
device_id = 0
torch.cuda.set_…
-
When working on #2905, I had to manually set the threshold to 0.01, which seemed to be good enough up to 6 GPUs. testValidate doesn't work out of the box.
Here are several reasons that I'm aware of:…
-
Hello,
Thank you for sharing your fantastic work.
I'm curious if you have any intentions to share the training and inference configurations using fp16 to accommodate limited GPU memories.