docs: update operator list in torch support's documentation section #486
Conversation
| - [`torch.nn.BatchNorm2d`](https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm2d.html) | ||
| - [`torch.nn.BatchNorm3d`](https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm3d.html) | ||
| - [`torch.erf, torch.special.erf`](https://pytorch.org/docs/stable/special.html#torch.special.erf) |
There was a problem hiding this comment.
I have a doubt in putting these two here (erf + pad), but I think it's ok
| - [`torch.ge, torch.greater_equal`](https://pytorch.org/docs/stable/generated/torch.ge.html) | ||
| - [`torch.lt, torch.less`](https://pytorch.org/docs/stable/generated/torch.lt.html) | ||
| - [`torch.le, torch.less_equal`](https://pytorch.org/docs/stable/generated/torch.le.html) | ||
| - [`torch.eq`](https://pytorch.org/docs/stable/generated/torch.eq.html) |
There was a problem hiding this comment.
side note: torch.eq is not the same as torch.equal , and I think we support torch.eq but to be confirmed maybe
|
|
||
| - `brevitas.nn.QuantLinear` | ||
| - `brevitas.nn.QuantConv1d` |
There was a problem hiding this comment.
I have a doubt here : is supporting torch.conv1d enough for supporting brevitas.nn.QuantConv1d ?
There was a problem hiding this comment.
I think we support all Brevitas operators that are equivalent to those already supported in Torch.
So the first sentence seems more appropriate.
Should we add: to this list: qnn.QuantLinear, qnn.QuantIdentity, qnn.QuantReLU
There was a problem hiding this comment.
qnn.QuantLinear is in this list and qnn.QuantIdentity is below in the "Quantizers" sub-section
but for other like activation functions, @andrei-stoian-zama pointed out to me that we should not suggest users to use them as :
- it's less certain that we indeed do support them
- we most likely never need them after all, as activation functions don't require quantization (done with a TLU), we can just use
torch.relu
so I don't think there's much to add here !
|
|
||
| - `brevitas.nn.QuantLinear` | ||
| - `brevitas.nn.QuantConv1d` |
RomanBredehoft
deleted the
docs/update_torch_support_operator_list_4268
branch
A user pointed that that this list was not up to date. Turns out we indeed added/changed quite some things since we've updated this section.
closes https://github.com/zama-ai/concrete-ml-internal/issues/4268