-
Hi,
I'm trying to port some models from Kornia. I was able to port NetVlad and LightGlue.
When it comes to [Disk](https://kornia.readthedocs.io/en/v0.6.12/feature.html#kornia.feature.DISK), the …
-
When a tensor is evaluated during compile, we currently raise an error or print a warning. However, we could make Tripy functionally correct even in the case of evaluation by simply not updating the f…
-
```py
import torch
import thunder
@thunder.jit
def f(x):
x.add_(1)
return x.copy_(x.sin())
f(torch.tensor(0.0, device='cuda'))
```
The above results in the following error from nv…
-
During the computation of cos/sin in [llama_rope#L119](https://github.com/tenstorrent/tt-metal/blob/skhorasgani/vllm_llama32_mm/models/demos/t3000/llama2_70b/tt/llama_rope.py#L119), when batch size is…
-
When I'm exporting disk+lightglue, does this warning mean that this variable is fixed to the value it had when exporting to ONNX? If that's the case, will there be any issues when running disk?
/…
-
```python
import torch
import thunder
from contextvars import ContextVar
_compile_data = ContextVar("compile_data", default=1)
def fn(x):
v = _compile_data.get()
return x + v
jfn…
-
This issue is related to `tp.mean` and `tp.var` failures when implementing `BatchNorm` using **Tripy** for Resnet50 model.
```python
class TPBatchNorm(tp.Module):
def __init__(self, num_features,…
-
I am trying compile LLaVA 1.5 7B to Neuron. As far as I can tell, the way to do this is to select some specific inputs and then trace the model execution with those inputs. However, when I try to trac…
-
### 🐛 Describe the bug
I tried to export and compile LSTM model and it's performance finally is much worse than in CUDA in terms of total kernel time and in number of operations
```(python)
impor…
-
**Describe the bug**
GraphCapture doesn't throw an error for L1 overflow in NO_DISPATCH mode
**To Reproduce**
E.g., the following block
```cpp
const auto input_tensor_a = ttnn::zeros(params.a_Shape, …