Aamir Nazir

Make NNCF common utils code pass mypy checks

5

### Changes Made all the mypy checks pass for nncf/common/utils ### Related tickets Closes Issue #2491

[Torch FX] Stable Diffusion v3

1

Notebook which includes quantizing and compressing Stable diffusion v3 model with Torch FX.

[Torch FX] Add Special FX Metatype in Torch Operator Metatypes

3

### Changes 1. Create a new registry for FX metatypes 2. Create new metatype `FXEmbeddingMetatype` for torch FX embedding metatype ### Reason for changes The current torch embedding metatype, which...

[Torch FX] Support for ExportedProgram from torch.export.export in Weights Compression

1

### 🚀 Feature request Currently, the Torch FX backend supports FX representation(`torch.fx.GraphModule()` object) extracted using `capture_pre_autograd_graph()` which has a slightly different representation and is a deprecated feature. This feature request...

DeformConv2d Cannot be Quantized

3

### 🐛 Describe the bug Operator Metatype `DeformConv2dOp` is mentioned in `nncf/nncf/torch/graph/operator_metatypes.py` which uses the namespace `torch.nn.functional` whereas the function `deform_conv2d` belongs to `torchvision.ops.deform_conv2d`. As seen in the code output...

### Details: With the latest conformance test runs, it was found that the operator `torch.ops.aten.bernoulli.p` was causing graph breaks due to being unsupported. This PR aims to add support for...

[Torch FX][SD3] Add Dynamic Shapes

6

Add Dynamic shapes to export of transformer model in SD3 Torch FX notebook.

[Examples][FX] WC Example for TorchFX Backend

2

### Changes Add example script for TorchFX backend in `examples/`. ### Reason for changes This is to show a sample script for applying WC to LLMs which are exported using...

[Torch FX] Map Namespace Names to Metatype

3

### Changes Created a new dictionary in operator metatypes registry to maintain mapping of namespace and function name to metatype. ### Reason for changes For a more accurate retrieval of...

### Changes The core idea of this change is to first unsqueeze the weights so that it becomes 3D. Even the 2D weights. Then the rest of the algorithm implementation...