| Titre | pytorch pytorch (in torch.jit.jit_module_from_flatbuffer) torch 2.6.0 Memory Leak |
|---|
| Description | I encountered a Segmentation fault when running a simple PyTorch script that uses torch.jit.script along with Flatbuffer serialization. The issue occurs when loading the saved model from the Flatbuffer file and passing an input tensor to it.
# Code to Reproduce
```
import torch
from torch import nn
simple_model = nn.Sequential(
nn.Linear(10, 20),
nn.BatchNorm2d(5),
nn.ReLU()
)
scripted_model = torch.jit.script(simple_model)
torch.jit.save_jit_module_to_flatbuffer(scripted_model, 'model.ff')
loaded_model = torch.jit.jit_module_from_flatbuffer('model.ff')
sample_input = torch.rand(1, 5, 3, 10)
_ = loaded_model(sample_input)
```
|
|---|
| La source | ⚠️ https://github.com/pytorch/pytorch/issues/149800 |
|---|
| Utilisateur | Default436352 (UID 81891) |
|---|
| Soumission | 22/03/2025 10:08 (il y a 2 ans) |
|---|
| Modérer | 02/04/2025 15:35 (11 days later) |
|---|
| Statut | Accepté |
|---|
| Entrée VulDB | 303012 [PyTorch 2.6.0 torch.jit.jit_module_from_flatbuffer buffer overflow] |
|---|
| Points | 20 |
|---|