-
`git grep 'def copy'` reveals a number of classes that define ad-hoc `copy` methods but not `__copy__` methods. Hence, they do not implement the copy protocol https://docs.python.org/3/library/copy.…
-
Hi, thank you for your awesome works. However, when I was trying to run the M3DClip model using code on huggingface I have some errors related to the einops lib. I noticed you use the monai ViT layers…
-
## Problem Description
When using the `seamless_communication` library and specifically its `translator.predict()` method, I've encapsulated the core inference logic into reusable interfaces. Howev…
-
when I use cutlass template to write my own gemm kernel, I meet a Internal error, even I follow the settings provided by cutlass profiler.
The full code is as below:
#include
#include…
-
### 🐛 Describe the bug
version:
python 3.10
torch 2.1.0
cuda 12.1
put breakpoint in row x = x.detach()
```python
import torch
import torchvision
class BoQ_DinoV2(torch.nn.Module):
…
-
### 🐛 Describe the bug
`_copy` ops all accept an `out=` parameter. The usual behavior is this:
1. If the `out=` parameter is the correct size, the result is copied into it.
2. If the `out=` par…
-
### Your current environment
AMD radon + kubernetes
### Model Input Dumps
`vllm serve mistralai/Mistral-7B-Instruct-v0.3 --trust-remote-code --enable-chunked-prefill --max_num_batch…
-
### 🚀 The feature, motivation and pitch
Currently, when we try to save anything using `torch.save` method for which the full path doesn't exist, it throws an error.
We then need to first create the…
-
环境如下
系统：wsl2 + ubuntu 20.04
python版本：3.8
pytorch: 2.0.1
cuda: 11.7
出错运行代码如下，前面stage 0 - 4步运行都成功了
===================================
./run.sh --stage 5 --stop-stage 5
finetune...
rank: 0, pid…
-
### 🚀 The feature, motivation and pitch
**Overview**
The goal of this RFC is to discuss the integration of distributed inference into TorchChat. Distributed inference leverages tensor parallelism …