-
WSL2 +ROCm
xformers @ git+https://github.com/facebookresearch/xformers.git@133d7f1cb0a050f9ee8b8cd02fd9c906247e6c4e
```
got prompt
[rgthree] Using rgthree's optimized recursive execution.
[rgth…
-
### Python -VV
```shell
Python 3.11.9 (main, Apr 19 2024, 16:48:06) [GCC 11.2.0]
```
### Pip Freeze
conda env export
```shell
name: test
channels:
- defaults
dependencies:
-…
-
# ❓ Questions and Help
Hi! I work on vision transformers and implemented the window attention, where the feature map will be divided into local windows and self-attention among tokens will only happe…
-
# ❓ Questions and Help
Hi, I did a simple profiling of xformers cutlass implementation of attention vs flash attn. I was assuming they are the same algorithm with different implementation on Ampere+,…
-
-
ValueError: Incompatible shapes for attention inputs:
query.shape: torch.Size([32, 14400, 8, 40])
key.shape : torch.Size([2, 77, 8, 40])
value.shape: torch.Size([2, 77, 8, 40])
…
-
I am trying to use the HF model "mlfoundations/dclm-7b-it" for inference, simply using the code below:
```
model = AutoModelForCausalLM.from_pretrained("mlfoundations/dclm-7b-it")
gen_kwargs = {"…
-
The main use for pallas is for tpu, but right now, a lot of options related to TPU compiler is left undocumented and require guesswork.
Documentation for `PrefetchScalarGridSpec` is incomplete, la…
-
Hello
I'm trying to extract the results of the net in its second to last and last layers to build two attention modules for semantic segmentation. This is the code:
```
class efficientnet_b5(torc…
-
我在本地运行 cli_demo.py 时出现如下错误:
init model ...
**A matching Triton is not available, some optimizations will not be enabled.
Error caught was: DLL load failed while importing libtriton: 找不到指定的模块。**
…