Feng Chen
Feng Chen
The gpu will split multi-instance drawing into multiple drawing commands, and the current drawing scheduling scheme incorrectly identifies this kind of drawing, resulting in the vertex shader unable to obtain...
Close #7171 Before：  After：  Before：  After：  This should fix more other games, more tests are welcome
An optimization for PR #7375, the inconsistency of location was not handled well at that time. With PR #8858 DQB can playable now. Before:  After:  Vulkan  GLSL...
It seems that some games use drawing to generate mipmap textures, and the emulator does not currently support this behaviour very well. When this behaviour is triggered, the viewport transform...
The moon of LIVE A LIVE is a 9-layer astc texture, but the data to be decoded is very strange, only the first two layers are normal, from the third...
The previous implementation would cause opengl to crash in non-unified_memory mode, and the generated buffer would not be recycled. Now we have a unified implementation of vulkan and opengl.
Hi Stability-AI team, Thank you for your outstanding work! I looked at the code base and only found the test code, are there any plans to publish the training code?...
### Describe the feature I found that the train_dreambooth_colossalai script you provide no longer supports train_text_encoder parameter, but it's gives much better results especially on faces, so I would like...
Awesome work! I have read your paper and noticed that your training data is extracted from VGG2 and Voxceleb, and substantial efforts have been put into it. Could you please...
**Description** Using tritonserver to delay loading(--model-control-mode=explicit) the llava-mixtral-8x7b model, there is a probability that when my client initiates load_model, it triggers the server to load the same model multiple times(There...