issues Search Results · repo:mit-han-lab/nunchaku language:Cuda
Filter by
214 results
(71 ms)214 results
inmit-han-lab/nunchaku (press backspace or delete to remove)Some acceleration techniques, such as TeaCache rely on having access for inference to individual layers in the FLUX
transformer. To make them interoperable with Nunchaku, the Nunchaku library should export ...
andreaferretti
- Opened yesterday
- #233
When using the fp4 model, an error crash will be reported when the generation width is greater than 1500. The error is
ssertion failed: k = MAXK, file D:\Project\nunchaku-dev\src\kernels\misc_kernels.cu, ...
CGF985
- Opened yesterday
- #232
I need to use some slider lora for flux to adjust a few things, they work very well on specific concept, such as skin
tone. The nature of slider lora requires it to be used from -4 to +4, but I noticed ...
frankyifei
- Opened 2 days ago
- #231
GPU5070ti,The other parameters of the same set of prompt words are the same, but there are subtle differences in the
results. This will not happen if the model with fp8 is used.
Image Image
- 2
- Opened 5 days ago
- #229
Since Flex-Alpha is a de-distilled version of the Flux Schnell, maybe it could be supported. I think the original unet
is pruned, but maybe many things remain unchanged. Since it has a more permissive ...
vgabbo
- 3
- Opened 7 days ago
- #228
Hey there! I ve got a super simple question after doing a ton of code searching. Which code can show that the activation
s quantization is based on a per group size of 64? Since I learned from
quantize_w4a4_from_fpsum_warp(https://github.com/mit-han-lab/nunchaku/blob/main/src/kernels/zgemm/gemm_w4a4.cuh#L460) ...
question
unbelievable3513
- Opened 8 days ago
- #227
I have followed the steps and installed nightly pytorch, but still got this error Unsupported SM 120
pip install -e . --no-build-isolation
Obtaining file:///root/nunchaku
Checking if build backend supports ...
valiantljk
- 1
- Opened 8 days ago
- #226
Loading weights from F:\ComfyUI\ComfyUI\models\diffusion_models\svdq-int4-flux.1-fill-dev\transformer_blocks.safetensors
[2025-03-25 09:58:45.044] [warning] Failed to load safetensors using method MIO: ...
xueqing0622
- 3
- Opened 8 days ago
- #225
Hi @lmxyy
I just update ComfyUI to Pytorch 2.8 however I can t install Nunchaku anymore, I try to re compile Nunchaku for
Pytorch 2.8 but failed. This is the error log:
Nunchaku Error Log.docx
This ...
pivtienduc
- 3
- Opened 9 days ago
- #224
Hello, I am doing an experiment about lora-like layer s performance under different precision and hardware platform. I
would appreciate it if you can provide a single fused kernel mentioned in original ...
Ther-nullptr
- Opened 10 days ago
- #223
Learn how you can use GitHub Issues to plan and track your work.
Save views for sprints, backlogs, teams, or releases. Rank, sort, and filter issues to suit the occasion. The possibilities are endless.Learn more about GitHub IssuesProTip!
Press the /
key to activate the search input again and adjust your query.Learn how you can use GitHub Issues to plan and track your work.
Save views for sprints, backlogs, teams, or releases. Rank, sort, and filter issues to suit the occasion. The possibilities are endless.Learn more about GitHub IssuesProTip!
Restrict your search to the title by using the in:title qualifier.