issues
search
aredden
/
flux-fp8-api
Flux diffusion model implementation using quantized fp8 matmul & remaining layers use faster half precision accumulate, which is ~2x faster on consumer devices.
Apache License 2.0
105
stars
12
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Docker image support.
#17
ShivamB25
opened
6 days ago
5
How to save a "prequantized_flow" safetensor?
#16
smuelpeng
opened
6 days ago
3
Any plans for controlnet + inpainting support?
#15
0xtempest
opened
6 days ago
1
add benchmarks numbers for rtx4000ada (non-sff)
#14
flowpoint
closed
2 days ago
1
LoRA loading fails if only trained on specific blocks
#13
fblissjr
opened
1 week ago
17
Consider adding a license to the code
#12
flowpoint
closed
1 week ago
4
add h100
#11
ClashLuke
closed
1 week ago
1
Where is the code about "remaining layers use faster half precision accumulate"?
#10
goldhuang
opened
1 week ago
3
Potential LoRA performance issue
#9
ashakoen
opened
2 weeks ago
7
`NotImplementedError: Cannot copy out of meta tensor; no data!`
#8
montyanderson
opened
2 weeks ago
3
WHL files for torch-cublas-hgemm.git and ao?
#7
SoftologyPro
closed
2 weeks ago
2
[feature] Support for ControlNet from x-lab
#6
George0726
closed
2 weeks ago
3
Error No module named 'cublas_ops'
#5
ankitsiliconithub
closed
1 week ago
2
No issue - just a thank you!
#4
ashakoen
closed
1 week ago
2
Improved but configurable precision
#3
aredden
closed
3 weeks ago
0
python 3.10 compatibility
#2
XmYx
closed
3 weeks ago
1
update README
#1
dsingal0
closed
3 weeks ago
1