-
## Without fine-tuning the text encoder (Sayak)
* lr: [1e-6, 3e-6, 5e-6]
* max_train_steps: [800, 1000, 1200]
## With fine-tuning the text encoder (Chansung)
* lr: [2e-6, 5e-6]
* max_train_โฆ
-
Hello
Eureka is never avaiable.
Message in the log : "Eureka request failed to endpoint https://xxxx.dns:7553/eureka/apps, next server retry in 100000ms"
However we can connect to ZLUX. Inside โฆ
-
I currently get a warning that the extensions could not be found, and immediately a popup request to install when running `netlogo.load_model()`.
```
UserWarning: could not find default NetLogo exteโฆ
-
### ๐ Describe the bug
- I tried to use FSDP to wrap a Unet model, but somehow during `backward` some assert failed suggesting that the the FSDP model has unexpected states. Seems that the `pre_bacโฆ
-
cf: https://github.com/RadeonOpenCompute/ROCm/issues/1608
And **please don't close this issue** until we have a clear answer - has polaris support been intentionally dropped from ROCm
after only 6โฆ
-
No idea what can be wrong (except for using the same webui folder for almost 3 months), but 1.11.0+ version do not work for me - no "Autocomplete options" under checkpoint selection menu, no autocomplโฆ
-
is it possible to reduce size?
-
mixed_precision.set_global_policy('mixed_float16')
stable_diffusion_mixed_precision = keras_cv.models.StableDiffusion()
```
TypeError: Exception encountered when calling layer "clip_attentiโฆ
-
### ๐ Describe the bug
This bug occurs when trying to train using the train_collossalai.yaml
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ Traceback (most recent call last) โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฎ
โ โฆ
-
---------------------------------------------------------------------------
HTTPError Traceback (most recent call last)
[/usr/local/lib/python3.7/dist-packages/huโฆ