-
**Describe the bug**
Error while serving a toy model using EasyDel with the following exception.
**To Reproduce**
Context
* Using Google TPU VM
* Follow the instructions as suggested here ht…
-
Is there a versatile transformers-like API (like model.generate()) equivalent for this? I tried JAXServer but it is quite confusing, and I couldnt get flashattention to work. Could you maybe provide s…
-
An exception is generated while running a gpt2 format model as show below
**To Reproduce**
Prepare to serve a model
```
run examples/serving/causal-lm/artgpt2tox-chat.py --pretrained_model_nam…
-
Hi, I tried finetune a model on TPU VM v3-8. when not using flash attention, it works. However, when I set config.use_flash_attention =True, an error occurs: block_q=1024 should be smaller or equal to…
-
#34
I read this issue and tried it. but couldn't make it work :(
Hi, Thank you for your amazing work.
I've been trying few days to make tpu v3-32 to work.
I used tpu VM 'tpu-ubuntu220…
-
**To Reproduce**
Use a TPU v2_8 with vm architecture
```
Time For configure functions and sharding them (ms) : 2012.7429962158203
Action : Sharding Passed Parameters
Model Contain 1.1000483…
-
**Describe the bug**
I use the example code provided on the documentation (https://erfanzar.github.io/EasyDeL/Llama2/).
But I changed the model to 'hfl/chinese-alpaca-2-7b-16k'. (The model is base…
-
**Describe the bug**
A clear and concise description of what the bug is.
**To Reproduce**
Output from Hugging Face Transformer APIs on local env
```
You are an oracle who knows the anwer…
-
**Describe the bug**
pip install easydel
/usr/local/lib/python3.8/dist-packages/pkg_resources/__init__.py:123: PkgResourcesDeprecationWarning: 0.1.36ubuntu1 is an invalid version and will not be sup…
-
In one of the longer training runs that is now running on a tpu-v3-8 I noticed the training ETA kept getting later and later.
Also in the step-time wandb log (picture below) the higher the step numbe…