jina-ai / rungpt

An open-source cloud-native of large multi-modal models (LMMs) serving framework.
Apache License 2.0
156 stars 20 forks source link

Trying to deploy via JCloud #75

Open mvsoom opened 1 year ago

mvsoom commented 1 year ago

I installed pip install open_gpt_torch today. The opengpt serve stabilityai/stablelm-tuned-alpha-3b --precision fp16 --device_map balanced command fails because open_gpt/resources/flow.yml.jinja2 cannot be found. I think it might be missing from the pip install from because of a missing MANIFEST.in file?

mvsoom commented 1 year ago

When I add the file manually (copy it from this repo), the command runs succesfully, but it returns this error:

(.venv) marnix@hp:~/ART/CAMERA/code$ opengpt deploy stabilityai/stablelm-tuned-alpha-3b --precision fp16 --device_map balanced --cloud jina --replicas 1
[08/06/23 21:41:16] INFO     Successfully validated flow config.     flow.py:158
                             Proceeding to flow deployment...                   
⠴ Deploying /tmp/tmpl65rkej9 • Submitting... ━━━━━━━━━━╺━━━━━━━━━ 1/2 • 0:00:00[jcloud                          ]  Successfully validated flow config. Proceeding to flow deployment...
[08/06/23 21:41:17] INFO     Successfully submitted flow with ID     flow.py:177
                             striking-elk-0cbc352599                            
⠧ Deploying /tmp/tmpl65rkej9 • Submitting... ━━━━━━━━━━╺━━━━━━━━━ 1/2 • 0:00:01[jcloud                          ]  Successfully submitted flow with ID [bold][blue]striking-elk-0cbc352599[/blue][/bold]
                    INFO     Check the Flow deployment logs:         flow.py:720
                             https://cloud.jina.ai/ !                           
⠧ Deploying /tmp/tmpl65rkej9 • Submitting... ━━━━━━━━━━╺━━━━━━━━━ 1/2 • 0:00:01[jcloud                          ]  Check the Flow deployment logs: [link=https://cloud.jina.ai/user/flows?action=detail&id=striking-elk-0cbc352599&tab=logs]https://cloud.jina.ai/[/link] !
**Unexpected phase: Failed reached at Starting for Flow ID striking-elk-0cbc352599**

Any idea how to solve this? I can see the Failed Flow in https://cloud.jina.ai/user/flows, but the Logs tab is empty. I tried all the variations I could think of, including different resource tiers (G2, G3) and models and quantization.

open_gpt.__version__ == '0.0.13' (+ manually copied open_gpt/resources/flow.yml.jinja2 file from this repo)
jina.__version__ == '3.20.0'
jcloud.__version__ = '0.2.16'
Python 3.9.5
Ubuntu 20.04.6 LTS

I'm currently on the free plan.

mvsoom commented 1 year ago

I can confirm the same error (Unexpected phase: Failed reached at Starting for Flow ID) happens after doing a fresh install using

git clone https://github.com/jina-ai/opengpt.git
pip install -e .
jemmyshin commented 1 year ago

Thanks for feedback, how long did you get this error after you executed the command opengpt serve?

mvsoom commented 1 year ago

About 10 minutes for every combination that I tried.

jemmyshin commented 1 year ago

This might be some issues in JCloud since it takes too much time to convert the JCloud YAML file to k8s YAML. Unfortunately, there isn't too much we can do at this time, but I will keep you updated once we have some progress here. Sorry for the inconvenient. Will it be possible to deploy the model on your local machine?

mvsoom commented 1 year ago

Unfortunately not. Is it possible to upload and use it to JCloud without docker, i.e. with source code format?

mvsoom commented 1 year ago

Bump