History

Jeremy Lewi 2e6e891a5b Update the ArgoCD app to use the kubeflow/examples repo (#440 ) * We were using jlewi's fork because PRs hadn't been committed but all the relevant PRs have been merged and master is the source of truth.		2018-12-19 21:26:49 -08:00
..
cs-demo-1103	Update the ArgoCD app to use the kubeflow/examples repo (#440 )	2018-12-19 21:26:49 -08:00
README.md	Fix the web app. (#432 )	2018-12-12 09:24:40 -08:00
count_lines_of_code.py	Create a script to count lines of code. (#379 )	2018-12-19 09:42:25 -08:00

README.md

Demo

This directory contains assets for setting up a demo of the code search example. It is primarily intended for use by Kubeflow contributors working on the shared demo.

Users looking to run the example should follow the README.md in the parent directory.

GCP Resources

We are using the following project

org: kubeflow.org
project: code-search-demo
code-search-team@kubeflow.org Google group administering access

Deploying the services

Deploy the TFServing server

ks12 show  cs_demo -c t2t-code-search-serving

Deploy the UI and nmslib index server

ks12 apply  cs_demo -c search-index-server

Copy the GCP service account to the namespace where the servers run
- The serving piece runs in a different namespace from Kubeflow
- We need to copy the GCP service account to that namespace because the pod will try to mount it.
```
kubectl -n kubeflow get secret user-gcp-sa -o json | jq -r '.data["user-gcp-sa.json"]' | base64 -d > ${SECRET_FILE}
kubectl -n cs-web-app create secret generic user-gcp-sa --from-file=user-gcp-sa.json=${SECRET_FILE}
```

Install Argo CD

kubectl create namespace argocd
cd cs-demo-1103/k8s_specs
kubectl apply -f argocd.yaml

Create the app

SRC_URL=https://github.com/kubeflow/examples.git
KS_PATH=code_search/kubeflow
argocd app create code-search --name kubeflow --repo $SRC_URL --path ${KS_PATH} --env ${KS_ENV}

Results

2018-11-05

jlewi@ ran experiments that produced the following results

What	location	Description
Preprocessed data	gs://code-search-demo/20181104/data/func-doc-pairs-00???-of-00100.csv	This is the output of the Dataflow preprocessing job
Training data	gs://code-search-demo/20181104/data/kf_github_function_docstring-train-00???-of-00100	TFRecord files produced by running T2T datagen

Models

hparams	Location
transformer_tine	gs://code-search-demo/models/20181105-tinyparams/
transformer_base_single_gpu	gs://code-search-demo/models/20181105-single-gpu
transformer_base	gs://code-search-demo/models/20181107-dist-sync-gpu

Performance

hparams	Resources	Steps/sec
transformer_tiny	1 CPU worker	~1.8 global step /sec
transformer_base_single_gpu	1 GPU worker (K80)	~3.22611 global step /sec
transformer_base	1 chief with K80, 8 workers with 1 K80, sync training	~ 0.0588723 global step /sec
transformer_base	1 chief (no GPU), 8 workers (no GPU), sync training	~ 0.707014 global step /sec