Skip to content

Commit

Permalink
[WIP] Add new GPU runner for E2E tests
Browse files Browse the repository at this point in the history
Use previous GPU runner for unit tests

Signed-off-by: Nathan Weinberg <[email protected]>
  • Loading branch information
nathan-weinberg committed Jul 16, 2024
1 parent ae6097f commit ab0af26
Show file tree
Hide file tree
Showing 10 changed files with 300 additions and 85 deletions.
205 changes: 205 additions & 0 deletions .github/workflows/e2e-nvidia-a10g-x1.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,205 @@
# SPDX-License-Identifier: Apache-2.0

name: E2E (NVIDIA A10G x1)

on:
workflow_dispatch:
inputs:
pr_or_branch:
description: 'pull request number or branch name'
required: true
default: 'main'

jobs:
start-runner:
name: Start external EC2 runner
runs-on: ubuntu-latest
outputs:
label: ${{ steps.start-ec2-runner.outputs.label }}
ec2-instance-id: ${{ steps.start-ec2-runner.outputs.ec2-instance-id }}
steps:
- name: Configure AWS credentials
uses: aws-actions/configure-aws-credentials@e3dd6a429d7300a6a4c196c26e071d42e0343502 # v4.0.2
with:
aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
aws-region: ${{ secrets.AWS_REGION }}
- name: Start EC2 runner
id: start-ec2-runner
uses: machulav/ec2-github-runner@fcfb31a5760dad1314a64a0e172b78ec6fc8a17e # v2.3.6
with:
mode: start
github-token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
ec2-image-id: ami-00c51d9c1374eda97
ec2-instance-type: g5.2xlarge
subnet-id: subnet-02d230cffd9385bd4
security-group-id: sg-06300447c4a5fbef3
iam-role-name: instructlab-ci-runner
aws-resource-tags: >
[
{"Key": "Name", "Value": "instructlab-ci-github-runner"},
{"Key": "GitHubRepository", "Value": "${{ github.repository }}"}
]
e2e:
name: E2E Test
needs: start-runner
runs-on: ${{ needs.start-runner.outputs.label }}

permissions:
pull-requests: write

steps:
- name: Checkout instructlab/eval
uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
with:
# https://github.com/actions/checkout/issues/249
fetch-depth: 0

- name: Checkout instructlab/instructlab
uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
with:
repository: "instructlab/instructlab"
path: "instructlab"
fetch-depth: 0

- name: Determine if pr_or_branch is a PR number
id: check_pr
run: |
nvidia-smi
if [[ "${{ github.event.inputs.pr_or_branch }}" =~ ^[0-9]+$ ]]; then
echo "is_pr=true" >> "$GITHUB_OUTPUT"
else
echo "is_pr=false" >> "$GITHUB_OUTPUT"
fi
- name: Check if gh cli is installed
id: gh_cli
run: |
if command -v gh &> /dev/null ; then
echo "gh_cli_installed=true" >> "$GITHUB_OUTPUT"
else
echo "gh_cli_installed=false" >> "$GITHUB_OUTPUT"
fi
- name: Install gh CLI
if: steps.gh_cli.outputs.gh_cli_installed == 'false'
run: |
sudo dnf install 'dnf-command(config-manager)' -y
sudo dnf config-manager --add-repo https://cli.github.com/packages/rpm/gh-cli.repo
sudo dnf install gh --repo gh-cli -y
- name: test gh CLI
run: |
gh --version
- name: set default repo
run: |
gh repo set-default ${{ github.server_url }}/${{ github.repository }}
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

- name: Add comment to PR
if: steps.check_pr.outputs.is_pr == 'true'
run: |
gh pr comment "${{ github.event.inputs.pr_or_branch }}" -b "${{ github.workflow }} workflow launched on this PR: [View run](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }})"
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

- name: Fetch and checkout PR
if: steps.check_pr.outputs.is_pr == 'true'
run: |
gh pr checkout ${{ github.event.inputs.pr_or_branch }}
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

- name: Checkout branch
if: steps.check_pr.outputs.is_pr == 'false'
run: |
git checkout ${{ github.event.inputs.pr_or_branch }}
- name: Install Packages
run: |
cat /etc/os-release
sudo dnf install -y gcc gcc-c++ make git python3.11 python3.11-devel
- name: Install ilab
run: |
export CUDA_HOME="/usr/local/cuda"
export LD_LIBRRY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64"
export PATH="$PATH:$CUDA_HOME/bin"
python3.11 -m venv venv
. venv/bin/activate
nvidia-smi
sed 's/\[.*\]//' requirements.txt > constraints.txt
python3.11 -m pip cache remove llama_cpp_python
CMAKE_ARGS="-DLLAMA_CUBLAS=on" python3.11 -m pip install --force-reinstall --no-binary llama_cpp_python -c constraints.txt llama_cpp_python
python3.11 -m pip install bitsandbytes
# TODO This should be added to instructlab-training
python3.11 -m pip install packaging wheel
python3.11 -m pip install instructlab-training[cuda]
# Install the local version of eval before installing the CLI so PR changes are included
python3.11 -m pip install .
python3.11 -m pip install instructlab
- name: Run e2e test
run: |
nvidia-smi
# This env variable is used on GPUs with less vRAM. It only allows cuda to alloc small chunks of vRAM at a time, usually helping to avoid OOM errors.
# This is not a good solution for production code, as setting ENV variables for users isn't best practice. However, it is a helpful manual workaround.
export PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.6,max_split_size_mb:128
. venv/bin/activate
# TODO: for some reason we need to reinstall DS in order to get fused adam support
# This means we need to manually rm and re-install a bunch of packages. Investigate why this is.
python3.11 -m pip uninstall -y deepspeed
python3.11 -m pip cache purge
DS_BUILD_CPU_ADAM=1 BUILD_UTILS=1 python3.11 -m pip install deepspeed
nvidia-smi
python3.11 -m pip show nvidia-nccl-cu12
cd instructlab
./scripts/basic-workflow-tests.sh -em
- name: Add comment to PR if the workflow failed
if: failure() && steps.check_pr.outputs.is_pr == 'true'
run: |
gh pr comment "${{ github.event.inputs.pr_or_branch }}" -b "e2e workflow failed on this PR: [View run](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}), please investigate."
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

- name: Add comment to PR if the workflow succeeded
if: success() && steps.check_pr.outputs.is_pr == 'true'
run: |
gh pr comment "${{ github.event.inputs.pr_or_branch }}" -b "e2e workflow succeeded on this PR: [View run](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}), congrats!"
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

stop-runner:
name: Stop external EC2 runner
needs:
- start-runner
- e2e
runs-on: ubuntu-latest
if: ${{ always() }}
steps:
- name: Configure AWS credentials
uses: aws-actions/configure-aws-credentials@e3dd6a429d7300a6a4c196c26e071d42e0343502 # v4.0.2
with:
aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
aws-region: ${{ secrets.AWS_REGION }}
- name: Stop EC2 runner
uses: machulav/ec2-github-runner@fcfb31a5760dad1314a64a0e172b78ec6fc8a17e # v2.3.6
with:
mode: stop
github-token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
label: ${{ needs.start-runner.outputs.label }}
ec2-instance-id: ${{ needs.start-runner.outputs.ec2-instance-id }}
58 changes: 32 additions & 26 deletions .github/workflows/e2e.yml → .github/workflows/test.yml
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
# SPDX-License-Identifier: Apache-2.0

name: E2E test
name: Test

on:
push:
Expand All @@ -11,7 +11,7 @@ on:
- '**.py'
- 'pyproject.toml'
- 'requirements*.txt'
- '.github/workflows/e2e.yml'
- '.github/workflows/test.yml'
pull_request:
branches:
- "main"
Expand All @@ -20,19 +20,15 @@ on:
- '**.py'
- 'pyproject.toml'
- 'requirements*.txt'
- '.github/workflows/e2e.yml'
- '.github/workflows/test.yml'

concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
cancel-in-progress: true

jobs:
e2e:
unit:
runs-on: ubuntu-gpu

permissions:
pull-requests: write

steps:
# No step-security/harden-runner since this is a self-hosted runner
- name: Checkout instructlab/eval
Expand All @@ -41,14 +37,23 @@ jobs:
# https://github.com/actions/checkout/issues/249
fetch-depth: 0

# this is needed for branch tests
- name: Checkout instructlab/taxonomy
uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
with:
repository: "instructlab/taxonomy"
path: "taxonomy"
fetch-depth: 0

# this is needed for judge_answer tests
- name: Checkout instructlab/instructlab
uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
with:
repository: "instructlab/instructlab"
path: "instructlab"
fetch-depth: 0

- name: Install Packages
- name: Install system packages
run: |
sudo apt-get install -y cuda-toolkit git cmake build-essential virtualenv
nvidia-smi
Expand All @@ -67,34 +72,35 @@ jobs:
run: |
pip cache remove llama_cpp_python
- name: Cache huggingface
uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
with:
path: ~/.cache/huggingface
# config contains DEFAULT_MODEL
key: huggingface-${{ hashFiles('src/instructlab/configuration.py') }}
- name: Install dependencies
run: |
python3.11 -m venv venv
. venv/bin/activate
python3.11 -m pip install .
python3.11 -m pip install pytest
- name: Install instructlab and instructlab-eval
- name: Start inference server
run: |
export PATH="/home/runner/.local/bin:/usr/local/cuda/bin:$PATH"
python3 -m venv venv
. venv/bin/activate
cd instructlab
python3.11 -m venv cli_venv
. cli_venv/bin/activate
sed 's/\[.*\]//' requirements.txt > constraints.txt
python3 -m pip cache remove llama_cpp_python
python3.11 -m pip cache remove llama_cpp_python
CMAKE_ARGS="-DLLAMA_CUBLAS=on" python3 -m pip install --no-binary llama_cpp_python -c constraints.txt llama_cpp_python
# needed for --4-bit-quant option to ilab train
python3 -m pip install bitsandbytes
python3.11 -m pip install bitsandbytes
# install instructlab
python3 -m pip install .
cd ..
# Install instructlab-eval
python3 -m pip install .
python3.11 -m pip install .
# start llama-cpp server
ilab model download --repository instructlab/granite-7b-lab-GGUF --filename granite-7b-lab-Q4_K_M.gguf
ilab model serve --model-path models/granite-7b-lab-Q4_K_M.gguf
- name: Run e2e test
- name: Run unit tests
run: |
export INSTRUCTLAB_EVAL_FIRST_N_QUESTIONS=5
. venv/bin/activate
./instructlab/scripts/basic-workflow-tests.sh -cm
python3.11 -m pytest
- name: Remove llama-cpp-python from cache
if: always()
Expand Down
1 change: 1 addition & 0 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,7 @@ coverage.xml
.hypothesis/
.pytest_cache/
cover/
eval_output/

# Translations
*.mo
Expand Down
1 change: 1 addition & 0 deletions README.md
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
# eval

![Lint](https://github.com/instructlab/eval/actions/workflows/lint.yml/badge.svg?branch=main)
![Test](https://github.com/instructlab/eval/actions/workflows/test.yml/badge.svg?branch=main)
![Build](https://github.com/instructlab/eval/actions/workflows/pypi.yaml/badge.svg?branch=main)
![Release](https://img.shields.io/github/v/release/instructlab/eval)
![License](https://img.shields.io/github/license/instructlab/eval)
Expand Down
16 changes: 9 additions & 7 deletions tests/test_branch_gen_answers.py
Original file line number Diff line number Diff line change
@@ -1,10 +1,12 @@
# First Party
from instructlab.eval.mt_bench import MTBenchBranchEvaluator

mt_bench_branch = MTBenchBranchEvaluator(
"instructlab/granite-7b-lab",
"instructlab/granite-7b-lab",
"../taxonomy",
"main",
)
mt_bench_branch.gen_answers("http://localhost:8000/v1")

def test_branch_gen_answers():
mt_bench_branch = MTBenchBranchEvaluator(
"instructlab/granite-7b-lab",
"instructlab/granite-7b-lab",
"taxonomy",
"main",
)
mt_bench_branch.gen_answers("http://localhost:8000/v1")
Loading

0 comments on commit ab0af26

Please sign in to comment.