feat: update nightly gsm8k eval (#1304)

This commit is contained in:
Yineng Zhang
2024-09-03 01:18:41 +10:00
committed by GitHub
parent 9999442756
commit 2561ed012c
3 changed files with 100 additions and 38 deletions

View File

@@ -15,9 +15,9 @@ concurrency:
cancel-in-progress: true
jobs:
meta-llama-31-8b-instruct:
nightly-eval-2-gpu:
if: github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request'
runs-on: 1-gpu-runner
runs-on: 2-gpu-runner
steps:
- name: Checkout code
uses: actions/checkout@v3
@@ -25,42 +25,11 @@ jobs:
- name: Install dependencies
run: |
pip install --upgrade pip
pip install -e "python[dev]"
pip install -e "python[all]"
pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.4/ --force-reinstall
git clone https://github.com/EleutherAI/lm-evaluation-harness
pushd lm-evaluation-harness
pip install -e .
pip install lm_eval[api]
popd
- name: Run eval
timeout-minutes: 20
- name: Nightly gsm8k Accuracy
timeout-minutes: 60
run: |
python3 -m sglang.launch_server --model meta-llama/Meta-Llama-3.1-8B-Instruct --disable-radix-cache &
echo "Waiting for server to start..."
for i in {1..120}; do
if curl -s http://127.0.0.1:30000/health; then
echo "Server is up!"
break
fi
if [ $i -eq 120 ]; then
echo "Server failed to start within 120 seconds"
exit 1
fi
sleep 1
done
lm_eval --model local-completions --tasks gsm8k --model_args model=meta-llama/Meta-Llama-3.1-8B-Instruct,base_url=http://127.0.0.1:30000/v1/completions,num_concurrent=128,max_retries=3,tokenized_requests=False
echo "Stopping server..."
kill -9 $(ps aux | grep sglang | grep Meta-Llama-3.1-8B-Instruct | grep -v grep | awk '{print $2}')
finish:
needs: [
meta-llama-31-8b-instruct
]
runs-on: ubuntu-latest
steps:
- name: Finish
run: echo "This is an empty step to ensure that all jobs are completed."
cd test/srt
python3 test_nightly_gsm8k_eval.py