50dfcaa8fa
ppl test and harness test are using the same update_in_parent_folder file. To reduce the repetition, change the ppl update HTML file to the same one under the harness folder and delete the HTML file under the ppl folder.
279 lines
10 KiB
YAML
279 lines
10 KiB
YAML
name: LLM Perplexity Evaluation
|
|
|
|
# Cancel previous runs in the PR when you push new commits
|
|
concurrency:
|
|
group: ${{ github.workflow }}-llm-nightly-test-${{ github.event.pull_request.number || github.run_id }}
|
|
cancel-in-progress: true
|
|
|
|
permissions:
|
|
contents: read
|
|
|
|
# Controls when the action will run.
|
|
on:
|
|
schedule:
|
|
- cron: "00 12 * * *" # GMT time, 12:00 GMT == 20:00 China
|
|
pull_request:
|
|
branches: [main]
|
|
paths:
|
|
- ".github/workflows/llm-ppl-evaluation.yml"
|
|
# Allows you to run this workflow manually from the Actions tab
|
|
workflow_dispatch:
|
|
inputs:
|
|
seq_len:
|
|
description: 'sequence length, input one number and must be quoted.'
|
|
required: true
|
|
type: string
|
|
model_name:
|
|
description: 'Model names, separated by comma and must be quoted.'
|
|
required: true
|
|
type: string
|
|
precision:
|
|
description: 'Precisions, separated by comma and must be quoted.'
|
|
required: true
|
|
type: string
|
|
runs-on:
|
|
description: 'Labels to filter the runners, separated by comma and must be quoted.'
|
|
default: "accuracy"
|
|
required: false
|
|
type: string
|
|
|
|
|
|
# A workflow run is made up of one or more jobs that can run sequentially or in parallel
|
|
jobs:
|
|
llm-cpp-build:
|
|
uses: ./.github/workflows/llm-binary-build.yml
|
|
with:
|
|
platform: 'Linux'
|
|
set-matrix:
|
|
runs-on: ubuntu-latest
|
|
outputs:
|
|
seq_len: ${{ steps.set-matrix.outputs.seq_len }}
|
|
model_name: ${{ steps.set-matrix.outputs.model_name }}
|
|
precision: ${{ steps.set-matrix.outputs.precision }}
|
|
runner: ${{ steps.set-matrix.outputs.runner }}
|
|
steps:
|
|
- name: set-nightly-env
|
|
if: ${{github.event_name == 'schedule'}}
|
|
env:
|
|
NIGHTLY_MATRIX_SEQ_LEN: '["512"]'
|
|
NIGHTLY_MATRIX_MODEL_NAME: '["Llama-2-7b-chat-hf", "mpt-7b-chat",
|
|
"falcon-7b-instruct-with-patch", "Mistral-7B-v0.1",
|
|
"chatglm2-6b", "chatglm3-6b", "Baichuan2-7B-Chat"]'
|
|
NIGHTLY_MATRIX_PRECISION: '["sym_int4", "fp8"]'
|
|
NIGHTLY_LABELS: '["self-hosted", "llm", "accuracy-nightly"]'
|
|
run: |
|
|
echo "seq_len=$NIGHTLY_MATRIX_SEQ_LEN" >> $GITHUB_ENV
|
|
echo "model_name=$NIGHTLY_MATRIX_MODEL_NAME" >> $GITHUB_ENV
|
|
echo "precision=$NIGHTLY_MATRIX_PRECISION" >> $GITHUB_ENV
|
|
echo "runner=$NIGHTLY_LABELS" >> $GITHUB_ENV
|
|
|
|
- name: set-pr-env
|
|
if: ${{github.event_name == 'pull_request'}}
|
|
env:
|
|
PR_MATRIX_SEQ_LEN: '["512"]'
|
|
PR_MATRIX_MODEL_NAME: '["Llama-2-7b-chat-hf", "chatglm3-6b"]'
|
|
PR_MATRIX_PRECISION: '["sym_int4"]'
|
|
PR_LABELS: '["self-hosted", "llm", "temp-arc01"]'
|
|
run: |
|
|
echo "seq_len=$PR_MATRIX_SEQ_LEN" >> $GITHUB_ENV
|
|
echo "model_name=$PR_MATRIX_MODEL_NAME" >> $GITHUB_ENV
|
|
echo "precision=$PR_MATRIX_PRECISION" >> $GITHUB_ENV
|
|
echo "runner=$PR_LABELS" >> $GITHUB_ENV
|
|
- name: set-manual-env
|
|
if: ${{github.event_name == 'workflow_dispatch'}}
|
|
env:
|
|
MANUAL_MATRIX_SEQ_LEN: ${{format('[ {0} ]', inputs.seq_len)}}
|
|
MANUAL_MATRIX_MODEL_NAME: ${{format('[ {0} ]', inputs.model_name)}}
|
|
MANUAL_MATRIX_PRECISION: ${{format('[ {0} ]', inputs.precision)}}
|
|
MANUAL_LABELS: ${{format('["self-hosted", "llm", {0}]', inputs.runs-on)}}
|
|
run: |
|
|
echo "seq_len=$MANUAL_MATRIX_SEQ_LEN" >> $GITHUB_ENV
|
|
echo "model_name=$MANUAL_MATRIX_MODEL_NAME" >> $GITHUB_ENV
|
|
echo "precision=$MANUAL_MATRIX_PRECISION" >> $GITHUB_ENV
|
|
echo "runner=$MANUAL_LABELS" >> $GITHUB_ENV
|
|
- name: set-matrix
|
|
id: set-matrix
|
|
run: |
|
|
echo "seq_len=$seq_len" >> $GITHUB_OUTPUT
|
|
echo "model_name=$model_name" >> $GITHUB_OUTPUT
|
|
echo "precision=$precision" >> $GITHUB_OUTPUT
|
|
echo "runner=$runner" >> $GITHUB_OUTPUT
|
|
llm-ppl-evaluation:
|
|
timeout-minutes: 1000
|
|
needs: [llm-cpp-build, set-matrix]
|
|
strategy:
|
|
fail-fast: false
|
|
matrix:
|
|
python-version: ["3.9"]
|
|
model_name: ${{ fromJson(needs.set-matrix.outputs.model_name) }}
|
|
precision: ${{ fromJson(needs.set-matrix.outputs.precision) }}
|
|
seq_len: ${{ fromJson(needs.set-matrix.outputs.seq_len) }}
|
|
device: [xpu]
|
|
|
|
runs-on: ${{ fromJson(needs.set-matrix.outputs.runner) }}
|
|
env:
|
|
ANALYTICS_ZOO_ROOT: ${{ github.workspace }}
|
|
ORIGIN_DIR: /mnt/disk1/models
|
|
DATASET_DIR: /mnt/disk1/datasets/THUDM___long_bench/
|
|
steps:
|
|
- uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # actions/checkout@v3
|
|
- name: Set up Python ${{ matrix.python-version }}
|
|
uses: actions/setup-python@v4
|
|
with:
|
|
python-version: ${{ matrix.python-version }}
|
|
- name: Install dependencies
|
|
shell: bash
|
|
run: |
|
|
python -m pip install --upgrade pip
|
|
python -m pip install --upgrade setuptools==58.0.4
|
|
python -m pip install --upgrade wheel
|
|
|
|
- name: Download llm binary
|
|
uses: ./.github/actions/llm/download-llm-binary
|
|
with:
|
|
platform: 'Linux'
|
|
|
|
- name: Run LLM install (all) test
|
|
uses: ./.github/actions/llm/setup-llm-env
|
|
with:
|
|
extra-dependency: "xpu_2.1"
|
|
|
|
- name: Download models
|
|
shell: bash
|
|
run: |
|
|
echo "MODEL_PATH=${ORIGIN_DIR}/${{ matrix.model_name }}/" >> "$GITHUB_ENV"
|
|
MODEL_PATH=${ORIGIN_DIR}/${{ matrix.model_name }}/
|
|
wget -r -nH -nc --no-verbose --cut-dirs=1 ${LLM_FTP_URL}/llm/${{ matrix.model_name }} -P ${ORIGIN_DIR}
|
|
|
|
- name: Upgrade packages
|
|
shell: bash
|
|
run: |
|
|
pip install --upgrade datasets==2.14.6
|
|
if [ "${{ matrix.model_name }}" = "Mistral-7B-v0.1" ]; then
|
|
pip install --upgrade transformers==4.36
|
|
else
|
|
pip install --upgrade transformers==4.31
|
|
fi
|
|
|
|
- name: Run perplexity
|
|
shell: bash
|
|
working-directory: ${{ github.workspace }}/python/llm/dev/benchmark/perplexity
|
|
env:
|
|
USE_XETLA: OFF
|
|
SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS: 1
|
|
run: |
|
|
source /opt/intel/oneapi/setvars.sh
|
|
if [[ "${{ matrix.model_name }}" == *"chatglm"* || "${{ matrix.model_name }}" == *"Baichuan"* ]]; then
|
|
LANGUAGE="zh"
|
|
else
|
|
LANGUAGE="en"
|
|
fi
|
|
python run.py \
|
|
--seq_len ${{ matrix.seq_len }} \
|
|
--model_path ${MODEL_PATH} \
|
|
--precisions ${{ matrix.precision }} \
|
|
--device ${{ matrix.device }} \
|
|
--dataset_path ${DATASET_DIR} \
|
|
--language ${LANGUAGE} \
|
|
--output_path results
|
|
|
|
- uses: actions/upload-artifact@v3
|
|
with:
|
|
name: ppl_results
|
|
path:
|
|
${{ github.workspace }}/python/llm/dev/benchmark/perplexity/results/**
|
|
|
|
- name: echo single result
|
|
shell: bash
|
|
|
|
working-directory: ${{ github.workspace }}/python/llm/dev/benchmark/perplexity/results/
|
|
run: |
|
|
if [[ "${{ matrix.model_name }}" == *"chatglm"* || "${{ matrix.model_name }}" == *"Baichuan"* ]]; then
|
|
LANGUAGE="zh"
|
|
else
|
|
LANGUAGE="en"
|
|
fi
|
|
cat ${{ matrix.model_name }}/${{ matrix.device }}/${{ matrix.precision }}/${LANGUAGE}/result.json
|
|
|
|
llm-ppl-summary:
|
|
if: ${{ always() }}
|
|
needs: llm-ppl-evaluation
|
|
runs-on: ubuntu-latest
|
|
steps:
|
|
- uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # actions/checkout@v3
|
|
- name: Set up Python 3.9
|
|
uses: actions/setup-python@v4
|
|
with:
|
|
python-version: 3.9
|
|
- name: Install dependencies
|
|
shell: bash
|
|
run: |
|
|
pip install --upgrade pip
|
|
pip install jsonlines pytablewriter regex
|
|
- name: Download all results
|
|
uses: actions/download-artifact@v3
|
|
with:
|
|
name: ppl_results
|
|
path: results
|
|
- name: Summarize the results
|
|
shell: bash
|
|
run: |
|
|
ls results
|
|
python ${{ github.workspace }}/python/llm/dev/benchmark/perplexity/make_table.py results
|
|
|
|
llm-ppl-html:
|
|
if: ${{github.event_name == 'schedule' || github.event_name == 'pull_request'}}
|
|
needs: [llm-ppl-evaluation]
|
|
runs-on: ["self-hosted", "llm", "accuracy1", "accuracy-nightly"]
|
|
steps:
|
|
- uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # actions/checkout@v3
|
|
- name: Set up Python 3.9
|
|
uses: actions/setup-python@v4
|
|
with:
|
|
python-version: 3.9
|
|
- name: Install dependencies
|
|
shell: bash
|
|
run: |
|
|
pip install --upgrade pip
|
|
pip install jsonlines pytablewriter regex
|
|
pip install pandas==1.5.3
|
|
|
|
- name: Set output path
|
|
shell: bash
|
|
run: |
|
|
echo "DATE=$(date +%Y-%m-%d)" >> $GITHUB_ENV
|
|
if ${{github.event_name == 'pull_request'}}; then
|
|
echo 'ACC_FOLDER=/home/arda/action-runners/ppl/pr-accuracy-data' >> $GITHUB_ENV
|
|
fi
|
|
if ${{github.event_name == 'schedule'}}; then
|
|
echo 'ACC_FOLDER=/home/arda/action-runners/ppl/nightly-accuracy-data' >> $GITHUB_ENV
|
|
fi
|
|
|
|
- name: Download ppl results
|
|
uses: actions/download-artifact@v3
|
|
with:
|
|
name: ppl_results
|
|
path: ${{ env.ACC_FOLDER}}/${{ env.DATE }}
|
|
|
|
# Save fp16.csv in the parent folder of env.nightly_folder
|
|
- name: Download fp16.results
|
|
shell: bash
|
|
run: |
|
|
wget https://raw.githubusercontent.com/intel-analytics/ipex-llm/main/python/llm/test/benchmark/perplexity/fp16.csv -O $ACC_FOLDER/../fp16.csv
|
|
ls $ACC_FOLDER/..
|
|
|
|
- name: Write to CSV
|
|
working-directory: ${{ github.workspace }}/python/llm/dev/benchmark/perplexity
|
|
shell: bash
|
|
run: |
|
|
ls $ACC_FOLDER/$DATE
|
|
python make_csv.py $ACC_FOLDER/$DATE $ACC_FOLDER
|
|
|
|
- name: Update HTML
|
|
working-directory: ${{ github.workspace }}/python/llm/test/benchmark/perplexity
|
|
shell: bash
|
|
run: |
|
|
python ppl_csv_to_html.py -f $ACC_FOLDER
|
|
if ${{github.event_name == 'schedule'}}; then
|
|
python ../harness/update_html_in_parent_folder.py -f $ACC_FOLDER
|
|
fi
|