LLM: only run arc perf test nightly (#9448)

* LLM: only run arc perf test nightly

* deleted unused python scripts

* rebase main

.github/workflows/llm_performance_tests.yml

@@ -9,72 +9,23 @@ concurrency:
 on:
   schedule:
     - cron: "00 13 * * *" # GMT time, 13:00 GMT == 21:00 China
-  pull_request:
-    branches: [main]
-    paths:
-      - ".github/workflows/llm_performance_tests.yml"
-      - "python/llm/test/benchmark/**"
-      - "python/llm/dev/benchmark/all-in-one/**"
+  # pull_request:
+  #   branches: [main]
+  #   paths:
+  #     - ".github/workflows/llm_performance_tests.yml"
+  #     - "python/llm/test/benchmark/**"
+  #     - "python/llm/dev/benchmark/all-in-one/**"
   workflow_dispatch:
   workflow_call:
 
 # A workflow run is made up of one or more jobs that can run sequentially or in parallel
 jobs:
   llm-cpp-build:
+    if: ${{ github.event.schedule || github.event.inputs.artifact == 'llm-cpp-build' || github.event.inputs.artifact == 'all' }}
     uses: ./.github/workflows/llm-binary-build.yml
-  llm-performance-test:
-    if: false # skip cpu performance test for now; may add it back with separated runner
-    needs: llm-cpp-build
-    strategy:
-      fail-fast: false
-      matrix:
-        python-version: ["3.9"]
-        instruction: ["AVX512"]
-    runs-on: [self-hosted, llm, perf]
-    env:
-      THREAD_NUM: 24
-    steps:
-      - name: Set environment variables
-        shell: bash
-        run: |
-          echo "LLAMA2_7B_ORIGIN_PATH=${ORIGIN_DIR}/Llama-2-7b-chat-hf" >> "$GITHUB_ENV"
-      - uses: actions/checkout@v2
-      - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v2
-        with:
-          python-version: ${{ matrix.python-version }}
-      - name: Install dependencies
-        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
-
-      - name: Run LLM install (all) test
-        uses: ./.github/actions/llm/setup-llm-env
-        env:
-          ANALYTICS_ZOO_ROOT: ${{ github.workspace }}
-
-      - name: Download LLMs
-        shell: bash
-        run: |
-          if [ ! -d $LLAMA2_7B_ORIGIN_PATH ]; then
-            echo "Directory $LLAMA2_7B_ORIGIN_PATH not found. Downloading from FTP server..."
-            wget -r -nH --no-verbose --cut-dirs=1 $LLM_FTP_URL/llm/Llama-2-7b-chat-hf -P $ORIGIN_DIR
-          fi
-
-      - name: Run LLM Performance test
-        env:
-          ANALYTICS_ZOO_ROOT: ${{ github.workspace }}
-        run: bash python/llm/dev/benchmark/run-benchmark-tests.sh
-
-      # - name: Clean up test environment
-      #   uses: ./.github/actions/llm/remove-llm-env
-      #   env:
-      #     ANALYTICS_ZOO_ROOT: ${{ github.workspace }}
   llm-performance-test-on-arc:
+    if: ${{ github.event.schedule || github.event.inputs.artifact == 'llm-performance-test-on-arc' || github.event.inputs.artifact == 'all' }}
     needs: llm-cpp-build
     strategy:
       fail-fast: false
@@ -142,6 +93,7 @@ jobs:
           fi
           
   llm-performance-test-on-spr:
+    if: ${{ github.event.schedule || github.event.inputs.artifact == 'llm-performance-test-on-spr' || github.event.inputs.artifact == 'all' }}
     needs: llm-cpp-build
     strategy:
       fail-fast: false
@@ -190,6 +142,7 @@ jobs:
           python csv_to_html.py -f /mnt/disk1/nightly_perf_cpu/
 
   llm-performance-test-on-core:
+    if: ${{ github.event.schedule || github.event.inputs.artifact == 'llm-performance-test-on-core' || github.event.inputs.artifact == 'all' }}
     needs: llm-cpp-build
     strategy:
       fail-fast: false

python/llm/dev/benchmark/pipelines/llama2_test.py

@@ -1,94 +0,0 @@
-#
-# Copyright 2016 The BigDL Authors.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-
-
-# this code is copied from llama2 example test, and added performance test
-import torch
-import time
-import argparse
-
-from bigdl.llm.transformers import AutoModelForCausalLM
-from transformers import LlamaTokenizer
-
-
-import os
-benchmark_util_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), '..')
-import sys
-sys.path.append(benchmark_util_path)
-from benchmark_util import BenchmarkWrapper
-
-# you could tune the prompt based on your own model,
-# here the prompt tuning refers to https://huggingface.co/georgesung/llama2_7b_chat_uncensored#prompt-style
-LLAMA2_PROMPT_FORMAT = """### HUMAN:
-{prompt}
-
-### RESPONSE:
-"""
-
-if __name__ == '__main__':
-
-    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Llama2 model')
-    parser.add_argument('--repo-id-or-model-path', type=str, default="meta-llama/Llama-2-7b-chat-hf",
-                        help='The huggingface repo id for the Llama2 (e.g. `meta-llama/Llama-2-7b-chat-hf` and `meta-llama/Llama-2-13b-chat-hf`) to be downloaded'
-                                ', or the path to the huggingface checkpoint folder')
-    parser.add_argument('--prompt', type=str, default="What is AI?",
-                        help='Prompt to infer')
-    parser.add_argument('--n-predict', type=int, default=32,
-                        help='Max tokens to predict')
-
-    args = parser.parse_args()
-    model_path = args.repo_id_or_model_path
-
-    # Load model in 4 bit,
-    # which convert the relevant layers in the model into INT4 format
-    model = AutoModelForCausalLM.from_pretrained(model_path,
-                                                    load_in_4bit=True,
-                                                    trust_remote_code=True)
-
-
-    model = BenchmarkWrapper(model, do_print=False)
-
-    # Load tokenizer
-    tokenizer = LlamaTokenizer.from_pretrained(model_path, trust_remote_code=True)
-
-    # Generate predicted tokens
-    with torch.inference_mode():
-        prompt = LLAMA2_PROMPT_FORMAT.format(prompt=args.prompt)
-        input_ids = tokenizer.encode(prompt, return_tensors="pt")
-        st = time.time()
-        # if your selected model is capable of utilizing previous key/value attentions
-        # to enhance decoding speed, but has `"use_cache": false` in its model config,
-        # it is important to set `use_cache=True` explicitly in the `generate` function
-        # to obtain optimal performance with BigDL-LLM INT4 optimizations
-        output = model.generate(input_ids,
-                                max_new_tokens=args.n_predict)
-        end = time.time()
-        output_str = tokenizer.decode(output[0], skip_special_tokens=True)
-        print(f'Inference time: {end-st} s')
-        print('-'*20, 'Prompt', '-'*20)
-        print(prompt)
-        print('-'*20, 'Output', '-'*20)
-        print(output_str)
-        
-        assert "AI is a term" in output_str, "output is not as expected, the correctness may be wrong."
-        llama2_baseline = os.getenv('LLAMA2_BASELINE')
-        if llama2_baseline is None:
-            print('baseline is not set, skipping baseline validation')
-        else:
-            llama2_baseline = float(llama2_baseline)
-            ratio = model.rest_cost_mean / llama2_baseline
-            assert ratio < 1.1, f"performance did not meet baseline, the cost is {(ratio - 1) * 100}% higher than the baseline"
-

python/llm/dev/benchmark/run-benchmark-tests.sh

@@ -1,17 +0,0 @@
-# Performance tests usually use dedicated machines, see below to set env vars, e.g. model paths
-# The following environment variables should be ready
-# ORIGINAL_LLAMA2_PATH
-# LLAMA2_BASELINE
-# LLM_DIR
-
-if [ -z "$THREAD_NUM" ]; then
-  THREAD_NUM=2
-fi
-export OMP_NUM_THREADS=$THREAD_NUM
-
-######## LLAMA2
-# transformers
-
-echo ">>> Testing LLAMA2 transformers API"
-taskset -c 0-$((THREAD_NUM - 1)) python python/llm/dev/benchmark/pipelines/llama2_test.py --repo-id-or-model-path $LLAMA2_7B_ORIGIN_PATH
-