[CI][Benchmark] Optimize performance benchmark workflow

.github/workflows/nightly_benchmarks.yaml

@@ -15,20 +15,22 @@
 # limitations under the License.
 #
 
-name: 'run benchmarks main'
+name: 'Benchmarks / Performance'
+# This workflow runs nightly benchmarks for vllm-ascend.
 
 on:
   schedule:
+    # Run at 24:00 everyday
     - cron: '00 16 * * *'
   workflow_dispatch:
-
-  # after merged, secrets will be available
-  # pull_request:
-  #   branches:
-  #       - 'main'
-  #       - '*-dev'
-  #   paths:
-  #       - '.github/workflows/nightly_benchmarks.yaml'
+  pull_request:
+    branches:
+        - 'main'
+        - '*-dev'
+    paths:
+        - '.github/workflows/nightly_benchmarks.yaml'
+  pull_request_target:
+    types: [labeled]
 
 
 # Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
@@ -38,9 +40,13 @@ defaults:
   run:
     shell: bash -el {0}
 
+concurrency:
+  group: pr-${{ github.event.pull_request.number }}
+  cancel-in-progress: true
+
 jobs:
   test:
-    name: run benchmarks main
+    name: Benchmarks/vLLM=${{ matrix.vllm_branch }}, vLLM-Ascend=${{ matrix.vllm_ascend_branch }}
     runs-on: 'linux-arm64-npu-static-8'
     strategy:
       matrix:
@@ -64,6 +70,7 @@ jobs:
       env:
         HF_ENDPOINT: https://hf-mirror.com
         HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        HF_HOME: /github/home/.cache/huggingface
         ES_OM_DOMAIN: ${{ secrets.ES_OM_DOMAIN }}
         ES_OM_AUTHORIZATION: ${{ secrets.ES_OM_AUTHORIZATION }}
     steps:
@@ -90,7 +97,8 @@ jobs:
       - name: Checkout vllm-project/vllm-ascend repo
         uses: actions/checkout@v4
         with:
-          ref: ${{ matrix.vllm_ascend_branch }}
+          ref: dev-bench
+          repository: Potabk/vllm-ascend
 
       - name: Checkout vllm-project/vllm repo
         uses: actions/checkout@v4
@@ -109,25 +117,43 @@ jobs:
           pip install -e .
           pip install -r benchmarks/requirements-bench.txt
 
-      - name: Checkout cosdt/elastic-tool
-        uses: actions/checkout@v4
+      - name: Run current commit benchmarks
+        if: github.event_name != 'schedule'
+        run: |
+          # Sometimes we only want to run benchmarks on the current commit
+          # This is useful for debugging or a release benchmark
+          bash benchmarks/scripts/run-performance-benchmarks.sh
+          # Convert the benchmark results to markdown format
+          python3 benchmarks/scripts/convert_json_to_markdown.py
+
+      - name: Generate step summary
+        if: github.event_name != 'schedule'
+        run: |
+          cat ./benchmarks/results/benchmark_results.md >> $GITHUB_STEP_SUMMARY
+
+      - name: Upload benchmark artifacts
+        if: github.event_name != 'schedule'
+        uses: actions/upload-artifact@v4
         with:
-          repository: cosdt/elastic-tool
-          path: ./elastic_tool
-          ref: 0.1.0-dev
+          name: "benchmark-performance-${{ matrix.vllm_branch }}-${{ matrix.vllm_ascend_branch }}-report"
+          path: ./benchmarks/results/benchmark_results.md
+          if-no-files-found: warn
+          retention-days: 90
+          overwrite: true
 
       - name: Install elastic_tool
-        working-directory: ./elastic_tool
         run: |
-          pip install -e .
+          pip install escli-tool==0.2.0
 
       - name: Collect pr info from vllm-project/vllm-ascend
+        if: github.event_name == 'schedule'
         run: |
           # Only get the pull request which may influences performance
           git log --pretty=format:"%H %s" -- '**/*.py' ':!docs/*' ':!tests/*' ':!examples/*' > commit_log.txt
           escli check commit_log.txt
 
       - name: Run benchmark iteration
+        if: github.event_name == 'schedule'
         run: |
           while IFS= read -r line || [[ -n "$line" ]]; do
             commit_id=${line%% *}

benchmarks/requirements-bench.txt

@@ -1,3 +1,5 @@
 pandas
 datasets
-modelscope
+modelscope
+libcst
+tabulate

benchmarks/scripts/convert_json_to_markdown.py

@@ -0,0 +1,183 @@
+import argparse
+import json
+import os
+from pathlib import Path
+
+import pandas as pd
+from tabulate import tabulate
+
+CUR_PATH = Path(__file__).parent.resolve()
+# latency results and the keys that will be printed into markdown
+latency_results = []
+latency_column_mapping = {
+    "test_name": "Test name",
+    "avg_latency": "Mean latency (ms)",
+    "P50": "Median latency (ms)",
+    "P99": "P99 latency (ms)",
+}
+
+# throughput tests and the keys that will be printed into markdown
+throughput_results = []
+throughput_results_column_mapping = {
+    "test_name": "Test name",
+    "num_requests": "Num of reqs",
+    "total_num_tokens": "Total num of tokens",
+    "elapsed_time": "Elapsed time (s)",
+    "requests_per_second": "Tput (req/s)",
+    "tokens_per_second": "Tput (tok/s)",
+}
+
+# serving results and the keys that will be printed into markdown
+serving_results = []
+serving_column_mapping = {
+    "test_name": "Test name",
+    "request_rate": "Request rate (req/s)",
+    "request_throughput": "Tput (req/s)",
+    "output_throughput": "Output Tput (tok/s)",
+    "median_ttft_ms": "TTFT (ms)",
+    "median_tpot_ms": "TPOT (ms)",
+    "median_itl_ms": "ITL (ms)",
+}
+
+
+def read_markdown(file):
+    if os.path.exists(file):
+        with open(file) as f:
+            return f.read() + "\n"
+    else:
+        return f"{file} not found.\n"
+
+
+def results_to_json(latency, throughput, serving):
+    return json.dumps({
+        'latency': latency.to_dict(),
+        'throughput': throughput.to_dict(),
+        'serving': serving.to_dict()
+    })
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Process the results of the benchmark tests.")
+    parser.add_argument(
+        "--results_folder",
+        type=str,
+        default="../results/",
+        help="The folder where the benchmark results are stored.")
+    parser.add_argument(
+        "--output_folder",
+        type=str,
+        default="../results/",
+        help="The folder where the benchmark results are stored.")
+    parser.add_argument("--markdown_template",
+                        type=str,
+                        default="./perf_result_template.md",
+                        help="The template file for the markdown report.")
+    parser.add_argument("--tag",
+                        default="main",
+                        help="Tag to be used for release message.")
+    parser.add_argument("--commit_id",
+                        default="",
+                        help="Commit ID to be used for release message.")
+
+    args = parser.parse_args()
+    results_folder = (CUR_PATH / args.results_folder).resolve()
+    output_folder = (CUR_PATH / args.output_folder).resolve()
+    markdown_template = (CUR_PATH / args.markdown_template).resolve()
+
+    # collect results
+    for test_file in results_folder.glob("*.json"):
+
+        with open(test_file) as f:
+            raw_result = json.loads(f.read())
+
+        if "serving" in str(test_file):
+            # this result is generated via `benchmark_serving.py`
+
+            # update the test name of this result
+            raw_result.update({"test_name": test_file.stem})
+
+            # add the result to raw_result
+            serving_results.append(raw_result)
+            continue
+
+        elif "latency" in f.name:
+            # this result is generated via `benchmark_latency.py`
+
+            # update the test name of this result
+            raw_result.update({"test_name": test_file.stem})
+
+            # get different percentiles
+            for perc in [10, 25, 50, 75, 90, 99]:
+                # Multiply 1000 to convert the time unit from s to ms
+                raw_result.update(
+                    {f"P{perc}": 1000 * raw_result["percentiles"][str(perc)]})
+            raw_result["avg_latency"] = raw_result["avg_latency"] * 1000
+
+            # add the result to raw_result
+            latency_results.append(raw_result)
+            continue
+
+        elif "throughput" in f.name:
+            # this result is generated via `benchmark_throughput.py`
+
+            # update the test name of this result
+            raw_result.update({"test_name": test_file.stem})
+
+            # add the result to raw_result
+            throughput_results.append(raw_result)
+            continue
+
+        print(f"Skipping {test_file}")
+    serving_results.sort(key=lambda x: (len(x['test_name']), x['test_name']))
+
+    latency_results = pd.DataFrame.from_dict(latency_results)
+    serving_results = pd.DataFrame.from_dict(serving_results)
+    throughput_results = pd.DataFrame.from_dict(throughput_results)
+
+    raw_results_json = results_to_json(latency_results, throughput_results,
+                                       serving_results)
+
+    # remapping the key, for visualization purpose
+    if not latency_results.empty:
+        latency_results = latency_results[list(
+            latency_column_mapping.keys())].rename(
+                columns=latency_column_mapping)
+    if not serving_results.empty:
+        serving_results = serving_results[list(
+            serving_column_mapping.keys())].rename(
+                columns=serving_column_mapping)
+    if not throughput_results.empty:
+        throughput_results = throughput_results[list(
+            throughput_results_column_mapping.keys())].rename(
+                columns=throughput_results_column_mapping)
+
+    processed_results_json = results_to_json(latency_results,
+                                             throughput_results,
+                                             serving_results)
+
+    # get markdown tables
+    latency_md_table = tabulate(latency_results,
+                                headers='keys',
+                                tablefmt='pipe',
+                                showindex=False)
+    serving_md_table = tabulate(serving_results,
+                                headers='keys',
+                                tablefmt='pipe',
+                                showindex=False)
+    throughput_md_table = tabulate(throughput_results,
+                                   headers='keys',
+                                   tablefmt='pipe',
+                                   showindex=False)
+
+    # document the result
+    print(output_folder)
+    with open(output_folder / "benchmark_results.md", "w") as f:
+
+        results = read_markdown(markdown_template)
+        results = results.format(
+            latency_tests_markdown_table=latency_md_table,
+            throughput_tests_markdown_table=throughput_md_table,
+            serving_tests_markdown_table=serving_md_table,
+            benchmarking_results_in_json_string=processed_results_json)
+        f.write(results)

benchmarks/scripts/patch_benchmark_dataset.py

@@ -0,0 +1,67 @@
+from argparse import ArgumentParser
+
+import libcst as cst
+import libcst.matchers as m
+
+# Patch the benchmark_dataset.py file to set streaming=False in load_dataset calls
+
+
+class StreamingFalseTransformer(cst.CSTTransformer):
+
+    def __init__(self):
+        self.in_target_class = False
+        self.in_target_func = False
+
+    def visit_ClassDef(self, node):
+        if node.name.value == "HuggingFaceDataset":
+            self.in_target_class = True
+
+    def leave_ClassDef(self, original_node, updated_node):
+        self.in_target_class = False
+        return updated_node
+
+    def visit_FunctionDef(self, node):
+        if self.in_target_class and node.name.value == "load_data":
+            self.in_target_func = True
+
+    def leave_FunctionDef(self, original_node, updated_node):
+        self.in_target_func = False
+        return updated_node
+
+    def leave_Call(self, original_node, updated_node):
+        if self.in_target_class and self.in_target_func:
+            if m.matches(updated_node.func, m.Name("load_dataset")):
+                new_args = []
+                for arg in updated_node.args:
+                    if arg.keyword and arg.keyword.value == "streaming":
+                        new_arg = arg.with_changes(value=cst.Name("False"))
+                        new_args.append(new_arg)
+                    else:
+                        new_args.append(arg)
+                return updated_node.with_changes(args=new_args)
+        return updated_node
+
+
+def patch_file(path):
+    with open(path, "r", encoding="utf-8") as f:
+        source = f.read()
+
+    module = cst.parse_module(source)
+    modified = module.visit(StreamingFalseTransformer())
+
+    with open(path, "w", encoding="utf-8") as f:
+        f.write(modified.code)
+
+    print(f"Patched: {path}")
+
+
+if __name__ == '__main__':
+    parser = ArgumentParser(
+        description=
+        "Patch benchmark_dataset.py to set streaming=False in load_dataset calls"
+    )
+    parser.add_argument("--path",
+                        type=str,
+                        help="Path to the benchmark_dataset.py file")
+    args = parser.parse_args()
+    patch_file(args.path)