feat: add benchmarks_entrypoint.py (#34495)

* feat: add `benchmarks_entrypoint.py`

Adding `benchmarks_entrypoint.py` file, which will be run from the
benchmarks CI.

This python script will list all python files from the `benchmark/`
folder and run the included `run_benchmark` function, allowing people to
add new benchmarks scripts.

* feat: add `MetricsRecorder`

* feat: update dashboard

* fix: add missing arguments to `MetricsRecorder`

* feat: update dash & add datasource + `default.yml`

* fix: move responsibility to create `MetricsRecorder` in bench script

* fix: update incorrect datasource UID

* fix: incorrect variable values

* debug: benchmark entrypoint script

* refactor: update log level

* fix: update broken import

* feat: add debug log in `MetricsRecorder`

* debug: set log level to debug

* fix: set connection `autocommit` to `True`

.github/workflows/benchmark.yml

@@ -63,7 +63,7 @@ jobs:
             commit_id=$GITHUB_SHA
           fi
           commit_msg=$(git show -s --format=%s | cut -c1-70)
-          python3 benchmark/llama.py "${{ github.head_ref || github.ref_name }}" "$commit_id" "$commit_msg"
+          python3 benchmark/benchmarks_entrypoint.py "${{ github.head_ref || github.ref_name }}" "$commit_id" "$commit_msg"
         env:
           HF_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
           # Enable this to see debug logs

benchmark/README.md

@@ -0,0 +1,49 @@
+# Benchmarks
+
+You might want to add new benchmarks.
+
+You will need to define a python function named `run_benchmark` in your python file and the file must be located in this `benchmark/` directory.
+
+The expected function signature is the following:
+
+```py
+def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str, num_tokens_to_generate=100):
+```
+
+## Writing metrics to the database
+
+`MetricRecorder` is thread-safe, in the sense of the python [`Thread`](https://docs.python.org/3/library/threading.html#threading.Thread). This means you can start a background thread to do the readings on the device measurements while not blocking the main thread to execute the model measurements.
+
+cf [`llama.py`](./llama.py) to see an example of this in practice.
+
+```py
+from benchmarks_entrypoint import MetricsRecorder
+import psycopg2
+
+def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str, num_tokens_to_generate=100):
+  metrics_recorder = MetricsRecorder(psycopg2.connect("dbname=metrics"), logger, branch, commit_id, commit_msg)
+  benchmark_id = metrics_recorder.initialise_benchmark({"gpu_name": gpu_name, "model_id": model_id})
+    # To collect device measurements
+    metrics_recorder.collect_device_measurements(
+        benchmark_id, cpu_util, mem_megabytes, gpu_util, gpu_mem_megabytes
+    )
+    # To collect your model measurements
+    metrics_recorder.collect_model_measurements(
+        benchmark_id,
+        {
+            "model_load_time": model_load_time,
+            "first_eager_forward_pass_time_secs": first_eager_fwd_pass_time,
+            "second_eager_forward_pass_time_secs": second_eager_fwd_pass_time,
+            "first_eager_generate_time_secs": first_eager_generate_time,
+            "second_eager_generate_time_secs": second_eager_generate_time,
+            "time_to_first_token_secs": time_to_first_token,
+            "time_to_second_token_secs": time_to_second_token,
+            "time_to_third_token_secs": time_to_third_token,
+            "time_to_next_token_mean_secs": mean_time_to_next_token,
+            "first_compile_generate_time_secs": first_compile_generate_time,
+            "second_compile_generate_time_secs": second_compile_generate_time,
+            "third_compile_generate_time_secs": third_compile_generate_time,
+            "fourth_compile_generate_time_secs": fourth_compile_generate_time,
+        },
+    )
+```

benchmark/benchmarks_entrypoint.py

@@ -0,0 +1,144 @@
+import argparse
+import importlib.util
+import logging
+import os
+from typing import Dict
+import psycopg2
+import sys
+
+from psycopg2.extras import Json
+from psycopg2.extensions import register_adapter
+
+
+register_adapter(dict, Json)
+
+
+class ImportModuleException(Exception):
+    pass
+
+
+class MetricsRecorder:
+    def __init__(self, connection, logger: logging.Logger, branch: str, commit_id: str, commit_msg: str):
+        self.conn = connection
+        self.conn.autocommit = True
+        self.logger = logger
+        self.branch = branch
+        self.commit_id = commit_id
+        self.commit_msg = commit_msg
+
+    def initialise_benchmark(self, metadata: Dict[str, str]) -> int:
+        """
+        Creates a new benchmark, returns the benchmark id
+        """
+        # gpu_name: str, model_id: str
+        with self.conn.cursor() as cur:
+            cur.execute(
+                "INSERT INTO benchmarks (branch, commit_id, commit_message, metadata) VALUES (%s, %s, %s, %s) RETURNING benchmark_id",
+                (self.branch, self.commit_id, self.commit_msg, metadata),
+            )
+            benchmark_id = cur.fetchone()[0]
+            logger.debug(f"initialised benchmark #{benchmark_id}")
+            return benchmark_id
+
+    def collect_device_measurements(self, benchmark_id: int, cpu_util, mem_megabytes, gpu_util, gpu_mem_megabytes):
+        """
+        Collect device metrics, such as CPU & GPU usage. These are "static", as in you cannot pass arbitrary arguments to the function.
+        """
+        with self.conn.cursor() as cur:
+            cur.execute(
+                "INSERT INTO device_measurements (benchmark_id, cpu_util, mem_megabytes, gpu_util, gpu_mem_megabytes) VALUES (%s, %s, %s, %s, %s)",
+                (benchmark_id, cpu_util, mem_megabytes, gpu_util, gpu_mem_megabytes),
+            )
+        self.logger.debug(
+            f"inserted device measurements for benchmark #{benchmark_id} [CPU util: {cpu_util}, mem MBs: {mem_megabytes}, GPU util: {gpu_util}, GPU mem MBs: {gpu_mem_megabytes}]"
+        )
+
+    def collect_model_measurements(self, benchmark_id: int, measurements: Dict[str, float]):
+        with self.conn.cursor() as cur:
+            cur.execute(
+                """
+                INSERT INTO model_measurements (
+                    benchmark_id,
+                    measurements
+                ) VALUES (%s, %s)
+                """,
+                (
+                    benchmark_id,
+                    measurements,
+                ),
+            )
+        self.logger.debug(f"inserted model measurements for benchmark #{benchmark_id}: {measurements}")
+
+    def close(self):
+        self.conn.close()
+
+
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.INFO)
+
+handler = logging.StreamHandler(sys.stdout)
+handler.setLevel(logging.INFO)
+formatter = logging.Formatter("[%(levelname)s - %(asctime)s] %(message)s")
+handler.setFormatter(formatter)
+logger.addHandler(handler)
+
+
+def parse_arguments():
+    """
+    Parse command line arguments for the benchmarking CLI.
+    """
+    parser = argparse.ArgumentParser(description="CLI for benchmarking the huggingface/transformers.")
+
+    parser.add_argument(
+        "branch",
+        type=str,
+        help="The branch name on which the benchmarking is performed.",
+    )
+
+    parser.add_argument(
+        "commit_id",
+        type=str,
+        help="The commit hash on which the benchmarking is performed.",
+    )
+
+    parser.add_argument(
+        "commit_msg",
+        type=str,
+        help="The commit message associated with the commit, truncated to 70 characters.",
+    )
+
+    args = parser.parse_args()
+
+    return args.branch, args.commit_id, args.commit_msg
+
+
+def import_from_path(module_name, file_path):
+    try:
+        spec = importlib.util.spec_from_file_location(module_name, file_path)
+        module = importlib.util.module_from_spec(spec)
+        sys.modules[module_name] = module
+        spec.loader.exec_module(module)
+        return module
+    except Exception as e:
+        raise ImportModuleException(f"failed to load python module: {e}")
+
+
+if __name__ == "__main__":
+    benchmarks_folder_path = os.path.dirname(os.path.realpath(__file__))
+
+    branch, commit_id, commit_msg = parse_arguments()
+
+    for entry in os.scandir(benchmarks_folder_path):
+        try:
+            if not entry.name.endswith(".py"):
+                continue
+            if entry.path == __file__:
+                continue
+            logger.debug(f"loading: {entry.name}")
+            module = import_from_path(entry.name.split(".")[0], entry.path)
+            logger.info(f"runnning benchmarks in: {entry.name}")
+            module.run_benchmark(logger, branch, commit_id, commit_msg)
+        except ImportModuleException as e:
+            logger.error(e)
+        except Exception as e:
+            logger.error(f"error running benchmarks for {entry.name}: {e}")

benchmark/default.yml

@@ -0,0 +1,10 @@
+apiVersion: 1
+
+providers:
+  - name: 'Transformers Benchmarks'
+    orgId: 1
+    type: file
+    updateIntervalSeconds: 10
+    allowUiUpdates: true
+    options:
+      path: /etc/grafana/dashboards