[Frontend] add run batch to CLI (#18804)

Signed-off-by: reidliu41 <reid201711@gmail.com>
Co-authored-by: reidliu41 <reid201711@gmail.com>

examples/offline_inference/openai_batch/README.md

@@ -48,7 +48,19 @@ The batch running tool is designed to be used from the command line.
 You can run the batch with the following command, which will write its results to a file called `results.jsonl`
 
 ```console
-python -m vllm.entrypoints.openai.run_batch -i offline_inference/openai_batch/openai_example_batch.jsonl -o results.jsonl --model meta-llama/Meta-Llama-3-8B-Instruct
+python -m vllm.entrypoints.openai.run_batch \
+    -i offline_inference/openai_batch/openai_example_batch.jsonl \
+    -o results.jsonl \
+    --model meta-llama/Meta-Llama-3-8B-Instruct
+```
+
+or use command-line:
+
+```console
+vllm run-batch \
+    -i offline_inference/openai_batch/openai_example_batch.jsonl \
+    -o results.jsonl \
+    --model meta-llama/Meta-Llama-3-8B-Instruct
 ```
 
 ### Step 3: Check your results
@@ -68,7 +80,19 @@ The batch runner supports remote input and output urls that are accessible via h
 For example, to run against our example input file located at `https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl`, you can run
 
 ```console
-python -m vllm.entrypoints.openai.run_batch -i https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl -o results.jsonl --model meta-llama/Meta-Llama-3-8B-Instruct
+python -m vllm.entrypoints.openai.run_batch \
+    -i https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl \
+    -o results.jsonl \
+    --model meta-llama/Meta-Llama-3-8B-Instruct
+```
+
+or use command-line:
+
+```console
+vllm run-batch \
+    -i https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl \
+    -o results.jsonl \
+    --model meta-llama/Meta-Llama-3-8B-Instruct
 ```
 
 ## Example 3: Integrating with AWS S3
@@ -164,6 +188,15 @@ python -m vllm.entrypoints.openai.run_batch \
     --model --model meta-llama/Meta-Llama-3-8B-Instruct
 ```
 
+or use command-line:
+
+```console
+vllm run-batch \
+    -i "https://s3.us-west-2.amazonaws.com/MY_BUCKET/MY_INPUT_FILE.jsonl?AWSAccessKeyId=ABCDEFGHIJKLMNOPQRST&Signature=abcdefghijklmnopqrstuvwxyz12345&Expires=1715800091" \
+    -o "https://s3.us-west-2.amazonaws.com/MY_BUCKET/MY_OUTPUT_FILE.jsonl?AWSAccessKeyId=ABCDEFGHIJKLMNOPQRST&Signature=abcdefghijklmnopqrstuvwxyz12345&Expires=1715800091" \
+    --model --model meta-llama/Meta-Llama-3-8B-Instruct
+```
+
 ### Step 4: View your results
 
 Your results are now on S3. You can view them in your terminal by running

tests/entrypoints/openai/test_run_batch.py

@@ -2,7 +2,6 @@
 
 import json
 import subprocess
-import sys
 import tempfile
 
 from vllm.entrypoints.openai.protocol import BatchRequestOutput
@@ -35,9 +34,8 @@ def test_empty_file():
         input_file.write("")
         input_file.flush()
         proc = subprocess.Popen([
-            sys.executable, "-m", "vllm.entrypoints.openai.run_batch", "-i",
+            "vllm", "run-batch", "-i", input_file.name, "-o", output_file.name,
-            input_file.name, "-o", output_file.name, "--model",
+            "--model", "intfloat/multilingual-e5-small"
-            "intfloat/multilingual-e5-small"
         ], )
         proc.communicate()
         proc.wait()
@@ -54,9 +52,8 @@ def test_completions():
         input_file.write(INPUT_BATCH)
         input_file.flush()
         proc = subprocess.Popen([
-            sys.executable, "-m", "vllm.entrypoints.openai.run_batch", "-i",
+            "vllm", "run-batch", "-i", input_file.name, "-o", output_file.name,
-            input_file.name, "-o", output_file.name, "--model",
+            "--model", "NousResearch/Meta-Llama-3-8B-Instruct"
-            "NousResearch/Meta-Llama-3-8B-Instruct"
         ], )
         proc.communicate()
         proc.wait()
@@ -79,9 +76,8 @@ def test_completions_invalid_input():
         input_file.write(INVALID_INPUT_BATCH)
         input_file.flush()
         proc = subprocess.Popen([
-            sys.executable, "-m", "vllm.entrypoints.openai.run_batch", "-i",
+            "vllm", "run-batch", "-i", input_file.name, "-o", output_file.name,
-            input_file.name, "-o", output_file.name, "--model",
+            "--model", "NousResearch/Meta-Llama-3-8B-Instruct"
-            "NousResearch/Meta-Llama-3-8B-Instruct"
         ], )
         proc.communicate()
         proc.wait()
@@ -95,9 +91,8 @@ def test_embeddings():
         input_file.write(INPUT_EMBEDDING_BATCH)
         input_file.flush()
         proc = subprocess.Popen([
-            sys.executable, "-m", "vllm.entrypoints.openai.run_batch", "-i",
+            "vllm", "run-batch", "-i", input_file.name, "-o", output_file.name,
-            input_file.name, "-o", output_file.name, "--model",
+            "--model", "intfloat/multilingual-e5-small"
-            "intfloat/multilingual-e5-small"
         ], )
         proc.communicate()
         proc.wait()
@@ -117,9 +112,8 @@ def test_score():
         input_file.write(INPUT_SCORE_BATCH)
         input_file.flush()
         proc = subprocess.Popen([
-            sys.executable,
+            "vllm",
-            "-m",
+            "run-batch",
-            "vllm.entrypoints.openai.run_batch",
             "-i",
             input_file.name,
             "-o",

vllm/entrypoints/cli/main.py

@@ -7,6 +7,7 @@ import sys
 import vllm.entrypoints.cli.benchmark.main
 import vllm.entrypoints.cli.collect_env
 import vllm.entrypoints.cli.openai
+import vllm.entrypoints.cli.run_batch
 import vllm.entrypoints.cli.serve
 import vllm.version
 from vllm.entrypoints.utils import VLLM_SERVE_PARSER_EPILOG, cli_env_setup
@@ -17,6 +18,7 @@ CMD_MODULES = [
     vllm.entrypoints.cli.serve,
     vllm.entrypoints.cli.benchmark.main,
     vllm.entrypoints.cli.collect_env,
+    vllm.entrypoints.cli.run_batch,
 ]
 
 

vllm/entrypoints/cli/run_batch.py

@@ -0,0 +1,55 @@
+# SPDX-License-Identifier: Apache-2.0
+
+import argparse
+import asyncio
+
+from prometheus_client import start_http_server
+
+from vllm.entrypoints.cli.types import CLISubcommand
+from vllm.entrypoints.logger import logger
+from vllm.entrypoints.openai.run_batch import main as run_batch_main
+from vllm.entrypoints.openai.run_batch import make_arg_parser
+from vllm.utils import FlexibleArgumentParser
+from vllm.version import __version__ as VLLM_VERSION
+
+
+class RunBatchSubcommand(CLISubcommand):
+    """The `run-batch` subcommand for vLLM CLI."""
+
+    def __init__(self):
+        self.name = "run-batch"
+        super().__init__()
+
+    @staticmethod
+    def cmd(args: argparse.Namespace) -> None:
+        logger.info("vLLM batch processing API version %s", VLLM_VERSION)
+        logger.info("args: %s", args)
+
+        # Start the Prometheus metrics server.
+        # LLMEngine uses the Prometheus client
+        # to publish metrics at the /metrics endpoint.
+        if args.enable_metrics:
+            logger.info("Prometheus metrics enabled")
+            start_http_server(port=args.port, addr=args.url)
+        else:
+            logger.info("Prometheus metrics disabled")
+
+        asyncio.run(run_batch_main(args))
+
+    def subparser_init(
+            self,
+            subparsers: argparse._SubParsersAction) -> FlexibleArgumentParser:
+        run_batch_parser = subparsers.add_parser(
+            "run-batch",
+            help="Run batch prompts and write results to file.",
+            description=(
+                "Run batch prompts using vLLM's OpenAI-compatible API.\n"
+                "Supports local or HTTP input/output files."),
+            usage=
+            "vllm run-batch -i INPUT.jsonl -o OUTPUT.jsonl --model <model>",
+        )
+        return make_arg_parser(run_batch_parser)
+
+
+def cmd_init() -> list[CLISubcommand]:
+    return [RunBatchSubcommand()]

vllm/entrypoints/openai/run_batch.py

@@ -33,9 +33,7 @@ from vllm.utils import FlexibleArgumentParser, random_uuid
 from vllm.version import __version__ as VLLM_VERSION
 
 
-def parse_args():
+def make_arg_parser(parser: FlexibleArgumentParser):
-    parser = FlexibleArgumentParser(
-        description="vLLM OpenAI-Compatible batch runner.")
     parser.add_argument(
         "-i",
         "--input-file",
@@ -98,7 +96,13 @@ def parse_args():
         default=False,
         help="If set to True, enable prompt_tokens_details in usage.")
 
-    return parser.parse_args()
+    return parser
+
+
+def parse_args():
+    parser = FlexibleArgumentParser(
+        description="vLLM OpenAI-Compatible batch runner.")
+    return make_arg_parser(parser).parse_args()
 
 
 # explicitly use pure text format, with a newline at the end