[Frontend] Add Early Validation For Chat Template / Tool Call Parser (#9151)

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>
This commit is contained in:
Alex Brooks
2024-10-08 14:31:26 +00:00
committed by GitHub
parent a3691b6b5e
commit 069d3bd8d0
5 changed files with 153 additions and 70 deletions
+15
View File
@@ -10,6 +10,7 @@ import ssl
from typing import List, Optional, Sequence, Union
from vllm.engine.arg_utils import AsyncEngineArgs, nullable_str
from vllm.entrypoints.chat_utils import validate_chat_template
from vllm.entrypoints.openai.serving_engine import (LoRAModulePath,
PromptAdapterPath)
from vllm.entrypoints.openai.tool_parsers import ToolParserManager
@@ -231,6 +232,20 @@ def make_arg_parser(parser: FlexibleArgumentParser) -> FlexibleArgumentParser:
return parser
def validate_parsed_serve_args(args: argparse.Namespace):
"""Quick checks for model serve args that raise prior to loading."""
if hasattr(args, "subparser") and args.subparser != "serve":
return
# Ensure that the chat template is valid; raises if it likely isn't
validate_chat_template(args.chat_template)
# Enable auto tool needs a tool call parser to be valid
if args.enable_auto_tool_choice and not args.tool_call_parser:
raise TypeError("Error: --enable-auto-tool-choice requires "
"--tool-call-parser")
def create_parser_for_docs() -> FlexibleArgumentParser:
parser_for_docs = FlexibleArgumentParser(
prog="-m vllm.entrypoints.openai.api_server")