Files
FastDeploy/fastdeploy/input/utils.py
T
gongweibao 1e49855b0f [BugFix][DataProcessor] Add validate_model_path to fail fast on bad model path or unreachable network (#6713)
* fix

* add more endpoint

* fix some

---------

Co-authored-by: gongweibao <gognweibao@baidu.com>
2026-03-08 12:36:32 +08:00

106 lines
3.7 KiB
Python

"""
# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
#
# 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.
"""
__all__ = [
"IDS_TYPE_FLAG",
"MAX_IMAGE_DIMENSION",
]
import os
import socket
from typing import Any, Callable, Dict, List, Tuple
from urllib.parse import urlparse
from fastdeploy.utils import console_logger
IDS_TYPE_FLAG = {"text": 0, "image": 1, "video": 2, "audio": 3}
MAX_IMAGE_DIMENSION = 9999999
# Hub endpoints for connectivity check, keyed by DOWNLOAD_SOURCE value
_HUB_ENDPOINTS = {
"huggingface": ("huggingface.co", 443),
"modelscope": ("modelscope.cn", 443),
}
def _get_hub_endpoint():
"""Return (host, port, hub_name) for the active download hub."""
source = os.environ.get("DOWNLOAD_SOURCE", "huggingface")
if source == "aistudio":
url = os.environ.get("AISTUDIO_ENDPOINT", "http://git.aistudio.baidu.com")
parsed = urlparse(url)
host = parsed.hostname or "git.aistudio.baidu.com"
port = parsed.port or (443 if parsed.scheme == "https" else 80)
return host, port, "aistudio"
host, port = _HUB_ENDPOINTS.get(source, ("huggingface.co", 443))
return host, port, source
def validate_model_path(model_name_or_path):
"""
Validate model path before from_pretrained calls.
Give immediate feedback instead of letting users wait 50s+ for timeout.
"""
if os.path.isdir(model_name_or_path) or os.path.isfile(model_name_or_path):
return # Local path exists, no network needed
host, port, hub_name = _get_hub_endpoint()
console_logger.warning(
f"Model path '{model_name_or_path}' is not a local directory or file, "
f"will try to download from {hub_name} hub."
)
# Quick connectivity check — fail fast instead of waiting 50s
try:
sock = socket.create_connection((host, port), timeout=3)
sock.close()
except OSError:
console_logger.warning(
f"Cannot reach {host}. If the model is stored locally, "
f"please check the path '{model_name_or_path}'. Otherwise check "
f"network/proxy settings (DOWNLOAD_SOURCE={hub_name})."
)
def process_stop_token_ids(
request: Dict[str, Any],
update_stop_seq_fn: Callable[[List[str]], Tuple[List[List[int]], List[int]]],
) -> None:
stop_token_ids_final = []
if request.get("stop_token_ids") is not None:
stop_token_ids = request.get("stop_token_ids")
if isinstance(stop_token_ids, list) and len(stop_token_ids) > 0:
if isinstance(stop_token_ids[0], int):
# List[int] -> List[List[int]]
stop_token_ids_final.extend([[t] for t in stop_token_ids])
elif isinstance(stop_token_ids[0], list):
# Already List[List[int]]
stop_token_ids_final.extend(stop_token_ids)
stop_sequences = request.get("stop", [])
if stop_sequences:
stop_seqs, _ = update_stop_seq_fn(stop_sequences)
stop_token_ids_final.extend(stop_seqs)
# Update request
if stop_token_ids_final:
stop_seqs_len = [len(seq) for seq in stop_token_ids_final]
request["stop_token_ids"] = stop_token_ids_final
request["stop_seqs_len"] = stop_seqs_len