diff --git a/funasr/datasets/audio_datasets/jsonl2scp.py b/funasr/datasets/audio_datasets/jsonl2scp.py index 5e732596bc..7053ecd9e3 100644 --- a/funasr/datasets/audio_datasets/jsonl2scp.py +++ b/funasr/datasets/audio_datasets/jsonl2scp.py @@ -1,3 +1,4 @@ +import ast import os import json import torch @@ -58,7 +59,8 @@ def main_hydra(cfg: DictConfig): ("/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"), ) if isinstance(scp_file_list, str): - scp_file_list = eval(scp_file_list) + # 只允许字面量(路径列表/元组),禁止 eval 执行任意代码 + scp_file_list = ast.literal_eval(scp_file_list) data_type_list = kwargs.get("data_type_list", ("source", "target")) jsonl_file = kwargs.get( "jsonl_file_in", "/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl" diff --git a/funasr/datasets/audio_datasets/scp2jsonl.py b/funasr/datasets/audio_datasets/scp2jsonl.py index 1fbb34e41a..8a50448d26 100644 --- a/funasr/datasets/audio_datasets/scp2jsonl.py +++ b/funasr/datasets/audio_datasets/scp2jsonl.py @@ -1,3 +1,4 @@ +import ast import os import json import torch @@ -127,7 +128,8 @@ def main_hydra(cfg: DictConfig): ("/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"), ) if isinstance(scp_file_list, str): - scp_file_list = eval(scp_file_list) + # 只允许字面量(路径列表/元组),禁止 eval 执行任意代码 + scp_file_list = ast.literal_eval(scp_file_list) data_type_list = kwargs.get("data_type_list", ("source", "target")) jsonl_file_out = kwargs.get( "jsonl_file_out", "/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl" diff --git a/funasr/datasets/audio_datasets/sensevoice2jsonl.py b/funasr/datasets/audio_datasets/sensevoice2jsonl.py index f9c40128e9..6f25432ccd 100644 --- a/funasr/datasets/audio_datasets/sensevoice2jsonl.py +++ b/funasr/datasets/audio_datasets/sensevoice2jsonl.py @@ -1,3 +1,4 @@ +import ast import os import json import torch @@ -193,7 +194,8 @@ def main_hydra(cfg: DictConfig): ("/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"), ) if isinstance(scp_file_list, str): - scp_file_list = eval(scp_file_list) + # 只允许字面量(路径列表/元组),禁止 eval 执行任意代码 + scp_file_list = ast.literal_eval(scp_file_list) data_type_list = kwargs.get("data_type_list", ("source", "target")) jsonl_file_out = kwargs.get( "jsonl_file_out", "/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl" diff --git a/funasr/download/file.py b/funasr/download/file.py index 0375f0c877..ebeacab2a1 100644 --- a/funasr/download/file.py +++ b/funasr/download/file.py @@ -23,10 +23,12 @@ def download_from_url(url): storage = HTTPStorage() # bytes data = storage.read(url) - work_dir = tempfile.TemporaryDirectory().name - if not os.path.exists(work_dir): - os.makedirs(work_dir) - file_path = os.path.join(work_dir, os.path.basename(url)) + # mkdtemp 返回真实存在且不会被 GC 回收的目录; + # 原 TemporaryDirectory().name 未持有引用,目录随即被析构删除 + work_dir = tempfile.mkdtemp(prefix="funasr_download_") + # 用 URL path 的 basename,避免查询串(?a=b)进入文件名(Windows 非法字符) + file_name = os.path.basename(result.path) or "downloaded_file" + file_path = os.path.join(work_dir, file_name) with open(file_path, "wb") as fb: fb.write(data) assert file_path is not None, f"failed to download: {url}" @@ -166,7 +168,8 @@ def read(self, url): Args: url: TODO. """ - r = requests.get(url) + # 超时:防止恶意/故障服务器让下载线程永久挂起 + r = requests.get(url, timeout=(10, 60)) r.raise_for_status() return r.content @@ -176,7 +179,7 @@ def read_text(self, url): Args: url: TODO. """ - r = requests.get(url) + r = requests.get(url, timeout=(10, 60)) r.raise_for_status() return r.text diff --git a/funasr/models/data2vec/data2vec_encoder.py b/funasr/models/data2vec/data2vec_encoder.py index 652b477d18..37de311492 100644 --- a/funasr/models/data2vec/data2vec_encoder.py +++ b/funasr/models/data2vec/data2vec_encoder.py @@ -3,6 +3,7 @@ # This source code is licensed under the MIT license found in the # LICENSE file in the root directory of this source tree. +import ast import logging import math @@ -21,6 +22,48 @@ from funasr.models.transformer.utils.nets_utils import make_pad_mask +def _parse_conv_feature_layers(expr): + """安全解析卷积层配置字符串,替代 eval()。 + + 仅允许整数字面量、list/tuple 以及它们之间的 + / * 运算, + 例如 "[(512,10,5)] + [(512,3,2)] * 4"。 + 配置可能来自下载的模型 config.yaml,eval() 会执行其中任意代码。 + + Args: + expr: 卷积层配置表达式字符串。 + + Returns: + 解析后的层配置列表。 + """ + if not isinstance(expr, str): + return expr + + def _eval_node(node): + if isinstance(node, ast.Expression): + return _eval_node(node.body) + if isinstance(node, ast.Constant) and type(node.value) is int: + return node.value + if isinstance(node, ast.UnaryOp) and isinstance(node.op, ast.USub): + operand = _eval_node(node.operand) + if type(operand) is int: + return -operand + if isinstance(node, ast.List): + return [_eval_node(e) for e in node.elts] + if isinstance(node, ast.Tuple): + return tuple(_eval_node(e) for e in node.elts) + if isinstance(node, ast.BinOp): + left, right = _eval_node(node.left), _eval_node(node.right) + if isinstance(node.op, ast.Add): + return left + right + if isinstance(node.op, ast.Mult): + return left * right + raise ValueError( + f"Unsupported conv_feature_layers expression: {expr!r}" + ) + + return _eval_node(ast.parse(expr, mode="eval")) + + def get_annealed_rate(start, end, curr_step, total_steps): """Get annealed rate. @@ -164,7 +207,7 @@ def __init__( # ConvFeatureExtractionModel self.conv_feature_layers = conv_feature_layers - feature_enc_layers = eval(conv_feature_layers) + feature_enc_layers = _parse_conv_feature_layers(conv_feature_layers) self.extractor_embed = feature_enc_layers[-1][0] self.feature_extractor = ConvFeatureExtractionModel( conv_layers=feature_enc_layers, @@ -401,7 +444,7 @@ def _conv_out_length(input_length, kernel_size, stride): """ return torch.floor((input_length - kernel_size).to(torch.float32) / stride + 1) - conv_cfg_list = eval(self.conv_feature_layers) + conv_cfg_list = _parse_conv_feature_layers(self.conv_feature_layers) for i in range(len(conv_cfg_list)): input_lengths = _conv_out_length( diff --git a/funasr/models/llm_asr/model.py b/funasr/models/llm_asr/model.py index 5a1ccd5f41..3699bce062 100644 --- a/funasr/models/llm_asr/model.py +++ b/funasr/models/llm_asr/model.py @@ -1,3 +1,4 @@ +import ast import logging from typing import Union, Dict, List, Tuple, Optional @@ -1340,7 +1341,10 @@ def data_load_speech(self, contents: dict, tokenizer, frontend, meta_data={}, ** if sub_str.startswith("!"): sub_str = sub_str[1:] if sub_str.startswith("!"): # !!bytes - sub_str = eval(sub_str[1:]) + # 安全解析字节字面量(如 b'...')。 + # 原 eval() 会执行任意代码:训练数据/用户输入中的 + # <|startofspeech|>!!...<|endofspeech|> 片段可注入恶意表达式。 + sub_str = ast.literal_eval(sub_str[1:]) try: time1 = time.perf_counter() data_src = load_audio_text_image_video(sub_str, fs=frontend.fs) diff --git a/funasr/models/qwen_audio/audio.py b/funasr/models/qwen_audio/audio.py index 133ad86def..6d85401780 100644 --- a/funasr/models/qwen_audio/audio.py +++ b/funasr/models/qwen_audio/audio.py @@ -47,7 +47,7 @@ def get_T_after_cnn(L_in, dilation=1): L_in: TODO. dilation: TODO. """ - for padding, kernel_size, stride in eval("[(1,3,1)] + [(1,3,2)] "): + for padding, kernel_size, stride in [(1, 3, 1), (1, 3, 2)]: L_out = L_in + 2 * padding - dilation * (kernel_size - 1) - 1 L_out = 1 + L_out // stride L_in = L_out diff --git a/runtime/python/http/server.py b/runtime/python/http/server.py index 4f97821355..a85cdc444b 100644 --- a/runtime/python/http/server.py +++ b/runtime/python/http/server.py @@ -1,4 +1,6 @@ import argparse +import asyncio +import functools import logging import os import uuid @@ -7,6 +9,7 @@ import ffmpeg import uvicorn from fastapi import FastAPI, File, UploadFile +from fastapi.responses import JSONResponse from modelscope.utils.logger import get_logger from funasr import AutoModel @@ -52,6 +55,13 @@ parser.add_argument("--certfile", type=str, default=None, required=False, help="certfile for ssl") parser.add_argument("--keyfile", type=str, default=None, required=False, help="keyfile for ssl") parser.add_argument("--temp_dir", type=str, default="temp_dir/", required=False, help="temp dir") +parser.add_argument( + "--max_upload_mb", + type=int, + default=100, + required=False, + help="max upload audio size in MB (reject larger uploads to avoid memory/disk exhaustion)", +) args = parser.parse_args() logger.info("----------- Configuration Arguments -----------") for arg, value in vars(args).items(): @@ -91,30 +101,54 @@ @app.post("/recognition") async def api_recognition(audio: UploadFile = File(..., description="audio file")): - suffix = audio.filename.split(".")[-1] - audio_path = f"{args.temp_dir}/{str(uuid.uuid1())}.{suffix}" - async with aiofiles.open(audio_path, "wb") as out_file: - content = await audio.read() - await out_file.write(content) + # 扩展名只保留字母数字,防止客户端用带 ../ 的文件名做路径穿越 + raw_suffix = (audio.filename or "").rsplit(".", 1)[-1] + suffix = "".join(c for c in raw_suffix if c.isalnum())[:10] or "audio" + audio_path = os.path.join(args.temp_dir, f"{uuid.uuid1()}.{suffix}") + max_bytes = args.max_upload_mb * 1024 * 1024 + received = 0 try: - audio_bytes, _ = ( - ffmpeg.input(audio_path, threads=0) - .output("-", format="s16le", acodec="pcm_s16le", ac=1, ar=16000) - .run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True) + # 分块落盘并限制总大小,避免超大上传耗尽内存/磁盘 + async with aiofiles.open(audio_path, "wb") as out_file: + while True: + chunk = await audio.read(1024 * 1024) + if not chunk: + break + received += len(chunk) + if received > max_bytes: + return JSONResponse( + status_code=413, + content={ + "msg": f"音频文件过大(上限 {args.max_upload_mb}MB)", + "code": 1, + }, + ) + await out_file.write(chunk) + try: + audio_bytes, _ = ( + ffmpeg.input(audio_path, threads=0) + .output("-", format="s16le", acodec="pcm_s16le", ac=1, ar=16000) + .run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True) + ) + except Exception as e: + logger.error(f"读取音频文件发生错误,错误信息:{e}") + return {"msg": "读取音频文件发生错误", "code": 1} + # 推理是阻塞操作:放线程池执行,避免卡死事件循环导致所有客户端挂起 + loop = asyncio.get_running_loop() + rec_results = await loop.run_in_executor( + None, + functools.partial(model.generate, input=audio_bytes, is_final=True, **param_dict), ) - except Exception as e: - logger.error(f"读取音频文件发生错误,错误信息:{e}") - return {"msg": "读取音频文件发生错误", "code": 1} - rec_results = model.generate(input=audio_bytes, is_final=True, **param_dict) - # 结果为空 - if len(rec_results[0]["text"] ) == 0: - return {"text": "", "sentences": [], "code": 0} - elif len(rec_results[0]["text"] ) > 0: - # 解析识别结果 + # 结果为空 + if not rec_results: + return {"text": "", "sentences": [], "code": 0} rec_result = rec_results[0] - text = rec_result["text"] + text = rec_result.get("text", "") + if len(text) == 0: + return {"text": "", "sentences": [], "code": 0} + # 解析识别结果 sentences = [] - for sentence in rec_result["sentence_info"]: + for sentence in rec_result.get("sentence_info", []): # 每句话的时间戳 sentences.append( {"text": sentence["text"], "start": sentence["start"], "end": sentence["end"]} @@ -122,9 +156,12 @@ async def api_recognition(audio: UploadFile = File(..., description="audio file" ret = {"text": text, "sentences": sentences, "code": 0} logger.info(f"识别结果:{ret}") return ret - else: - logger.info(f"识别结果:{rec_results}") - return {"msg": "未知错误", "code": -1} + finally: + # 无论成功失败都清理临时文件,防止磁盘被上传文件占满 + try: + os.remove(audio_path) + except OSError: + pass if __name__ == "__main__": diff --git a/runtime/python/libtorch/funasr_torch/utils/utils.py b/runtime/python/libtorch/funasr_torch/utils/utils.py index ee438521c9..1c5b4a73d4 100644 --- a/runtime/python/libtorch/funasr_torch/utils/utils.py +++ b/runtime/python/libtorch/funasr_torch/utils/utils.py @@ -125,7 +125,8 @@ def read_yaml(yaml_path: Union[str, Path]) -> Dict: raise FileExistsError(f"The {yaml_path} does not exist.") with open(str(yaml_path), "rb") as f: - data = yaml.load(f, Loader=yaml.Loader) + # safe_load:模型配置可能来自远端下载,yaml.Loader 会执行任意 Python 标签 + data = yaml.safe_load(f) return data diff --git a/runtime/python/onnxruntime/funasr_onnx/utils/utils.py b/runtime/python/onnxruntime/funasr_onnx/utils/utils.py index 58d7dca97e..f06a2729a6 100644 --- a/runtime/python/onnxruntime/funasr_onnx/utils/utils.py +++ b/runtime/python/onnxruntime/funasr_onnx/utils/utils.py @@ -360,7 +360,8 @@ def read_yaml(yaml_path: Union[str, Path]) -> Dict: raise FileExistsError(f"The {yaml_path} does not exist.") with open(str(yaml_path), "rb") as f: - data = yaml.load(f, Loader=yaml.Loader) + # safe_load:模型配置可能来自远端下载,yaml.Loader 会执行任意 Python 标签 + data = yaml.safe_load(f) return data diff --git a/runtime/python/websocket/funASR-Service.py b/runtime/python/websocket/funASR-Service.py index a8d8803b80..9a4a00df89 100644 --- a/runtime/python/websocket/funASR-Service.py +++ b/runtime/python/websocket/funASR-Service.py @@ -14,6 +14,8 @@ from werkzeug.utils import secure_filename import torch import os +import uuid +import tempfile from openai import OpenAI from funasr import AutoModel @@ -51,8 +53,9 @@ def Register_Speaker(): if file.filename == '': return jsonify({"error": "Empty filename"}), 400 - # 保存上传文件 - filename = secure_filename(file.filename) + # 保存上传文件(uuid 前缀防止并发同名上传互相覆盖/误删) + filename = secure_filename(file.filename) or "upload.audio" + filename = f"{uuid.uuid4().hex}_{filename}" filepath = os.path.join(app.config['UPLOAD_FOLDER'], filename) file.save(filepath) try: @@ -95,18 +98,22 @@ def speech_recognition_Timestamp_cam_identify_speakers(): if file.filename == '': return jsonify({"error": "Empty filename"}), 400 - # 保存上传文件 - filename = secure_filename(file.filename) + # 保存上传文件(uuid 前缀防止并发同名上传互相覆盖/误删) + filename = secure_filename(file.filename) or "upload.audio" + filename = f"{uuid.uuid4().hex}_{filename}" filepath = os.path.join(app.config['UPLOAD_FOLDER'], filename) file.save(filepath) # 2. 从FormData读取参数并解析JSON # 读取identify_speakers(默认False) - identify_speakers_str = request.form.get('identify_speakers', 'false') - identify_speakers = json.loads(identify_speakers_str.lower()) # 转为bool + try: + identify_speakers_str = request.form.get('identify_speakers', 'false') + identify_speakers = json.loads(identify_speakers_str.lower()) # 转为bool - # 读取speaker_db(默认空字典) - speaker_db_str = request.form.get('speaker_db', '{}') - speaker_db = json.loads(speaker_db_str) + # 读取speaker_db(默认空字典) + speaker_db_str = request.form.get('speaker_db', '{}') + speaker_db = json.loads(speaker_db_str) + except (ValueError, TypeError, AttributeError) as e: + return jsonify({"error": f"invalid form JSON: {e}"}), 400 # 3. 验证声纹库(如果需要对比) if identify_speakers: @@ -158,7 +165,10 @@ def _extract_audio_segment(audio_path, start_sec, end_sec): segment = waveform[:, start_sample:end_sample] # 保存为临时文件(pipeline需要文件路径) - temp_path = f"/tmp/temp_segment_{start_sec}_{end_sec}.wav" + # 唯一临时文件:原固定 /tmp 路径会让并发请求互相覆盖片段(声纹比对错乱), + # 可被符号链接攻击写任意文件,且 /tmp 在 Windows 上不存在 + fd, temp_path = tempfile.mkstemp(suffix=".wav", prefix="funasr_seg_") + os.close(fd) torchaudio.save(temp_path, segment, sample_rate) return temp_path @@ -175,6 +185,8 @@ def process_cam_result_with_identify_speakers(result,speaker_db,filepath,identif # 创建JSON格式的输出 output = [] sentence_infos = data.get('sentence_info',[]) + if not sentence_infos: + return [] current_sentence = { "spk": sentence_infos[0]["spk"], @@ -275,4 +287,5 @@ def calculate_similarity(): }) if __name__ == '__main__': - app.run(host='0.0.0.0', port=10099, debug=True) + # debug=True 会在 0.0.0.0 上暴露 Werkzeug 调试器(可远程执行任意代码),必须关闭 + app.run(host='0.0.0.0', port=10099, debug=False) diff --git a/runtime/python/websocket/funasr_wss_server.py b/runtime/python/websocket/funasr_wss_server.py index 2a449e6423..88ef343274 100644 --- a/runtime/python/websocket/funasr_wss_server.py +++ b/runtime/python/websocket/funasr_wss_server.py @@ -218,7 +218,9 @@ def save_offline_wav_segment_sync(websocket, audio_bytes: bytes, reason: str = " base_dir = getattr(websocket, "offline_save_dir", args.save_offline_segments_dir) _ensure_dir(base_dir) - wav_name = (getattr(websocket, "wav_name", "microphone") or "microphone").replace("/", "_") + # 同时过滤 / 与 \(Windows 分隔符),并取 basename,防止客户端 wav_name 路径穿越 + wav_name = getattr(websocket, "wav_name", "microphone") or "microphone" + wav_name = os.path.basename(str(wav_name).replace("\\", "/")).replace("/", "_") ts = int(time.time() * 1000) fname = f"{wav_name}_{ts}_seg{seg_idx:04d}_{reason}_{duration_ms}ms.wav" out_path = os.path.join(base_dir, fname) @@ -478,8 +480,9 @@ async def finish_input(send_end_ack): end_of_input = not websocket.is_speaking if "chunk_interval" in messagejson: - websocket.chunk_interval = _safe_int( - messagejson["chunk_interval"], websocket.chunk_interval + # 至少为 1:客户端传 0 会让后续 len(...) % chunk_interval 除零崩溃 + websocket.chunk_interval = max( + 1, _safe_int(messagejson["chunk_interval"], websocket.chunk_interval) ) if "wav_name" in messagejson: @@ -489,7 +492,14 @@ async def finish_input(send_end_ack): chunk_size = messagejson["chunk_size"] if isinstance(chunk_size, str): chunk_size = [x.strip() for x in chunk_size.split(",") if x.strip()] - websocket.status_dict_asr_online["chunk_size"] = [int(x) for x in chunk_size] + # 校验客户端输入:非法类型只记录错误,不再让异常终止整个会话; + # 长度不足等问题仍由后续每帧的 VAD chunk_size 检查处理 + try: + websocket.status_dict_asr_online["chunk_size"] = [ + int(x) for x in chunk_size + ] + except (TypeError, ValueError): + record_error(f"invalid chunk_size: {messagejson['chunk_size']!r}") if "encoder_chunk_look_back" in messagejson: websocket.status_dict_asr_online["encoder_chunk_look_back"] = messagejson[ @@ -510,6 +520,8 @@ async def finish_input(send_end_ack): if "mode" in messagejson: requested_mode = messagejson["mode"] if requested_mode and requested_mode not in ("online", "offline", "2pass"): + # 故意赋值非法 mode:后续音频帧会在入口处被丢弃, + # 并在 end ack 中报告 unsupported mode(见测试契约) websocket.mode = requested_mode record_error(f"unsupported mode: {requested_mode!r}") else: diff --git a/runtime/triton_gpu/client/decode_manifest_triton_wo_cuts.py b/runtime/triton_gpu/client/decode_manifest_triton_wo_cuts.py index 379b28df00..4090fd5b9f 100644 --- a/runtime/triton_gpu/client/decode_manifest_triton_wo_cuts.py +++ b/runtime/triton_gpu/client/decode_manifest_triton_wo_cuts.py @@ -45,6 +45,7 @@ from pydub import AudioSegment import argparse +import ast import asyncio import math import time @@ -182,7 +183,8 @@ def load_manifest(fp): data = [] with open(fp) as f: for i, dp in enumerate(f.readlines()): - dp = eval(dp) + # manifest 每行是 dict 字面量;用 literal_eval 防止恶意文件执行任意代码 + dp = ast.literal_eval(dp) dp["id"] = i data.append(dp) return data diff --git a/runtime/triton_gpu/model_repo_paraformer_large_offline/feature_extractor/1/model.py b/runtime/triton_gpu/model_repo_paraformer_large_offline/feature_extractor/1/model.py index e6eb3b9355..34a2ec51f8 100644 --- a/runtime/triton_gpu/model_repo_paraformer_large_offline/feature_extractor/1/model.py +++ b/runtime/triton_gpu/model_repo_paraformer_large_offline/feature_extractor/1/model.py @@ -227,7 +227,7 @@ def initialize(self, args): value = value["string_value"] if key == "config_path": with open(str(value), "rb") as f: - config = yaml.load(f, Loader=yaml.Loader) + config = yaml.safe_load(f) if key == "cmvn_path": cmvn_path = str(value) diff --git a/runtime/triton_gpu/model_repo_paraformer_large_online/cif_search/1/model.py b/runtime/triton_gpu/model_repo_paraformer_large_online/cif_search/1/model.py index aeecd33016..a60b31c020 100755 --- a/runtime/triton_gpu/model_repo_paraformer_large_online/cif_search/1/model.py +++ b/runtime/triton_gpu/model_repo_paraformer_large_online/cif_search/1/model.py @@ -147,7 +147,7 @@ def init_vocab(self, parameters): def load_vocab(self, vocab_file): with open(str(vocab_file), "rb") as f: - config = yaml.load(f, Loader=yaml.Loader) + config = yaml.safe_load(f) return config["token_list"] async def execute(self, requests): diff --git a/runtime/triton_gpu/model_repo_paraformer_large_online/feature_extractor/1/model.py b/runtime/triton_gpu/model_repo_paraformer_large_online/feature_extractor/1/model.py index 716a075469..f86542d061 100755 --- a/runtime/triton_gpu/model_repo_paraformer_large_online/feature_extractor/1/model.py +++ b/runtime/triton_gpu/model_repo_paraformer_large_online/feature_extractor/1/model.py @@ -117,7 +117,7 @@ def initialize(self, args): value = value["string_value"] if key == "config_path": with open(str(value), "rb") as f: - config = yaml.load(f, Loader=yaml.Loader) + config = yaml.safe_load(f) opts = kaldifeat.FbankOptions() opts.frame_opts.dither = 0.0 diff --git a/runtime/triton_gpu/model_repo_sense_voice_small/feature_extractor/1/model.py b/runtime/triton_gpu/model_repo_sense_voice_small/feature_extractor/1/model.py index 21c2d723df..27fe4f1a08 100644 --- a/runtime/triton_gpu/model_repo_sense_voice_small/feature_extractor/1/model.py +++ b/runtime/triton_gpu/model_repo_sense_voice_small/feature_extractor/1/model.py @@ -227,7 +227,7 @@ def initialize(self, args): value = value["string_value"] if key == "config_path": with open(str(value), "rb") as f: - config = yaml.load(f, Loader=yaml.Loader) + config = yaml.safe_load(f) if key == "cmvn_path": cmvn_path = str(value) config["frontend_conf"]["cmvn_file"] = cmvn_path