问题现象

在Open WebUI里上传一个附件，然后针对这个附件做聊天或者问答，但是返回异常。

查看后台日志，有如下报错：

WARNI [python_multipart.multipart] Skipping data after last boundary
INFO  [open_webui.routers.files] file.content_type: text/plain
INFO  [open_webui.routers.retrieval] save_docs_to_vector_db: document doubao2.txt file-16bc9379-b8ff-4884-ba51-166efbd67e1a
Collection file-16bc9379-b8ff-4884-ba51-166efbd67e1a does not exist.
INFO  [open_webui.routers.retrieval] adding to collection file-16bc9379-b8ff-4884-ba51-166efbd67e1a
ERROR [open_webui.routers.retrieval] 'NoneType' object has no attribute 'encode'
Traceback (most recent call last):
  File "/root/miniconda3/lib/python3.12/site-packages/open_webui/routers/retrieval.py", line 784, in save_docs_to_vector_db
    embeddings = embedding_function(
                 ^^^^^^^^^^^^^^^^^^^
  File "/root/miniconda3/lib/python3.12/site-packages/open_webui/retrieval/utils.py", line 265, in <lambda>
    return lambda query, user=None: embedding_function.encode(query).tolist()
                                    ^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'encode'
ERROR [open_webui.routers.retrieval] 'NoneType' object has no attribute 'encode'

问题分析

从后台的错误上看应该是save_docs_to_vector_db的embedding_functionb出错了。
再具体看了一下代码应该是没的安装语义向量模型引擎sentence-transformers导致的。

解决方案

切回到前台的Open WebUI，用管理员身份登录进去。然后点右上角的“设置”，并找到“文档”，然后在该界面上点一下右这的下载的图标，如下图所示。

若下载成功了，应该就OK了。不过，实际上我这边的情况是下载失败。所以同志还要努力。继续往下走。

文档设置中下载sentence-transformers/all-MiniLM-L6-v2失败

查看后台日志，发现报错如下

ERROR [open_webui.retrieval.utils] Cannot determine model snapshot path: Cannot find an appropriate cached snapshot folder for the specified revision on the local disk and outgoing traffic has been disabled. To enable repo look-ups and downloads online, pass 'local_files_only=False' as input.
Traceback (most recent call last):
  File "/root/miniconda3/lib/python3.12/site-packages/open_webui/retrieval/utils.py", line 425, in get_model_path
    model_repo_path = snapshot_download(**snapshot_kwargs)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/miniconda3/lib/python3.12/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn
    return fn(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^
  File "/root/miniconda3/lib/python3.12/site-packages/huggingface_hub/_snapshot_download.py", line 219, in snapshot_download
    raise LocalEntryNotFoundError(
huggingface_hub.errors.LocalEntryNotFoundError: Cannot find an appropriate cached snapshot folder for the specified revision on the local disk and outgoing traffic has been disabled. To enable repo look-ups and downloads online, pass 'local_files_only=False' as input.
WARNI [sentence_transformers.SentenceTransformer] No sentence-transformers model found with name sentence-transformers/all-MiniLM-L6-v2. Creating a new one with mean pooling.
INFO:     127.0.0.1:56474 - "POST /api/v1/retrieval/embedding/update HTTP/1.1" 200 OK

从这个报错来看，应该是open_webui/retrieval/utils.py的get_model_path()在调用snapshot_download()连不上huggingface导致的。因此，我们需要再设置一下代码，翻个墙，或者设置一下huggingface的镜像站。

后台关了Open WebUI，然后设置一下代理及huggingface镜像。

设置代理

import subprocess
import os

result = subprocess.run('bash -c "source /etc/network_turbo && env | grep proxy"', shell=True, capture_output=True, text=True)
output = result.stdout
for line in output.splitlines():
    if '=' in line:
        var, value = line.split('=', 1)
        os.environ[var] = value

设置huggingface镜像
export HF_ENDPOINT=https://hf-mirror.com

重启 Open WebUI

./start.sh 后再chat with 附件，一切正常。

以下是我的start.sh的代码：

#!/usr/bin/env bash

# open-webui 不原生支持HOST和PORT环境变量，需手动传递参数
# https://docs.openwebui.com/getting-started/env-configuration/#port
# 若open-webui运行异常，可执行`rm -rf ./data`清除数据后重启服务并清理浏览器缓存

# https://docs.openwebui.com/getting-started/env-configuration/
export DATA_DIR="$(pwd)/data"
export ENABLE_OLLAMA_API=False
export ENABLE_OPENAI_API=True
export OPENAI_API_KEY="dont_change_this_cuz_openai_is_the_mcdonalds_of_ai"
export OPENAI_API_BASE_URL="http://127.0.0.1:8000/v1" # <--- 需与ktransformers/llama.cpp的API配置匹配
#export DEFAULT_MODELS="openai/foo/bar" # <--- 保留注释，此参数用于`litellm`接入
export WEBUI_AUTH=True
export DEFAULT_USER_ROLE="admin"
export HOST=127.0.0.1
export PORT=3000

# 如果仅加载了R1模型，可通过禁用以下功能节省时间：
#   * 标签生成
#   * 自动补全
#   * 标题生成
# https://github.com/kvcache-ai/ktransformers/issues/618#issuecomment-2681381587
export ENABLE_TAGS_GENERATION=False
export ENABLE_AUTOCOMPLETE_GENERATION=False
# 或许目前需手动在界面禁用该功能？？？
export TITLE_GENERATION_PROMPT_TEMPLATE=""

open-webui serve \
      --host $HOST \
        --port $PORT

Open WebUI上传附件聊天及本地知识库功能异常问题