feat: 实现文件附件功能，前端管理上传状态，后端解析文本、PDF、Word文件内容并注入聊天。
This commit is contained in:
SuperManTouX
parent
2d4e777056
commit
ec16cc93ed
|
|
@ -15,6 +15,7 @@ uploads
|
|||
.env
|
||||
.venv
|
||||
__pycache__
|
||||
.claude
|
||||
|
||||
# Editor directories and files
|
||||
.vscode/*
|
||||
|
|
|
|||
|
|
@ -34,6 +34,211 @@ upload_dir = Path("uploads")
|
|||
upload_dir.mkdir(exist_ok=True)
|
||||
|
||||
|
||||
def _extract_text_from_docmind(obj, depth: int = 0) -> str:
|
||||
"""
|
||||
递归提取 DocMind JSON 结构中的可读文本。
|
||||
DashScopeParse 返回的 text 字段是 JSON 字符串，内部为文档智能解析的树形结构。
|
||||
"""
|
||||
if depth > 15:
|
||||
return ""
|
||||
|
||||
if isinstance(obj, str):
|
||||
s = obj.strip()
|
||||
# 过滤极短、URL、base64 等非正文字符串
|
||||
if len(s) > 3 and not s.startswith(('http://', 'https://', 'data:', 'oss://')):
|
||||
return s
|
||||
return ""
|
||||
|
||||
if isinstance(obj, list):
|
||||
parts = [_extract_text_from_docmind(item, depth + 1) for item in obj]
|
||||
return "\n".join(p for p in parts if p)
|
||||
|
||||
if isinstance(obj, dict):
|
||||
# 优先处理文本相关字段
|
||||
priority_keys = ['content', 'text', 'paragraph', 'caption', 'value', 'title']
|
||||
# 跳过纯元数据字段
|
||||
skip_keys = {'backlink', 'pos', 'index', 'style', 'font', 'color',
|
||||
'size', 'hash', 'id_', 'id', 'layouts', 'type', 'link'}
|
||||
parts = []
|
||||
for key in priority_keys:
|
||||
if key in obj:
|
||||
t = _extract_text_from_docmind(obj[key], depth + 1)
|
||||
if t:
|
||||
parts.append(t)
|
||||
for key, val in obj.items():
|
||||
if key not in priority_keys and key not in skip_keys:
|
||||
t = _extract_text_from_docmind(val, depth + 1)
|
||||
if t:
|
||||
parts.append(t)
|
||||
return "\n".join(parts)
|
||||
|
||||
return ""
|
||||
|
||||
|
||||
def _read_file_content(file_url: str):
|
||||
"""
|
||||
【路线一：本地文本提取】对纯文本格式文件，直接读取内容注入消息。
|
||||
【路线二：DashScopeParse】对 doc/docx/pdf 文件，返回 (local_path, suffix) 供异步调用。
|
||||
|
||||
返回值：
|
||||
- str：文本内容（路线一成功）
|
||||
- tuple(Path, str)：(本地路径, 扩展名)，需异步调用 DashScopeParse（路线二）
|
||||
- None：不支持的文件类型
|
||||
"""
|
||||
try:
|
||||
from urllib.parse import urlparse
|
||||
parsed = urlparse(file_url)
|
||||
relative_path = parsed.path.lstrip('/')
|
||||
local_path = Path(relative_path)
|
||||
|
||||
if not local_path.exists():
|
||||
return f"[文件不存在: {local_path}]"
|
||||
|
||||
suffix = local_path.suffix.lower()
|
||||
|
||||
# 路线一：纯文本格式直接读取
|
||||
text_extensions = {'.txt', '.md', '.csv', '.json', '.xml',
|
||||
'.yaml', '.yml', '.log', '.py', '.js', '.ts', '.html', '.css'}
|
||||
if suffix in text_extensions:
|
||||
with open(local_path, 'r', encoding='utf-8', errors='replace') as f:
|
||||
content = f.read()
|
||||
max_len = 8000
|
||||
if len(content) > max_len:
|
||||
content = content[:max_len] + f"\n\n[...文件内容过长，已截断，共 {len(content)} 字符]"
|
||||
return content
|
||||
|
||||
# 路线二：doc/docx/pdf 使用 DashScopeParse 云端解析
|
||||
dashscope_extensions = {'.doc', '.docx', '.pdf'}
|
||||
if suffix in dashscope_extensions:
|
||||
return (local_path, suffix) # 交给异步函数处理
|
||||
|
||||
# 其余格式（xlsx、pptx 等）暂不支持内容读取
|
||||
return None
|
||||
|
||||
except Exception as e:
|
||||
print(f"[WARNING] 读取文件内容失败: {e}")
|
||||
return f"[文件读取失败: {str(e)}]"
|
||||
|
||||
|
||||
|
||||
async def _parse_with_dashscope(local_path: Path) -> str:
|
||||
"""
|
||||
【路线二：DashScopeParse】使用阿里云文档智能解析 doc/docx/pdf 文件。
|
||||
在线程池中运行（避免阻塞 FastAPI 事件循环）。
|
||||
仅支持 .doc/.docx/.pdf，文件大小 ≤100MB，页数 ≤1000 页。
|
||||
"""
|
||||
import asyncio
|
||||
|
||||
def _sync_parse():
|
||||
try:
|
||||
from llama_index.readers.dashscope.base import DashScopeParse
|
||||
from llama_index.readers.dashscope.utils import ResultType
|
||||
import json
|
||||
|
||||
api_key = os.getenv("ALIYUN_API_KEY")
|
||||
parser = DashScopeParse(
|
||||
result_type=ResultType.DASHSCOPE_DOCMIND,
|
||||
api_key=api_key,
|
||||
num_workers=1,
|
||||
)
|
||||
print(f"[INFO] DashScopeParse: 开始解析 {local_path.name} ...")
|
||||
documents = parser.load_data(file_path=[str(local_path)])
|
||||
|
||||
if not documents:
|
||||
return f"[DashScopeParse: {local_path.name} 解析结果为空]"
|
||||
|
||||
texts = []
|
||||
for doc in documents:
|
||||
try:
|
||||
content = json.loads(doc.text)
|
||||
extracted = _extract_text_from_docmind(content)
|
||||
texts.append(extracted if extracted else doc.text[:6000])
|
||||
except Exception:
|
||||
texts.append(doc.text[:6000] if doc.text else "")
|
||||
|
||||
result = "\n\n".join(t for t in texts if t)
|
||||
print(f"[INFO] DashScopeParse: {local_path.name} 解析完成，提取 {len(result)} 字符")
|
||||
return result or f"[DashScopeParse: {local_path.name} 未能提取到文本内容]"
|
||||
|
||||
except ImportError:
|
||||
return "[错误：DashScopeParse 未安装，请运行: pip install llama-index-core llama-index-readers-dashscope]"
|
||||
except Exception as e:
|
||||
print(f"[ERROR] DashScopeParse 解析失败: {e}")
|
||||
return f"[DashScopeParse 解析失败: {str(e)}]"
|
||||
|
||||
loop = asyncio.get_event_loop()
|
||||
return await loop.run_in_executor(None, _sync_parse)
|
||||
|
||||
|
||||
async def _inject_files_into_messages(messages: list, files: list) -> list:
|
||||
"""
|
||||
将文件内容异步注入到消息列表中。
|
||||
- 文本类文件（路线一）：读取内容并追加到最后一条 user 消息中
|
||||
- doc/docx/pdf（路线二）：调用 DashScopeParse 云端解析后注入
|
||||
- 其他二进制文件：仅告知 AI 文件名和类型
|
||||
"""
|
||||
if not files:
|
||||
return messages
|
||||
|
||||
file_context_parts = []
|
||||
for file_url in files:
|
||||
from urllib.parse import urlparse
|
||||
parsed = urlparse(file_url)
|
||||
filename = parsed.path.split('/')[-1]
|
||||
suffix = Path(filename).suffix.lower()
|
||||
|
||||
result = _read_file_content(file_url)
|
||||
|
||||
if isinstance(result, str):
|
||||
# 路线一：文本内容，直接嵌入
|
||||
file_context_parts.append(
|
||||
f"--- 附件文件内容（{filename}）---\n{result}\n--- 附件结束 ---"
|
||||
)
|
||||
elif isinstance(result, tuple):
|
||||
# 路线二：doc/docx/pdf → 调用 DashScopeParse
|
||||
local_path, _ = result
|
||||
print(f"[INFO] 路线二：调用 DashScopeParse 解析 {filename}")
|
||||
parsed_text = await _parse_with_dashscope(local_path)
|
||||
file_context_parts.append(
|
||||
f"--- 附件文件内容（{filename}，阿里云文档智能解析）---\n"
|
||||
f"{parsed_text}\n--- 附件结束 ---"
|
||||
)
|
||||
else:
|
||||
# 其他不支持的格式：仅告知文件信息
|
||||
file_context_parts.append(
|
||||
f"[用户上传了一个文件: {filename}，类型: {suffix}，暂不支持自动读取内容，请告知用户。]"
|
||||
)
|
||||
|
||||
if not file_context_parts:
|
||||
return messages
|
||||
|
||||
file_context_text = "\n\n" + "\n\n".join(file_context_parts)
|
||||
|
||||
# 把文件内容追加到最后一条 user 消息
|
||||
messages = list(messages) # 复制，避免修改原始列表
|
||||
for i in range(len(messages) - 1, -1, -1):
|
||||
msg = messages[i]
|
||||
if isinstance(msg, dict) and msg.get('role') == 'user':
|
||||
content = msg.get('content', '')
|
||||
if isinstance(content, str):
|
||||
messages[i] = dict(msg, content=content + file_context_text)
|
||||
elif isinstance(content, list):
|
||||
# 找到现有的 text 项，追加内容
|
||||
new_content = list(content)
|
||||
appended = False
|
||||
for j, item in enumerate(new_content):
|
||||
if isinstance(item, dict) and item.get('type') == 'text':
|
||||
new_content[j] = dict(item, text=item['text'] + file_context_text)
|
||||
appended = True
|
||||
break
|
||||
if not appended:
|
||||
new_content.append({'type': 'text', 'text': file_context_text})
|
||||
messages[i] = dict(msg, content=new_content)
|
||||
break
|
||||
|
||||
return messages
|
||||
|
||||
|
||||
async def chat_endpoint_handler(body: dict):
|
||||
"""
|
||||
聊天接口处理器 - 与阿里云百炼API兼容的接口
|
||||
|
|
@ -56,24 +261,33 @@ async def chat_endpoint_handler(body: dict):
|
|||
stream = body.get('stream', True)
|
||||
temperature = body.get('temperature', 0.7)
|
||||
max_tokens = body.get('max_tokens', 2000)
|
||||
|
||||
# 处理 files 附件：将文件内容注入到最后一条 user 消息中
|
||||
files = body.get('files', [])
|
||||
if files:
|
||||
messages = await _inject_files_into_messages(messages, files)
|
||||
# 调试：打印注入后最后一条 user 消息的内容（截断显示 500 字）
|
||||
for msg in reversed(messages):
|
||||
if isinstance(msg, dict) and msg.get('role') == 'user':
|
||||
content_preview = str(msg.get('content', ''))[:500]
|
||||
print(f"[DEBUG] 注入文件后 user 消息内容预览: {content_preview}")
|
||||
break
|
||||
else:
|
||||
# 否则是前端简化格式 (来自chat函数)
|
||||
# 需要将其转换为OpenAI兼容格式
|
||||
message_text = body.get('message', '')
|
||||
|
||||
# 检查message是否已经是格式化的列表（带图片的情况）
|
||||
if isinstance(message_text, list):
|
||||
user_content = message_text
|
||||
else:
|
||||
# 如果是字符串，转换为标准格式
|
||||
user_content = [{"type": "text", "text": message_text}]
|
||||
|
||||
messages = [
|
||||
{"role": "system", "content": body.get('systemPrompt', '你是一个支持视觉理解的助手。')},
|
||||
{"role": "system", "content": body.get('systemPrompt', '你是一个智能助手，可以分析用户发送的文本和文件内容。')},
|
||||
{"role": "user", "content": user_content}
|
||||
]
|
||||
model = body.get('model', 'qwen-plus')
|
||||
stream = body.get('stream', False) # 默认为非流式
|
||||
stream = body.get('stream', False)
|
||||
temperature = body.get('temperature', 0.7)
|
||||
max_tokens = body.get('maxTokens', 2000)
|
||||
|
||||
|
|
@ -602,13 +816,41 @@ async def delete_conversation_handler(conversation_id: str):
|
|||
async def upload_file_handler(file: UploadFile = File(...)):
|
||||
"""文件上传处理器"""
|
||||
try:
|
||||
# 检查文件类型
|
||||
allowed_types = ['image/jpeg', 'image/png', 'image/gif', 'image/webp', 'text/plain', 'application/pdf']
|
||||
if file.content_type not in allowed_types:
|
||||
raise HTTPException(status_code=400, detail=f"不支持的文件类型: {file.content_type}")
|
||||
# 允许的 MIME 类型（宽松策略）
|
||||
allowed_types = {
|
||||
# 图片
|
||||
'image/jpeg', 'image/png', 'image/gif', 'image/webp', 'image/bmp', 'image/svg+xml',
|
||||
# 文本类
|
||||
'text/plain', 'text/csv', 'text/markdown', 'text/html', 'text/xml',
|
||||
'application/json', 'application/xml',
|
||||
# PDF
|
||||
'application/pdf',
|
||||
# Office 文档
|
||||
'application/msword',
|
||||
'application/vnd.openxmlformats-officedocument.wordprocessingml.document',
|
||||
'application/vnd.ms-excel',
|
||||
'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet',
|
||||
'application/vnd.ms-powerpoint',
|
||||
'application/vnd.openxmlformats-officedocument.presentationml.presentation',
|
||||
}
|
||||
|
||||
# 允许的扩展名（兜底：MIME 类型可能被浏览器误判）
|
||||
allowed_extensions = {
|
||||
'.jpg', '.jpeg', '.png', '.gif', '.webp', '.bmp',
|
||||
'.txt', '.md', '.csv', '.json', '.xml', '.yaml', '.yml', '.log',
|
||||
'.pdf', '.doc', '.docx', '.xls', '.xlsx', '.ppt', '.pptx',
|
||||
'.py', '.js', '.ts', '.html', '.css'
|
||||
}
|
||||
|
||||
file_extension = Path(file.filename).suffix.lower()
|
||||
|
||||
if file.content_type not in allowed_types and file_extension not in allowed_extensions:
|
||||
raise HTTPException(
|
||||
status_code=400,
|
||||
detail=f"不支持的文件类型: {file.content_type}（{file_extension}）"
|
||||
)
|
||||
|
||||
# 生成唯一文件名
|
||||
file_extension = Path(file.filename).suffix.lower()
|
||||
unique_filename = f"{int(datetime.utcnow().timestamp())}_{generate_unique_id()}{file_extension}"
|
||||
file_path = upload_dir / unique_filename
|
||||
|
||||
|
|
|
|||
|
|
@ -5,4 +5,7 @@ python-multipart==0.0.18
|
|||
python-dotenv==1.0.1
|
||||
aiofiles==24.1.0
|
||||
pydantic==2.9.2
|
||||
typing-extensions==4.12.2
|
||||
typing-extensions==4.12.2
|
||||
# 路线二：阿里云文档智能解析（doc/docx/pdf）
|
||||
llama-index-core>=0.10.0
|
||||
llama-index-readers-dashscope>=0.1.0
|
||||
|
|
|
|||
|
|
@ -118,7 +118,7 @@ function handlePin() {
|
|||
// 发送消息 - 使用真实 API
|
||||
async function handleSend(text: string, attachments: Attachment[]) {
|
||||
// 检查是否还有正在上传的附件
|
||||
const uploadingAttachments = attachments.filter(a => a.uploading);
|
||||
const uploadingAttachments = attachments.filter((a) => a.uploading);
|
||||
if (uploadingAttachments.length > 0) {
|
||||
// 等待所有上传完成
|
||||
const uploads = uploadingAttachments.map(async (attachment) => {
|
||||
|
|
@ -126,7 +126,9 @@ async function handleSend(text: string, attachments: Attachment[]) {
|
|||
// 但更简单的方法是等待一小段时间，让上传有机会完成
|
||||
return new Promise<void>((resolve) => {
|
||||
const checkUpload = () => {
|
||||
const stillUploading = attachments.some(a => a.id === attachment.id && a.uploading);
|
||||
const stillUploading = attachments.some(
|
||||
(a) => a.id === attachment.id && a.uploading,
|
||||
);
|
||||
if (!stillUploading) {
|
||||
resolve();
|
||||
} else {
|
||||
|
|
@ -140,7 +142,7 @@ async function handleSend(text: string, attachments: Attachment[]) {
|
|||
try {
|
||||
await Promise.all(uploads);
|
||||
} catch (error) {
|
||||
console.error('等待上传完成时发生错误:', error);
|
||||
console.error("等待上传完成时发生错误:", error);
|
||||
}
|
||||
}
|
||||
|
||||
|
|
@ -174,13 +176,19 @@ async function handleSend(text: string, attachments: Attachment[]) {
|
|||
// 提取图片URL用于发送给API
|
||||
const imageUrls = attachments
|
||||
.filter((a) => a.type === "image")
|
||||
.map(a => a.url);
|
||||
.map((a) => a.url);
|
||||
|
||||
// 提取非图片文件URL（txt, pdf, docx 等）
|
||||
const fileUrls = attachments
|
||||
.filter((a) => a.type === "file")
|
||||
.map((a) => a.url);
|
||||
|
||||
const stream = chatApi.streamChat(
|
||||
{
|
||||
message: text,
|
||||
conversationId: currentConversation.value?.id || "",
|
||||
images: imageUrls, // 添加图片URL
|
||||
images: imageUrls,
|
||||
files: fileUrls, // 传递文件 URL，后端会读取内容
|
||||
model: settings.value.defaultModel,
|
||||
stream: true,
|
||||
},
|
||||
|
|
|
|||
|
|
@ -34,7 +34,7 @@ export interface ChatRequest {
|
|||
conversationId?: string;
|
||||
message: string;
|
||||
images?: string[];
|
||||
files?: string[];
|
||||
files?: string[]; // 非图片附件 URL 列表
|
||||
model?: string;
|
||||
temperature?: number;
|
||||
maxTokens?: number;
|
||||
|
|
@ -93,14 +93,12 @@ class ChatApi {
|
|||
let userContent;
|
||||
if (request.images && request.images.length > 0) {
|
||||
// 如果有图片，则构建内容数组（针对阿里云DashScope API的格式）
|
||||
userContent = [
|
||||
{ type: "text", text: request.message }
|
||||
];
|
||||
userContent = [{ type: "text", text: request.message }];
|
||||
// 添加图片URL到内容中（阿里云格式）
|
||||
request.images.forEach(imageUrl => {
|
||||
request.images.forEach((imageUrl) => {
|
||||
userContent.push({
|
||||
type: "image_url",
|
||||
image_url: imageUrl // 注意：阿里云格式不需要嵌套对象
|
||||
image_url: imageUrl, // 注意：阿里云格式不需要嵌套对象
|
||||
});
|
||||
});
|
||||
} else {
|
||||
|
|
@ -110,17 +108,21 @@ class ChatApi {
|
|||
|
||||
// 将前端简化的请求翻译为 OpenAI 兼容的规范请求体
|
||||
const openAiRequest = {
|
||||
model: request.model || "qwen-plus", // 可能需要指定支持视觉的模型
|
||||
model: request.model || "qwen-plus", // 可能需要指定支持视觉的模型
|
||||
messages: [
|
||||
{ role: "system", content: request.systemPrompt || "你是一个支持视觉理解的助手。" },
|
||||
{
|
||||
role: "system",
|
||||
content: request.systemPrompt || "你是一个智能助手，可以分析用户发送的文字，文件或图片内容，并进行回答。",
|
||||
},
|
||||
{
|
||||
role: "user",
|
||||
content: userContent
|
||||
}
|
||||
content: userContent,
|
||||
},
|
||||
],
|
||||
stream: true,
|
||||
temperature: request.temperature,
|
||||
max_tokens: request.maxTokens
|
||||
max_tokens: request.maxTokens,
|
||||
files: request.files || [], // 传递文件 URL 列表给后端
|
||||
};
|
||||
|
||||
const response = await fetch(
|
||||
|
|
@ -190,14 +192,12 @@ class ChatApi {
|
|||
let userContent;
|
||||
if (request.images && request.images.length > 0) {
|
||||
// 如果有图片，则构建内容数组
|
||||
userContent = [
|
||||
{ type: "text", text: request.message }
|
||||
];
|
||||
userContent = [{ type: "text", text: request.message }];
|
||||
// 添加图片URL到内容中
|
||||
request.images.forEach(imageUrl => {
|
||||
request.images.forEach((imageUrl) => {
|
||||
userContent.push({
|
||||
type: "image_url",
|
||||
image_url: { url: imageUrl }
|
||||
image_url: { url: imageUrl },
|
||||
});
|
||||
});
|
||||
} else {
|
||||
|
|
@ -207,7 +207,7 @@ class ChatApi {
|
|||
|
||||
const requestBody = {
|
||||
...request,
|
||||
message: userContent
|
||||
message: userContent,
|
||||
};
|
||||
|
||||
const response = await fetch(`${this.baseUrl}${API_ENDPOINTS.CHAT}`, {
|
||||
|
|
|
|||
|
|
@ -26,6 +26,7 @@ export interface Attachment {
|
|||
size?: number;
|
||||
mimeType?: string;
|
||||
thumbnail?: string;
|
||||
uploading?: boolean; // 标记附件是否正在上传中
|
||||
}
|
||||
|
||||
// 推荐选项
|
||||
|
|
|
|||
Loading…
Reference in New Issue