127 lines
5.0 KiB
Python
127 lines
5.0 KiB
Python
"""视觉语言模型客户端（主智能体专用）。"""
|
||
|
||
import base64
|
||
import mimetypes
|
||
from pathlib import Path
|
||
from typing import Dict, List
|
||
|
||
import httpx
|
||
from openai import OpenAI
|
||
|
||
from config import OCR_API_BASE_URL, OCR_API_KEY, OCR_MODEL_ID, OCR_MAX_TOKENS
|
||
from modules.file_manager import FileManager
|
||
|
||
|
||
class OCRClient:
|
||
"""封装外部 VLM 调用逻辑。"""
|
||
|
||
def __init__(self, project_path: str, file_manager: FileManager):
|
||
self.project_path = Path(project_path).resolve()
|
||
self.file_manager = file_manager
|
||
|
||
# 懒加载：httpx.Client() 构造会加载 CA 证书并初始化 SSL 上下文，
|
||
# 在 Windows（Defender 实时扫描）下可耗时数秒；每个对话级 terminal
|
||
# 都会构造 OCRClient，因此推迟到首次真正调用 VLM 时再创建。
|
||
self.http_client = None
|
||
self.client = None
|
||
self._client_ready = False
|
||
self.model = OCR_MODEL_ID
|
||
self.max_tokens = OCR_MAX_TOKENS or 4096
|
||
|
||
# 默认大小上限（10MB），超出则警告并拒绝
|
||
self.max_image_size = 10 * 1024 * 1024
|
||
|
||
def _ensure_client(self):
|
||
"""首次使用时创建 httpx / OpenAI 客户端（线程安全由 GIL 保证最坏情况重复创建一次）。"""
|
||
if self._client_ready:
|
||
return
|
||
# 补全 base_url，兼容是否包含 /v1
|
||
base_url = (OCR_API_BASE_URL or "").rstrip("/")
|
||
if base_url and not base_url.endswith("/v1"):
|
||
base_url = f"{base_url}/v1"
|
||
|
||
# httpx 0.28 起不再支持 proxies 参数，显式传入 http_client 以避免默认封装报错
|
||
self.http_client = httpx.Client()
|
||
if OCR_API_KEY:
|
||
self.client = OpenAI(
|
||
api_key=OCR_API_KEY,
|
||
base_url=base_url or None,
|
||
http_client=self.http_client,
|
||
)
|
||
self._client_ready = True
|
||
|
||
def _validate_image_path(self, path: str):
|
||
"""复用 FileManager 的路径校验，确保在项目内。"""
|
||
valid, error, full_path = self.file_manager._validate_path(path)
|
||
if not valid:
|
||
return False, error, None
|
||
if not full_path.exists():
|
||
return False, "文件不存在", None
|
||
if not full_path.is_file():
|
||
return False, "不是文件", None
|
||
return True, "", full_path
|
||
|
||
def vlm_analyze(self, path: str, prompt: str) -> Dict:
|
||
"""使用大参数视觉语言模型分析图片：文字、物体、布局等。"""
|
||
warnings: List[str] = []
|
||
|
||
valid, error, full_path = self._validate_image_path(path)
|
||
if not valid:
|
||
return {"success": False, "error": error, "warnings": warnings}
|
||
|
||
if not prompt or not str(prompt).strip():
|
||
return {"success": False, "error": "prompt 不能为空", "warnings": warnings}
|
||
if not OCR_API_KEY or not OCR_API_BASE_URL or not self.model:
|
||
return {"success": False, "error": "VLM 配置缺失，请设置 OCR_API_BASE_URL / OCR_API_KEY / OCR_MODEL_ID", "warnings": warnings}
|
||
self._ensure_client()
|
||
if not self.client:
|
||
return {"success": False, "error": "VLM 客户端初始化失败", "warnings": warnings}
|
||
|
||
try:
|
||
data = full_path.read_bytes()
|
||
except Exception as exc:
|
||
return {"success": False, "error": f"读取文件失败: {exc}", "warnings": warnings}
|
||
|
||
size = len(data)
|
||
if size <= 0:
|
||
return {"success": False, "error": "文件为空，无法识别", "warnings": warnings}
|
||
|
||
if size > self.max_image_size:
|
||
return {
|
||
"success": False,
|
||
"error": f"图片过大({size}字节)，上限为{self.max_image_size}字节",
|
||
"warnings": warnings,
|
||
}
|
||
|
||
mime_type, _ = mimetypes.guess_type(str(full_path))
|
||
if not mime_type or not mime_type.startswith("image/"):
|
||
warnings.append("无法确定图片类型，已按 JPEG 处理")
|
||
mime_type = "image/jpeg"
|
||
|
||
base64_image = base64.b64encode(data).decode("utf-8")
|
||
data_url = f"data:{mime_type};base64,{base64_image}"
|
||
|
||
try:
|
||
response = self.client.chat.completions.create(
|
||
model=self.model,
|
||
messages=[
|
||
{
|
||
"role": "user",
|
||
"content": [
|
||
{"type": "image_url", "image_url": {"url": data_url}},
|
||
{"type": "text", "text": prompt},
|
||
],
|
||
}
|
||
],
|
||
max_tokens=self.max_tokens,
|
||
temperature=0,
|
||
)
|
||
content = response.choices[0].message.content if response.choices else ""
|
||
return {"success": True, "content": content or "", "warnings": warnings}
|
||
except Exception as exc:
|
||
return {"success": False, "error": f"VLM 调用失败: {exc}", "warnings": warnings}
|
||
|
||
def ocr_image(self, path: str, prompt: str) -> Dict:
|
||
"""兼容旧名，转发到 vlm_analyze。"""
|
||
return self.vlm_analyze(path, prompt)
|