Download agy2api/src/converter/utils.py from a3216/ds2api: direct link, hf CLI and curl.
- Browser
- Download file 5.89 kB
-
https://huggingface.co/spaces/a3216/ds2api/resolve/main/agy2api/src/converter/utils.py
- Command line
-
hf download hf://spaces/a3216/ds2api/agy2api/src/converter/utils.py
-
curl -L -o utils.py https://huggingface.co/spaces/a3216/ds2api/resolve/main/agy2api/src/converter/utils.py
5.89 kB
| from typing import Any, Dict | |
| from src.converter.thoughtSignature_fix import ( | |
| is_internal_placeholder_text, | |
| is_skip_thought_signature_placeholder, | |
| ) | |
| def extract_content_and_reasoning(parts: list) -> tuple: | |
| """从Gemini响应部件中提取内容和推理内容 | |
| Args: | |
| parts: Gemini 响应中的 parts 列表 | |
| Returns: | |
| (content, reasoning_content, images): 文本内容、推理内容和图片数据的元组 | |
| - content: 文本内容字符串 | |
| - reasoning_content: 推理内容字符串 | |
| - images: 图片数据列表,每个元素格式为: | |
| { | |
| "type": "image_url", | |
| "image_url": { | |
| "url": "data:{mime_type};base64,{base64_data}" | |
| } | |
| } | |
| """ | |
| content = "" | |
| reasoning_content = "" | |
| images = [] | |
| for part in parts: | |
| if is_skip_thought_signature_placeholder(part): | |
| continue | |
| # 提取文本内容 | |
| text = part.get("text", "") | |
| if is_internal_placeholder_text(text): | |
| continue | |
| if text: | |
| if part.get("thought", False): | |
| reasoning_content += text | |
| else: | |
| content += text | |
| # 提取图片数据 | |
| if "inlineData" in part: | |
| inline_data = part["inlineData"] | |
| mime_type = inline_data.get("mimeType", "image/png") | |
| base64_data = inline_data.get("data", "") | |
| images.append({ | |
| "type": "image_url", | |
| "image_url": { | |
| "url": f"data:{mime_type};base64,{base64_data}" | |
| } | |
| }) | |
| return content, reasoning_content, images | |
| async def merge_system_messages(request_body: Dict[str, Any]) -> Dict[str, Any]: | |
| """ | |
| 处理请求体中的 system 消息: | |
| - 顶层 system 参数(Anthropic 格式)与开头的连续 system 消息合并为 systemInstruction | |
| - 中间的 system 消息转换为 user 消息(保持顺序) | |
| Args: | |
| request_body: OpenAI 或 Claude 格式的请求体,包含 messages 字段 | |
| Returns: | |
| 处理后的请求体 | |
| """ | |
| # 处理 Anthropic 格式的顶层 system 参数 | |
| # Anthropic API 规范: system 是顶层参数,不在 messages 中 | |
| system_content = request_body.get("system") | |
| if system_content: | |
| system_parts = [] | |
| if isinstance(system_content, str): | |
| if system_content.strip(): | |
| system_parts.append({"text": system_content}) | |
| elif isinstance(system_content, list): | |
| # system 可以是包含多个块的列表 | |
| for item in system_content: | |
| if isinstance(item, dict): | |
| if item.get("type") == "text" and item.get("text", "").strip(): | |
| system_parts.append({"text": item["text"]}) | |
| elif isinstance(item, str) and item.strip(): | |
| system_parts.append({"text": item}) | |
| if system_parts: | |
| request_body = request_body.copy() | |
| request_body.pop("system", None) | |
| request_body["systemInstruction"] = {"parts": system_parts} | |
| messages = request_body.get("messages", []) | |
| if not messages: | |
| return request_body | |
| # 提取开头的连续 system 消息合并为 systemInstruction | |
| system_parts = [] | |
| # 如果已经从顶层 system 参数创建了 systemInstruction,获取现有的 parts | |
| if "systemInstruction" in request_body: | |
| existing_instruction = request_body.get("systemInstruction", {}) | |
| if isinstance(existing_instruction, dict): | |
| system_parts = existing_instruction.get("parts", []).copy() | |
| remaining_messages = [] | |
| collecting_system = True | |
| # developer 角色与 system 同语义处理(OpenAI 新版指令角色),上游仅支持 user/model | |
| instruction_roles = ("system", "developer") | |
| for message in messages: | |
| role = message.get("role", "") | |
| content = message.get("content", "") | |
| if role in instruction_roles and collecting_system: | |
| # 提取 system 消息的文本内容 | |
| if isinstance(content, str): | |
| if content.strip(): | |
| system_parts.append({"text": content}) | |
| elif isinstance(content, list): | |
| # 处理列表格式的 content | |
| for item in content: | |
| if isinstance(item, dict): | |
| if item.get("type") == "text" and item.get("text", "").strip(): | |
| system_parts.append({"text": item["text"]}) | |
| elif isinstance(item, str) and item.strip(): | |
| system_parts.append({"text": item}) | |
| else: | |
| # 遇到非 system 消息,停止收集 | |
| collecting_system = False | |
| if role in instruction_roles: | |
| # 将后续的 system 消息转换为 user 消息 | |
| converted_message = message.copy() | |
| converted_message["role"] = "user" | |
| remaining_messages.append(converted_message) | |
| else: | |
| remaining_messages.append(message) | |
| # 如果没有收集到任何 systemInstruction 内容: | |
| # 仅当 messages 未发生变化时才返回原始请求体; | |
| # 存在被转换为 user 的中间 system 消息时,返回更新后的 messages | |
| if not system_parts: | |
| if remaining_messages == messages: | |
| return request_body | |
| result = request_body.copy() | |
| result["messages"] = remaining_messages | |
| return result | |
| # 构建新的请求体 | |
| result = request_body.copy() | |
| # 添加或更新 systemInstruction | |
| result["systemInstruction"] = {"parts": system_parts} | |
| # 更新 messages 列表(移除已处理的 system 消息) | |
| result["messages"] = remaining_messages | |
| return result | |