dekun
1cb3c7fad5
系统设置可关闭成绩复盘 AI；学生详情增加作文区（OCR/手动题目、方案与范文、历史与 MD 下载）；导出改用 UTF-8 文件名响应。 Co-authored-by: Cursor <cursoragent@cursor.com>
386 lines
14 KiB
Python
386 lines
14 KiB
Python
import httpx
|
||
from sqlalchemy.orm import Session
|
||
|
||
from app.core.config import settings as app_settings
|
||
from app.models.user import SchoolLevel, SystemSettings
|
||
from app.services.school_level import school_level_label
|
||
from app.services.url_sanitize import sanitize_http_url, sanitize_model_name
|
||
|
||
CURRICULUM_JUNIOR = """初中课程标准：代数、几何（全等/相似/勾股）、一次函数与简单二次函数、基础概率统计。
|
||
严禁使用：高中导数、向量、解析几何、排列组合进阶、复数、微积分、大学线性代数等。"""
|
||
|
||
CURRICULUM_SENIOR = """高中课程标准：课内函数、三角、向量、解析几何、概率统计、导数（课内范围）等。
|
||
严禁使用：大学数学分析、抽象代数、高等几何、超出课内的竞赛高阶技巧。"""
|
||
|
||
CURRICULUM_JUNIOR_OLYMPIAD = """初中奥数培优范围：整数/整除、因数分解、简单数论、代数恒等变形、几何辅助线与全等相似、简单组合计数。
|
||
严禁使用：高中及以上方法（导数、向量、解析几何、微积分、复数运算等）。"""
|
||
|
||
CURRICULUM_SENIOR_OLYMPIAD = """高中奥数/竞赛入门范围：课内知识+常规竞赛技巧（不等式、构造、归纳、简单数论等）。
|
||
严禁使用：大学数学、超出高中奥数培优体系的 IMO 高阶理论。"""
|
||
|
||
|
||
def _curriculum_block(level: SchoolLevel | str | None, olympiad: bool) -> str:
|
||
label = school_level_label(level)
|
||
is_senior = level == SchoolLevel.senior_high or level == "senior_high"
|
||
if olympiad:
|
||
return CURRICULUM_SENIOR_OLYMPIAD if is_senior else CURRICULUM_JUNIOR_OLYMPIAD
|
||
return CURRICULUM_SENIOR if is_senior else CURRICULUM_JUNIOR
|
||
|
||
|
||
QUESTION_PROMPT = """你是一位{stage}老师。以下是从试卷 OCR 识别出的文字，可能含有噪声。
|
||
科目：{subject}
|
||
请整理出清晰的题目内容（保留题号、选项、公式），只输出题目正文，不要解释。
|
||
|
||
OCR 原文：
|
||
{ocr_text}
|
||
"""
|
||
|
||
SOLUTION_PROMPT = """你是一位耐心的{stage}{subject}老师。请像「作业帮」一样，先讲清楚解题思路，再给出完整解答。
|
||
|
||
【学段要求 — 严禁超纲】
|
||
{curriculum}
|
||
|
||
题目：
|
||
{question_text}
|
||
|
||
请严格按以下 Markdown 结构输出：
|
||
|
||
## 解题思路
|
||
（2-5 句话：这题考什么、从哪里入手、关键一步是什么，让学生先懂「怎么想」）
|
||
|
||
## 详细解答
|
||
（分步骤完整推导，每步说明依据）
|
||
|
||
## 易错点
|
||
（指出常见错误及正确做法）
|
||
|
||
严禁使用超纲方法；若原题超纲，请给出{stage}课内可理解的解法。
|
||
"""
|
||
|
||
OLYMPIAD_SOLUTION_PROMPT = """你是一位{stage}奥数教练。请像优秀辅导老师一样，先讲解题思路，再完整解答。
|
||
|
||
【奥数学段要求 — 严禁超纲】
|
||
{curriculum}
|
||
|
||
题目：
|
||
{question_text}
|
||
|
||
请严格按以下 Markdown 结构输出：
|
||
|
||
## 解题思路
|
||
（点明题型、突破口、{stage}奥数常用技巧）
|
||
|
||
## 详细解答
|
||
（完整步骤）
|
||
|
||
## 关键技巧
|
||
（总结，仅限{stage}奥数范围）
|
||
|
||
严禁超纲；过难题给出{stage}可接受的培优思路。
|
||
"""
|
||
|
||
ERROR_DETECT_PROMPT = """你是{stage}{subject}老师。以下是试卷/作业 OCR 识别结果，每行前有编号。
|
||
请找出「学生答错的部分」：错误答案、被打叉的作答、明显不正确的计算结果等。
|
||
|
||
{numbered_lines}
|
||
|
||
只输出 JSON，不要其他文字：
|
||
{{"wrong_line_ids": [行编号整数列表]}}
|
||
若整张图就是一道错题，请标注含有错误答案或作答的行；找不到则标注最后作答行。
|
||
"""
|
||
|
||
REVIEW_INSIGHT_PROMPT = """你是一位{stage}{subject}学习顾问。请仅根据下方「复盘数据」做分析，不得编造未出现的考试或状态。
|
||
|
||
【学段】{stage}
|
||
【科目】{subject}（所有建议必须贴合本科目，禁止套用其他科目的说法）
|
||
|
||
【复盘数据】
|
||
{review_records}
|
||
|
||
【状态含义（结合本科目理解）】
|
||
- 粗心：{careless_hint}
|
||
- 不会：该科知识点/题型尚未掌握
|
||
- 紧张：心态影响，发挥低于平时水平
|
||
- 正常发挥：状态稳定
|
||
|
||
【科目建议方向】
|
||
{subject_hints}
|
||
|
||
【必须遵守】
|
||
1. 解读时必须写清具体考试日期（如 2026-06-21），按时间从早到晚分析，不得把「第1次」说成最近一场
|
||
2. 得分率 = 得分÷总分；95% 以上才可称「接近满分」，85% 左右应如实描述为「良好但仍有失分空间」，禁止夸大
|
||
3. 改进建议必须针对 {subject}，禁止出现与本科目无关的表述（如英语科禁止写「计算验算」）
|
||
4. 只分析数据中列出的复盘状态，不要臆测未勾选的原因
|
||
|
||
请用 Markdown 输出：
|
||
|
||
## 情况解读
|
||
（2-4 句：按时间顺序说明每次考试得分率、失分与复盘状态的关系，以及是否有改善或反复）
|
||
|
||
## 改进建议
|
||
（3-5 条，针对出现最多的问题状态，具体可操作，仅限 {stage}{subject} 范围）
|
||
|
||
## 近期重点
|
||
（1-2 条本周可落实的小目标）
|
||
|
||
语气务实，不要空泛鸡汤。
|
||
"""
|
||
|
||
SUBJECT_REVIEW_HINTS: dict[str, dict[str, str]] = {
|
||
"语文": {
|
||
"careless": "看错题干、漏读要求、作文偏题或漏写要点",
|
||
"hints": "阅读审题、文言文/语言运用、作文结构与素材积累",
|
||
},
|
||
"数学": {
|
||
"careless": "审题不清、计算或抄错、步骤跳步",
|
||
"hints": "错题归类、计算验算、典型题型归纳与限时练习",
|
||
},
|
||
"英语": {
|
||
"careless": "看错词义/时态、漏读题干、拼写与语法笔误",
|
||
"hints": "词汇语法、阅读完形、听力与写作模板，禁止建议计算类训练",
|
||
},
|
||
"物理": {
|
||
"careless": "审题漏条件、公式代错、单位换算失误",
|
||
"hints": "概念理解、建模分析、实验题与计算规范",
|
||
},
|
||
"化学": {
|
||
"careless": "方程式配平/条件遗漏、计算失误",
|
||
"hints": "方程式、物质性质、实验与推断题",
|
||
},
|
||
"生物": {
|
||
"careless": "概念混淆、漏答得分点",
|
||
"hints": "教材概念、图表分析、实验设计表述",
|
||
},
|
||
"历史": {
|
||
"careless": "材料题漏读、时间/人物混淆",
|
||
"hints": "时间线、材料分析、论述题答题模板",
|
||
},
|
||
"地理": {
|
||
"careless": "读图漏信息、术语使用不当",
|
||
"hints": "地图判读、区域分析、综合题答题条理",
|
||
},
|
||
"政治": {
|
||
"careless": "漏答采分点、概念表述不准",
|
||
"hints": "时政结合、材料分析、观点表述规范",
|
||
},
|
||
}
|
||
|
||
|
||
def _subject_review_hints(subject: str) -> tuple[str, str]:
|
||
block = SUBJECT_REVIEW_HINTS.get(subject)
|
||
if block:
|
||
return block["careless"], block["hints"]
|
||
return "审题或作答细节失误", f"针对{subject}常见失分点制定练习与错题巩固"
|
||
|
||
class AIConfig:
|
||
def __init__(
|
||
self,
|
||
provider: str,
|
||
ollama_base_url: str,
|
||
ollama_model: str,
|
||
openai_base_url: str,
|
||
openai_model: str,
|
||
openai_api_key: str | None,
|
||
):
|
||
self.provider = provider
|
||
self.ollama_base_url = ollama_base_url
|
||
self.ollama_model = ollama_model
|
||
self.openai_base_url = openai_base_url
|
||
self.openai_model = openai_model
|
||
self.openai_api_key = openai_api_key
|
||
|
||
|
||
def load_ai_config(db: Session) -> AIConfig:
|
||
row = db.get(SystemSettings, 1)
|
||
if row is None:
|
||
return AIConfig(
|
||
provider="ollama",
|
||
ollama_base_url=sanitize_http_url(app_settings.OLLAMA_BASE_URL),
|
||
ollama_model=sanitize_model_name(app_settings.OLLAMA_MODEL),
|
||
openai_base_url=sanitize_http_url(app_settings.OPENAI_BASE_URL),
|
||
openai_model=sanitize_model_name(app_settings.OPENAI_MODEL),
|
||
openai_api_key=None,
|
||
)
|
||
return AIConfig(
|
||
provider=row.ai_provider or "ollama",
|
||
ollama_base_url=sanitize_http_url(row.ollama_base_url or app_settings.OLLAMA_BASE_URL),
|
||
ollama_model=sanitize_model_name(row.ollama_model or app_settings.OLLAMA_MODEL),
|
||
openai_base_url=sanitize_http_url(row.openai_base_url or app_settings.OPENAI_BASE_URL),
|
||
openai_model=sanitize_model_name(row.openai_model or app_settings.OPENAI_MODEL),
|
||
openai_api_key=row.openai_api_key,
|
||
)
|
||
|
||
|
||
async def _ollama_generate(prompt: str, cfg: AIConfig, *, temperature: float = 0.3) -> str:
|
||
url = f"{cfg.ollama_base_url.rstrip('/')}/api/generate"
|
||
payload = {
|
||
"model": cfg.ollama_model,
|
||
"prompt": prompt,
|
||
"stream": False,
|
||
"options": {"temperature": temperature},
|
||
}
|
||
async with httpx.AsyncClient(timeout=180.0) as client:
|
||
response = await client.post(url, json=payload)
|
||
response.raise_for_status()
|
||
return (response.json().get("response") or "").strip()
|
||
|
||
|
||
async def _openai_generate(prompt: str, cfg: AIConfig, *, temperature: float = 0.3) -> str:
|
||
if not cfg.openai_api_key:
|
||
raise ValueError("未配置 OpenAI API Key")
|
||
url = f"{cfg.openai_base_url.rstrip('/')}/chat/completions"
|
||
headers = {"Authorization": f"Bearer {cfg.openai_api_key}"}
|
||
payload = {
|
||
"model": cfg.openai_model,
|
||
"messages": [{"role": "user", "content": prompt}],
|
||
"temperature": temperature,
|
||
}
|
||
async with httpx.AsyncClient(timeout=180.0) as client:
|
||
response = await client.post(url, json=payload, headers=headers)
|
||
response.raise_for_status()
|
||
data = response.json()
|
||
return (data["choices"][0]["message"]["content"] or "").strip()
|
||
|
||
|
||
async def generate_text(prompt: str, cfg: AIConfig, *, temperature: float = 0.3) -> str:
|
||
if cfg.provider == "openai":
|
||
return await _openai_generate(prompt, cfg, temperature=temperature)
|
||
return await _ollama_generate(prompt, cfg, temperature=temperature)
|
||
|
||
|
||
async def format_question(
|
||
cfg: AIConfig,
|
||
subject: str,
|
||
ocr_text: str,
|
||
school_level=None,
|
||
) -> str:
|
||
stage = school_level_label(school_level)
|
||
prompt = QUESTION_PROMPT.format(stage=stage, subject=subject, ocr_text=ocr_text)
|
||
return await generate_text(prompt, cfg)
|
||
|
||
|
||
async def generate_solution(
|
||
cfg: AIConfig,
|
||
subject: str,
|
||
question_text: str,
|
||
school_level=None,
|
||
*,
|
||
olympiad: bool = False,
|
||
) -> str:
|
||
stage = school_level_label(school_level)
|
||
curriculum = _curriculum_block(school_level, olympiad)
|
||
template = OLYMPIAD_SOLUTION_PROMPT if olympiad else SOLUTION_PROMPT
|
||
prompt = template.format(
|
||
stage=stage,
|
||
subject=subject,
|
||
curriculum=curriculum,
|
||
question_text=question_text,
|
||
)
|
||
return await generate_text(prompt, cfg)
|
||
|
||
|
||
async def detect_wrong_line_ids(
|
||
cfg: AIConfig,
|
||
subject: str,
|
||
ocr_lines: list[dict],
|
||
school_level=None,
|
||
) -> str:
|
||
stage = school_level_label(school_level)
|
||
numbered = "\n".join(f"[{i}] {line.get('text', '')}" for i, line in enumerate(ocr_lines))
|
||
prompt = ERROR_DETECT_PROMPT.format(stage=stage, subject=subject, numbered_lines=numbered)
|
||
return await generate_text(prompt, cfg)
|
||
|
||
|
||
async def generate_review_insight(
|
||
cfg: AIConfig,
|
||
subject: str,
|
||
review_records: str,
|
||
school_level=None,
|
||
) -> str:
|
||
stage = school_level_label(school_level)
|
||
careless_hint, subject_hints = _subject_review_hints(subject)
|
||
prompt = REVIEW_INSIGHT_PROMPT.format(
|
||
stage=stage,
|
||
subject=subject,
|
||
review_records=review_records,
|
||
careless_hint=careless_hint,
|
||
subject_hints=subject_hints,
|
||
)
|
||
return await generate_text(prompt, cfg, temperature=0.2)
|
||
|
||
|
||
CURRICULUM_CHINESE_JUNIOR = """初中作文：记叙文、写人记事、简单议论文为主，通常 600-800 字。
|
||
语言平实，素材来自课内与日常生活，禁止成人化腔调与超纲典故堆砌。"""
|
||
|
||
CURRICULUM_CHINESE_SENIOR = """高中作文：记叙、议论、材料作文为主，通常 800-1000 字。
|
||
可适度展开论证，仍须符合课内要求，禁止大学论文式写法与超纲理论。"""
|
||
|
||
COMPOSITION_PROMPT = """你是一位{stage}语文老师，正在辅导{grade_text}学生完成作文。
|
||
|
||
【学段年级 — 严禁超纲】
|
||
{curriculum}
|
||
|
||
作文题目：
|
||
{topic}
|
||
|
||
请严格按以下 Markdown 结构输出（不要增加其他一级标题）：
|
||
|
||
## 写作方案
|
||
（审题、立意、结构提纲、段落安排、可用素材方向，分条列出，贴合{grade_text}水平）
|
||
|
||
## 范文
|
||
（完整作文一篇，字数与语言风格必须符合{grade_text}课内要求，禁止超纲）
|
||
|
||
注意：范文必须是可直接参考的学生习作水准，不要写成评论或教案。
|
||
"""
|
||
|
||
|
||
def _chinese_curriculum(level, grade: str | None) -> str:
|
||
is_senior = level == SchoolLevel.senior_high or level == "senior_high"
|
||
return CURRICULUM_CHINESE_SENIOR if is_senior else CURRICULUM_CHINESE_JUNIOR
|
||
|
||
|
||
def _grade_text(grade: str | None) -> str:
|
||
if grade and grade.strip():
|
||
return grade.strip()
|
||
return "该学段学生"
|
||
|
||
|
||
def split_composition_sections(text: str) -> tuple[str, str]:
|
||
import re
|
||
|
||
text = text.strip()
|
||
if "## 范文" not in text:
|
||
return text.replace("## 写作方案", "").strip(), ""
|
||
parts = re.split(r"\n##\s*范文\s*\n", text, maxsplit=1)
|
||
plan = parts[0].replace("## 写作方案", "").strip()
|
||
essay = parts[1].strip() if len(parts) > 1 else ""
|
||
return plan, essay
|
||
|
||
|
||
async def generate_composition(
|
||
cfg: AIConfig,
|
||
topic: str,
|
||
school_level=None,
|
||
grade: str | None = None,
|
||
) -> tuple[str, str]:
|
||
stage = school_level_label(school_level)
|
||
grade_text = _grade_text(grade)
|
||
curriculum = _chinese_curriculum(school_level, grade)
|
||
prompt = COMPOSITION_PROMPT.format(
|
||
stage=stage,
|
||
grade_text=grade_text,
|
||
curriculum=curriculum,
|
||
topic=topic.strip(),
|
||
)
|
||
full = await generate_text(prompt, cfg, temperature=0.35)
|
||
return split_composition_sections(full)
|
||
|
||
|
||
def composition_markdown(topic: str, writing_plan: str | None, sample_essay: str | None) -> str:
|
||
parts = [f"# 作文题目\n\n{topic.strip()}", ""]
|
||
if writing_plan:
|
||
parts.extend(["## 写作方案", "", writing_plan.strip(), ""])
|
||
if sample_essay:
|
||
parts.extend(["## 范文", "", sample_essay.strip(), ""])
|
||
return "\n".join(parts).strip() + "\n" |