Merge pull request 'feat: 提交作业2024103833' (#1) from zhxinyue/zhxin202509201409272:zhangxinyue2024103833 into master

Reviewed-on: https://openi.pcl.ac.cn/info-ruc/middle25projects/pulls/1

zhangxinyue2024103833/.gitignore

@@ -0,0 +1,2 @@
+deepseek1.5B/
+__pycache__/

zhangxinyue2024103833/2024103833.md

@@ -0,0 +1,74 @@
+# 财务问答中间件（自然语言 → SQL → 结果）
+
+一句话说明：这是一个**把中文问题自动翻译成 SQL，查询 SQLite，再用人话解释结果**的小系统，“中间件 + 大模型”的结合。
+
+---
+
+## 🧠 原理简述
+1. 输入中文问题（例：8 月份市场部的费用是多少？）  
+2. 大模型生成对应的 SQL  
+3. SQLite 执行 SQL，得到结果  
+4. 后端把结果转成通俗解释  
+5. 网页展示：问题 → 解释 → SQL → 结果表  
+
+---
+
+## 📂 目录结构
+<pre>
+中间件/
+├── train.jsonl                微调样本（每行：问题→SQL 的 JSON）
+├── finetune.py                微调脚本：用 train.jsonl 对基础模型做 LoRA 微调
+├── middleware.py              后端中间件（FastAPI）：问句→SQL→查询→解释→返回给前端
+├── init_db.py                 初始化示例库：创建/写入演示用财务数据
+├── finance_demo.db            SQLite 数据库（运行 init_db.py 生成）
+│
+├── deepseek1.5B/              基础模型目录（你本地已有）
+├── models/
+│   └── deepseek-sql-lora/     微调产物（LoRA 适配器）
+│
+├── ui/                        前端网页
+│   ├── index.html             主页面（输入→显示解释/SQL/结果）
+│   ├── screenshot.png         演示截图（下方“演示案例”会引用）
+│   └── ...
+│
+└── logs/
+    └── middleware.log         运行日志
+    └── finetune.log           微调日志
+</pre>
+
+---
+
+## 🚀 使用步骤（通俗说明）
+
+### 1）初始化数据库  
+运行 `python init_db.py`  
+👉 生成或覆盖 `finance_demo.db`，写入示例财务数据（费用、部门、日期、预算等）。  
+
+### 2）微调小模型（可选）  
+运行 `python finetune.py`  
+👉 读取 `train.jsonl`，在 `deepseek1.5B/` 上做 LoRA 微调，结果存到 `models/deepseek-sql-lora/`。  
+
+### 3）启动后端中间件  
+运行 `python middleware.py`  
+👉 启动 FastAPI 服务，默认访问地址是 [http://127.0.0.1:8000/ui/](http://127.0.0.1:8000/ui/)。  
+
+### 4）打开前端页面演示  
+用浏览器访问 `http://127.0.0.1:8000/ui/`，输入问题或点击示例按钮，页面会显示：解释 → SQL → 查询结果。  
+推荐测试问题：  
+- 8月份市场部的费用是多少？  
+- 2024年8月各部门费用从高到低排序 
+
+---
+
+## 🎬 演示案例
+
+
+<img src="ui/screenshot.png" alt="演示页面" width="40%">
+
+---
+
+## ✨ 项目亮点
+- **大模型驱动**：利用大模型的强大语义理解和生成能力，把自然语言问题准确转化为 SQL，降低非技术人员使用数据库的门槛。  
+- **结构完整**：自然语言 → SQL → 数据库 → 解释，完整演示中间件与大模型结合；  
+- **可用与可维护**：前端可视化输出让业务人员能直接理解查询结果，后端日志记录则方便技术人员排查问题、进行审计，保证系统可靠运行。
+

zhangxinyue2024103833/finetune.py

@@ -0,0 +1,185 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+# 本脚本用于在CPU上对本地模型进行LoRA微调
+# 任务：把自然语言问题转换成SQLite的SQL查询
+
+import os
+import json
+import logging
+import argparse
+from typing import Dict
+
+import torch
+from datasets import load_dataset, Dataset
+from transformers import (
+    AutoTokenizer,
+    AutoModelForCausalLM,
+    DataCollatorForLanguageModeling,
+    Trainer,
+    TrainingArguments,
+    set_seed,
+)
+from peft import LoraConfig, get_peft_model
+
+# 默认路径（根据你的目录结构）
+DEFAULT_BASE_MODEL = r"D:/人大/结课作业/中间件/deepseek1.5B"
+DEFAULT_TRAIN_FILE = r"D:/人大/结课作业/中间件/train.jsonl"
+DEFAULT_OUTPUT_DIR = r"models/deepseek-sql-lora"
+
+# 日志
+os.makedirs("logs", exist_ok=True)
+logging.basicConfig(
+    filename="logs/finetune.log",
+    level=logging.INFO,
+    format="%(asctime)s %(levelname)s %(message)s",
+)
+logging.getLogger("").addHandler(logging.StreamHandler())
+
+# 系统提示
+SYS_PROMPT = "你是一个SQL助手，只输出SQLite的SELECT语句，不要其他内容。"
+
+def format_example(ex: Dict, eos_token: str) -> str:
+    """把一条样本转成训练用文本"""
+    inst = (ex.get("instruction") or "").strip()
+    _in = (ex.get("input") or "").strip()
+    out = (ex.get("output") or "").strip()
+    user_block = inst if not _in else f"{inst}\n用户问题：{_in}"
+    return (
+        f"<s>[SYSTEM]\n{SYS_PROMPT}\n[/SYSTEM]\n"
+        f"[USER]\n{user_block}\n[/USER]\n"
+        f"[ASSISTANT]\n{out}\n[/ASSISTANT]{eos_token}"
+    )
+
+def robust_load_jsonl(path: str) -> Dataset:
+    """读取jsonl文件，忽略坏行"""
+    rows = []
+    with open(path, "r", encoding="utf-8") as f:
+        for i, line in enumerate(f, 1):
+            s = line.strip()
+            if not s:
+                continue
+            try:
+                rows.append(json.loads(s))
+            except Exception as e:
+                logging.warning(f"跳过第{i}行：{e}")
+    if not rows:
+        raise RuntimeError("train.jsonl 里没有有效样本")
+    ds = Dataset.from_list(rows)
+    return ds.add_column("raw", [dict(x) for x in ds])
+
+def tokenize_function(batch, tokenizer, max_len):
+    """分词"""
+    texts = [format_example(x, tokenizer.eos_token or "</s>") for x in batch["raw"]]
+    return tokenizer(texts, max_length=max_len, truncation=True, padding=False)
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--base_model", type=str, default=DEFAULT_BASE_MODEL)
+    ap.add_argument("--train_file", type=str, default=DEFAULT_TRAIN_FILE)
+    ap.add_argument("--output_dir", type=str, default=DEFAULT_OUTPUT_DIR)
+    ap.add_argument("--eval_ratio", type=float, default=0.1)
+    ap.add_argument("--max_steps", type=int, default=200)
+    ap.add_argument("--per_device_train_batch_size", type=int, default=1)   # CPU友好
+    ap.add_argument("--per_device_eval_batch_size", type=int, default=1)
+    ap.add_argument("--gradient_accumulation_steps", type=int, default=1)   # CPU友好
+    ap.add_argument("--learning_rate", type=float, default=1e-4)
+    ap.add_argument("--lr_scheduler_type", type=str, default="cosine")
+    ap.add_argument("--warmup_ratio", type=float, default=0.05)
+    ap.add_argument("--lora_r", type=int, default=4)
+    ap.add_argument("--lora_alpha", type=int, default=8)
+    ap.add_argument("--lora_dropout", type=float, default=0.05)
+    ap.add_argument("--target_modules", type=str,
+                    default="q_proj,k_proj,v_proj,o_proj,gate_proj,up_proj,down_proj")
+    ap.add_argument("--max_seq_length", type=int, default=512)
+    ap.add_argument("--seed", type=int, default=42)
+    args = ap.parse_args()
+
+    set_seed(args.seed)
+
+    # 读数据
+    ds = robust_load_jsonl(args.train_file)
+    if 0 < args.eval_ratio < 1:
+        split = ds.train_test_split(test_size=args.eval_ratio, seed=args.seed)
+        train_ds, eval_ds = split["train"], split["test"]
+    else:
+        train_ds, eval_ds = ds, None
+
+    # 加载分词器和模型
+    tokenizer = AutoTokenizer.from_pretrained(args.base_model, use_fast=True)
+    if tokenizer.pad_token is None:
+        tokenizer.pad_token = tokenizer.eos_token
+
+    model = AutoModelForCausalLM.from_pretrained(
+        args.base_model,
+        device_map=None,
+        torch_dtype=torch.float32,
+    ).to("cpu")
+
+    if hasattr(model, "enable_input_require_grads"):
+        model.enable_input_require_grads()
+    if hasattr(model, "config"):
+        model.config.use_cache = False
+
+    # LoRA配置
+    target_modules = [t.strip() for t in args.target_modules.split(",") if t.strip()]
+    lora_cfg = LoraConfig(
+        r=args.lora_r,
+        lora_alpha=args.lora_alpha,
+        lora_dropout=args.lora_dropout,
+        bias="none",
+        task_type="CAUSAL_LM",
+        target_modules=target_modules,
+    )
+    model = get_peft_model(model, lora_cfg)
+    model.print_trainable_parameters()
+
+    # 数据tokenize
+    def _map(b): return tokenize_function(b, tokenizer, args.max_seq_length)
+    train_tok = train_ds.map(_map, batched=True, remove_columns=train_ds.column_names)
+    eval_tok = eval_ds.map(_map, batched=True, remove_columns=eval_ds.column_names) if eval_ds else None
+    collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)
+
+    # 训练参数（CPU）
+    train_kwargs = dict(
+        output_dir=args.output_dir,
+        learning_rate=args.learning_rate,
+        lr_scheduler_type=args.lr_scheduler_type,
+        warmup_ratio=args.warmup_ratio,
+        per_device_train_batch_size=args.per_device_train_batch_size,
+        per_device_eval_batch_size=args.per_device_eval_batch_size,
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        num_train_epochs=1,
+        max_steps=args.max_steps,
+        logging_steps=10,
+        save_steps=200,
+        save_total_limit=2,
+        bf16=False,
+        fp16=False,
+        gradient_checkpointing=False,
+        report_to=["none"],
+        remove_unused_columns=False,
+        dataloader_pin_memory=False,
+    )
+    training_args = TrainingArguments(**train_kwargs)
+
+    # Trainer
+    trainer = Trainer(
+        model=model,
+        args=training_args,
+        data_collator=collator,
+        train_dataset=train_tok,
+        eval_dataset=eval_tok,
+        tokenizer=tokenizer,
+    )
+    trainer.train()
+
+    # 保存
+    os.makedirs(args.output_dir, exist_ok=True)
+    trainer.model.save_pretrained(args.output_dir)
+    tokenizer.save_pretrained(args.output_dir)
+
+    print(f"[完成] 模型已保存到: {args.output_dir}")
+
+if __name__ == "__main__":
+    main()

zhangxinyue2024103833/init_sqlite.py

@@ -0,0 +1,190 @@
+import os
+import sqlite3
+import argparse
+from typing import List, Tuple
+
+DB_FILE = "finance_demo.db"
+
+DEPARTMENTS = ["市场部", "研发部", "行政部", "财务部", "销售部"]
+CATEGORIES  = ["广告", "人力", "办公", "差旅", "培训"]
+VENDORS = {
+    "广告": ["XX传媒", "广告联盟", "新媒体投放"],
+    "人力": ["外包A", "外包B", "人才中介"],
+    "办公": ["办公城", "文具行", "设备商"],
+    "差旅": ["航空公司", "铁路公司", "酒店集团"],
+    "培训": ["培训机构", "认证中心"],
+}
+
+EXPENSES_ROWS: List[Tuple[str, str, str, str, float]] = [
+    # 2024-07
+    ("市场部", "2024-07-05", "广告", "XX传媒", 15000),
+    ("研发部", "2024-07-15", "人力", "外包A", 25000),
+    ("行政部", "2024-07-08", "办公", "办公城", 5000),
+    ("财务部", "2024-07-22", "差旅", "航空公司", 12000),
+    ("销售部", "2024-07-18", "培训", "培训机构", 6000),
+
+    # 2024-08
+    ("市场部", "2024-08-10", "广告", "XX传媒", 18000),
+    ("研发部", "2024-08-20", "人力", "外包B", 30000),
+    ("行政部", "2024-08-12", "办公", "办公城", 8000),
+    ("财务部", "2024-08-25", "差旅", "铁路公司", 9000),
+    ("销售部", "2024-08-03", "培训", "培训机构", 7000),
+    ("销售部", "2024-08-18", "广告", "广告联盟", 11000),
+    ("市场部", "2024-08-22", "广告", "新媒体投放", 12000),  # 便于供应商聚合/TopN
+
+    # 2024-09
+    ("市场部", "2024-09-06", "广告", "XX传媒", 12000),
+    ("研发部", "2024-09-10", "人力", "外包A", 28000),
+    ("行政部", "2024-09-13", "办公", "文具行", 3000),
+    ("财务部", "2024-09-21", "差旅", "酒店集团", 6000),
+    ("销售部", "2024-09-02", "培训", "认证中心", 5000),
+    ("销售部", "2024-09-27", "广告", "广告联盟", 9000),
+
+    # 假期段 2024-10-01 ~ 2024-10-07（用于节假日区间查询示例）
+    ("市场部", "2024-10-01", "广告", "XX传媒", 5000),
+    ("研发部", "2024-10-03", "人力", "外包B", 7000),
+    ("行政部", "2024-10-05", "办公", "设备商", 4000),
+]
+
+# 月度预算
+BUDGETS_ROWS: List[Tuple[str, str, float]] = [
+    # 2024-07
+    ("市场部", "2024-07", 20000),
+    ("研发部", "2024-07", 26000),
+    ("行政部", "2024-07", 6000),
+    ("财务部", "2024-07", 15000),
+    ("销售部", "2024-07", 8000),
+    # 2024-08
+    ("市场部", "2024-08", 22000),
+    ("研发部", "2024-08", 31000),
+    ("行政部", "2024-08", 9000),
+    ("财务部", "2024-08", 10000),
+    ("销售部", "2024-08", 12000),
+    # 2024-09
+    ("市场部", "2024-09", 16000),
+    ("研发部", "2024-09", 29000),
+    ("行政部", "2024-09", 7000),
+    ("财务部", "2024-09", 8000),
+    ("销售部", "2024-09", 11000),
+]
+
+CREATE_EXPENSES_SQL = """
+CREATE TABLE IF NOT EXISTS expenses (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    department TEXT NOT NULL,
+    date TEXT NOT NULL,           -- YYYY-MM-DD
+    category TEXT NOT NULL,       -- 差旅/广告/人力/办公/培训 等
+    vendor TEXT NOT NULL,
+    amount REAL NOT NULL
+);
+"""
+
+CREATE_BUDGETS_SQL = """
+CREATE TABLE IF NOT EXISTS budgets (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    department TEXT NOT NULL,
+    month TEXT NOT NULL,          -- YYYY-MM
+    budget_amount REAL NOT NULL
+);
+"""
+
+def ensure_db(reset: bool) -> sqlite3.Connection:
+    """创建/重置数据库文件并建表。"""
+    if reset and os.path.exists(DB_FILE):
+        os.remove(DB_FILE)
+        print(f"已删除旧数据库: {DB_FILE}")
+
+    need_create = not os.path.exists(DB_FILE)
+    conn = sqlite3.connect(DB_FILE)
+    cur = conn.cursor()
+
+    if need_create:
+        print(f"创建新数据库: {DB_FILE}")
+
+    cur.execute(CREATE_EXPENSES_SQL)
+    cur.execute(CREATE_BUDGETS_SQL)
+    conn.commit()
+    return conn
+
+def table_count(cur: sqlite3.Cursor, table: str) -> int:
+    cur.execute(f"SELECT COUNT(*) FROM {table}")
+    return cur.fetchone()[0]
+
+def insert_if_empty(conn: sqlite3.Connection):
+    """仅在表为空时插入初始化数据；若已有数据则跳过。"""
+    cur = conn.cursor()
+    exp_cnt = table_count(cur, "expenses")
+    bud_cnt = table_count(cur, "budgets")
+
+    if exp_cnt == 0:
+        cur.executemany(
+            "INSERT INTO expenses(department,date,category,vendor,amount) VALUES(?,?,?,?,?)",
+            EXPENSES_ROWS
+        )
+        print(f"插入 expenses: {len(EXPENSES_ROWS)} 条")
+    else:
+        print(f"expenses 已有 {exp_cnt} 条，跳过插入。")
+
+    if bud_cnt == 0:
+        cur.executemany(
+            "INSERT INTO budgets(department,month,budget_amount) VALUES(?,?,?)",
+            BUDGETS_ROWS
+        )
+        print(f"插入 budgets: {len(BUDGETS_ROWS)} 条")
+    else:
+        print(f"budgets 已有 {bud_cnt} 条，跳过插入。")
+
+    conn.commit()
+
+def quick_selfcheck(conn: sqlite3.Connection):
+    """跑几条典型 SQL 做快速自检，与你的 Demo 问题一致。"""
+    cur = conn.cursor()
+
+    checks = [
+        ("检查 2024-08 市场部费用总额",
+         "SELECT IFNULL(SUM(amount),0) FROM expenses WHERE department='市场部' AND date LIKE '2024-08%';"),
+        ("检查 2024-08 各部门费用从高到低",
+         "SELECT department, IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%' GROUP BY department ORDER BY total DESC;"),
+        ("检查 2024-08 预算与实际对比",
+         "SELECT b.department, b.budget_amount, IFNULL(e.actual,0) AS actual, (IFNULL(e.actual,0)-b.budget_amount) AS diff "
+         "FROM budgets b LEFT JOIN (SELECT department, SUM(amount) AS actual FROM expenses WHERE date LIKE '2024-08%' GROUP BY department) e "
+         "ON b.department=e.department WHERE b.month='2024-08' ORDER BY diff DESC;"),
+        ("检查 2024-10-01 到 2024-10-07 区间记录",
+         "SELECT date, department, category, vendor, amount FROM expenses WHERE date BETWEEN '2024-10-01' AND '2024-10-07' ORDER BY date;"),
+    ]
+
+    print("\n==== 快速自检（部分查询结果预览） ====")
+    for title, sql in checks:
+        cur.execute(sql)
+        rows = cur.fetchall()
+        print(f"- {title}: {len(rows)} 行")
+        # 打印最多前 3 行
+        for r in rows[:3]:
+            print("  ", tuple(r))
+    print("==== 自检结束 ====\n")
+
+def parse_args():
+    ap = argparse.ArgumentParser(description="初始化 finance_demo.db")
+    g = ap.add_mutually_exclusive_group()
+    g.add_argument("--reset", action="store_true", help="若存在旧数据库则删除重建（默认）")
+    g.add_argument("--keep", action="store_true", help="若存在旧数据库则保留，仅在表为空时插入")
+    return ap.parse_args()
+
+def main():
+    args = parse_args()
+    reset = True
+    if args.keep:
+        reset = False
+
+    conn = ensure_db(reset=reset)
+    insert_if_empty(conn)
+    quick_selfcheck(conn)
+    conn.close()
+
+    print(f"✅ 初始化完成，数据库文件: {DB_FILE}")
+    print("下一步：")
+    print("  1) 运行  uvicorn middleware:app --reload")
+    print("  2) 打开  http://127.0.0.1:8000/ui/  进行演示")
+
+if __name__ == "__main__":
+    main()

zhangxinyue2024103833/logs/finetune.log

@@ -0,0 +1,56 @@
+2025-09-15 20:24:23,055 INFO Base model   : D:/人大/结课作业/中间件/deepseek1.5B
+2025-09-15 20:24:23,057 INFO Train file   : D:/人大/结课作业/中间件/train.jsonl
+2025-09-15 20:24:23,057 INFO Output dir   : models/deepseek-sql-lora
+2025-09-15 20:24:23,057 INFO Loading dataset ...
+2025-09-15 20:29:15,899 INFO Base model   : D:/人大/结课作业/中间件/deepseek1.5B
+2025-09-15 20:29:15,899 INFO Train file   : D:/人大/结课作业/中间件/train.jsonl
+2025-09-15 20:29:15,899 INFO Output dir   : models/deepseek-sql-lora
+2025-09-15 20:29:15,899 INFO Loading dataset ...
+2025-09-15 20:29:16,746 INFO Loading tokenizer ...
+2025-09-15 20:29:17,191 INFO Loading base model ...
+2025-09-15 20:29:29,793 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 20:29:31,261 INFO Trainable params (LoRA): 18.46 M
+2025-09-15 20:29:31,262 INFO Tokenizing dataset ...
+2025-09-15 20:29:31,892 INFO Setting TrainingArguments ...
+2025-09-15 20:35:55,947 INFO Base model   : D:/人大/结课作业/中间件/deepseek1.5B
+2025-09-15 20:35:55,948 INFO Train file   : D:/人大/结课作业/中间件/train.jsonl
+2025-09-15 20:35:55,948 INFO Output dir   : models/deepseek-sql-lora
+2025-09-15 20:35:55,948 INFO Loading dataset ...
+2025-09-15 20:35:56,700 INFO Loading tokenizer ...
+2025-09-15 20:35:57,126 INFO Loading base model ...
+2025-09-15 20:36:07,530 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 20:36:08,720 INFO Trainable params (LoRA): 18.46 M
+2025-09-15 20:36:08,720 INFO Tokenizing dataset ...
+2025-09-15 20:36:09,158 INFO Setting TrainingArguments ...
+2025-09-15 20:40:51,009 INFO Base model   : D:/人大/结课作业/中间件/deepseek1.5B
+2025-09-15 20:40:51,010 INFO Train file   : D:/人大/结课作业/中间件/train.jsonl
+2025-09-15 20:40:51,010 INFO Output dir   : models/deepseek-sql-lora
+2025-09-15 20:40:51,010 INFO Loading dataset ...
+2025-09-15 20:40:51,815 INFO Loading tokenizer ...
+2025-09-15 20:40:52,189 INFO Loading base model ...
+2025-09-15 20:41:01,229 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 20:41:02,830 INFO Trainable params (LoRA): 18.46 M
+2025-09-15 20:41:02,831 INFO Tokenizing dataset ...
+2025-09-15 20:41:03,189 INFO Setting TrainingArguments ...
+2025-09-15 20:43:26,543 INFO Base model   : D:/人大/结课作业/中间件/deepseek1.5B
+2025-09-15 20:43:26,544 INFO Train file   : D:/人大/结课作业/中间件/train.jsonl
+2025-09-15 20:43:26,544 INFO Output dir   : models/deepseek-sql-lora
+2025-09-15 20:43:26,544 INFO Loading dataset ...
+2025-09-15 20:43:27,630 INFO Loading tokenizer ...
+2025-09-15 20:43:28,663 INFO Loading base model ...
+2025-09-15 20:43:46,265 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 20:43:52,811 INFO Trainable params (LoRA): 18.46 M
+2025-09-15 20:43:52,811 INFO Tokenizing dataset ...
+2025-09-15 20:43:53,799 INFO Setting TrainingArguments ...
+2025-09-15 20:44:45,101 INFO Base model   : D:/人大/结课作业/中间件/deepseek1.5B
+2025-09-15 20:44:45,101 INFO Train file   : D:/人大/结课作业/中间件/train.jsonl
+2025-09-15 20:44:45,102 INFO Output dir   : models/deepseek-sql-lora
+2025-09-15 20:44:45,102 INFO Loading dataset ...
+2025-09-15 20:44:45,831 INFO Loading tokenizer ...
+2025-09-15 20:44:46,189 INFO Loading base model ...
+2025-09-15 20:44:58,632 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 20:45:00,345 INFO Trainable params (LoRA): 18.46 M
+2025-09-15 20:45:00,345 INFO Tokenizing dataset ...
+2025-09-15 20:45:00,761 INFO Setting TrainingArguments ...
+2025-09-15 20:45:00,786 INFO Start training ...
+2025-09-15 20:54:43,614 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.

zhangxinyue2024103833/logs/middleware.log

@@ -0,0 +1,60 @@
+2025-09-15 21:24:10,404 INFO 加载分词器与模型...
+2025-09-15 21:24:19,835 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 21:24:21,772 INFO 模型就绪，模式：lora
+2025-09-15 21:24:30,223 INFO 加载分词器与模型...
+2025-09-15 21:24:41,658 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 21:24:45,478 INFO 模型就绪，模式：lora
+2025-09-15 21:24:45,641 INFO 加载分词器与模型...
+2025-09-15 21:24:54,390 INFO 模型就绪，模式：lora
+2025-09-15 21:29:16,420 INFO 加载分词器与模型...
+2025-09-15 21:29:27,926 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 21:29:31,834 INFO 模型就绪，模式：lora
+2025-09-15 21:29:32,033 INFO 加载分词器与模型...
+2025-09-15 21:29:43,891 INFO 模型就绪，模式：lora
+2025-09-15 22:55:21,180 INFO 鍔犺浇鍒嗚瘝鍣ㄤ笌妯″瀷锛圕PU锛�...
+2025-09-15 22:55:32,511 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 22:55:34,631 INFO 妯″瀷灏辩华锛屾ā寮忥細lora
+2025-09-15 23:03:09,623 INFO 加载分词器与模型...
+2025-09-15 23:03:19,099 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 23:03:22,261 INFO 模型就绪，模式：lora
+2025-09-15 23:03:30,455 INFO 加载分词器与模型...
+2025-09-15 23:03:43,372 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 23:03:47,665 INFO 模型就绪，模式：lora
+2025-09-15 23:03:48,008 INFO 加载分词器与模型...
+2025-09-15 23:38:43,199 INFO 加载分词器与模型...
+2025-09-15 23:38:57,493 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-15 23:39:01,160 INFO 模型就绪，模式：lora
+2025-09-15 23:39:01,469 INFO 加载分词器与模型...
+2025-09-16 00:07:10,421 INFO 加载分词器与模型...
+2025-09-16 00:07:21,365 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-16 00:07:23,732 INFO 模型就绪，模式：lora
+2025-09-16 00:07:34,461 INFO 加载分词器与模型...
+2025-09-16 00:07:47,042 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-16 00:07:49,841 INFO 模型就绪，模式：lora
+2025-09-16 00:07:50,038 INFO 加载分词器与模型...
+2025-09-16 00:14:14,875 INFO 启动时加载分词器与模型（一次）...
+2025-09-16 00:14:26,581 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-16 00:14:29,926 INFO 模型就绪，模式：lora
+2025-09-16 00:23:11,134 INFO 启动时加载分词器与模型（一次）...
+2025-09-16 00:23:21,760 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-16 00:23:25,757 INFO 模型就绪，模式：lora
+2025-09-16 00:24:39,344 INFO [OK] 8月份市场部的费用是多少 | SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%'; | 1行 | 55483.7ms | lora
+2025-09-16 00:36:16,271 INFO 启动时加载分词器与模型（一次）...
+2025-09-16 00:36:26,659 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-16 00:36:29,581 INFO 模型就绪，模式：lora
+2025-09-16 00:37:47,562 INFO [OK] 8月份市场部的费用是多少？ | SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%'; | 1行 | 69119.7ms | lora
+2025-09-16 22:08:16,139 INFO [OK] 8月份市场部的费用是多少？ | SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%'; | 1行 | 85474.0ms | lora
+2025-09-16 22:13:43,585 INFO [OK] 8月份市场部的费用是多少？ | SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%'; | 1行 | 76405.4ms | lora
+2025-09-16 22:25:11,903 INFO 启动时加载分词器与模型（一次）...
+2025-09-16 22:25:22,578 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-16 22:25:25,524 INFO 模型就绪，模式：lora
+2025-09-16 22:26:39,689 INFO [OK] 8月份市场部的费用是多少？ | SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%'; | 1行 | 63840.8ms | lora
+2025-09-20 13:26:41,948 INFO 启动时加载分词器与模型（一次）...
+2025-09-20 13:27:10,782 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-20 13:27:14,219 INFO 模型就绪，模式：lora
+2025-09-20 13:29:03,984 INFO [OK] 8月份市场部的费用是多少？ | SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%'; | 1行 | 46152.7ms | lora
+2025-09-20 14:18:08,500 INFO 启动时加载分词器与模型（一次）...
+2025-09-20 14:18:43,174 INFO 启动时加载分词器与模型（一次）...
+2025-09-20 14:20:08,883 INFO 启动时加载分词器与模型（一次）...
+2025-09-20 14:20:21,555 WARNING The 8-bit optimizer is not available on your device, only available on CUDA for now.
+2025-09-20 14:20:27,136 INFO 模型就绪，模式：lora

zhangxinyue2024103833/middleware.py

@@ -0,0 +1,252 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+# 中间件：自然语言 -> 生成SQL -> 执行SQLite -> 返回结果（含结果导向解释）
+import os, re, time, logging, sqlite3
+from typing import Optional, List, Dict, Any
+from contextlib import asynccontextmanager
+
+import torch
+from fastapi import FastAPI
+from fastapi.responses import JSONResponse, HTMLResponse, RedirectResponse
+from fastapi.staticfiles import StaticFiles
+from pydantic import BaseModel
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from peft import PeftModel
+
+# ====== 路径与常量（按你的目录）======
+BASE_MODEL_PATH   = r"./deepseek1.5B"
+LORA_ADAPTER_PATH = r"models/deepseek-sql-lora"
+DB_PATH           = r"finance_demo.db"
+MAX_NEW_TOKENS    = 128
+RESULT_LIMIT      = 1000
+
+# ====== 日志 ======
+os.makedirs("logs", exist_ok=True)
+logging.basicConfig(
+    filename="logs/middleware.log",
+    level=logging.INFO,
+    format="%(asctime)s %(levelname)s %(message)s",
+)
+logging.getLogger("").addHandler(logging.StreamHandler())
+
+# ====== UI 静态目录（绝对路径，避免工作目录问题）======
+BASE_DIR = os.path.dirname(os.path.abspath(__file__))
+UI_DIR   = os.path.join(BASE_DIR, "ui")
+
+# ====== 全局模型句柄 ======
+tokenizer: Optional[AutoTokenizer] = None
+model:     Optional[torch.nn.Module] = None
+MODE = "base"
+model_ready = False
+
+# ====== SQL 安全与提示模板 ======
+SYS_PROMPT = "你是一个SQL助手，只输出可在SQLite执行的SELECT语句，不要解释。"
+PAT_ALLOW  = re.compile(r"^\s*SELECT\b", re.IGNORECASE)
+PAT_FORBID = re.compile(r"\b(UPDATE|DELETE|INSERT|DROP|ALTER|TRUNCATE|PRAGMA)\b", re.IGNORECASE)
+
+def build_prompt(q: str) -> str:
+    return f"<s>[SYSTEM]\n{SYS_PROMPT}\n[/SYSTEM]\n[USER]\n{q}\n[/USER]\n[ASSISTANT]\n"
+
+def extract_sql(text: str) -> str:
+    # 抓第一行SELECT；否则全局兜底
+    for ln in (x.strip() for x in text.splitlines() if x.strip()):
+        if PAT_ALLOW.search(ln) and not PAT_FORBID.search(ln):
+            return re.sub(r";+\s*$", ";", ln)
+    m = re.search(r"(SELECT.+)", text, flags=re.IGNORECASE | re.DOTALL)
+    if not m: return ""
+    chunk = m.group(1).strip()
+    semi = chunk.find(";")
+    if semi != -1: chunk = chunk[:semi+1]
+    return re.sub(r";+\s*$", ";", chunk)
+
+def generate_sql(question: str) -> str:
+    assert tokenizer is not None and model is not None
+    inputs = tokenizer(build_prompt(question), return_tensors="pt")
+    inputs = {k: v.to("cpu") for k, v in inputs.items()}
+    with torch.no_grad():
+        out = model.generate(
+            **inputs, max_new_tokens=MAX_NEW_TOKENS, do_sample=False,
+            eos_token_id=tokenizer.eos_token_id, pad_token_id=tokenizer.eos_token_id
+        )
+    text = tokenizer.decode(out[0], skip_special_tokens=True)
+    sql = extract_sql(text)
+    if not sql or not PAT_ALLOW.match(sql) or PAT_FORBID.search(sql):
+        return ""
+    return sql
+
+def run_sql(sql: str) -> List[Dict[str, Any]]:
+    q = sql.strip()
+    if not re.search(r"\bLIMIT\b", q, re.IGNORECASE):
+        q = q.rstrip(";") + f" LIMIT {RESULT_LIMIT};"
+    conn = sqlite3.connect(DB_PATH)
+    conn.row_factory = sqlite3.Row
+    cur = conn.cursor()
+    cur.execute(q)
+    rows = [dict(r) for r in cur.fetchall()]
+    conn.close()
+    return rows
+
+# ====== 结果导向的人话解释 ======
+RE_MONTH = re.compile(r"date\s+LIKE\s*'(\d{4}-\d{2})%'", re.IGNORECASE)
+RE_RANGE = re.compile(r"date\s+BETWEEN\s*'(\d{4}-\d{2}-\d{2})'\s+AND\s*'(\d{4}-\d{2}-\d{2})'", re.IGNORECASE)
+RE_DEPT  = re.compile(r"department\s*=\s*'([^']+)'", re.IGNORECASE)
+RE_CAT   = re.compile(r"category\s*=\s*'([^']+)'", re.IGNORECASE)
+RE_VEND  = re.compile(r"vendor\s*=\s*'([^']+)'", re.IGNORECASE)
+
+def _first_numeric(row: Dict[str, Any]):
+    for k in ["total", "sum", "actual", "budget_amount", "diff", "SUM(amount)", "COUNT(*)"]:
+        if k in row and isinstance(row[k], (int, float)):
+            return k, row[k]
+    for k, v in row.items():
+        if isinstance(v, (int, float)):
+            return k, v
+    return None, None
+
+def _fmt_num(v):
+    if isinstance(v, (int, float)):
+        return int(v) if float(v).is_integer() else round(float(v), 2)
+    return v
+
+def _ctx_from_sql(sql: str) -> str:
+    parts = []
+    m = RE_MONTH.search(sql);  m and parts.append(f"{m.group(1)} 月")
+    r = RE_RANGE.search(sql);  r and parts.append(f"{r.group(1)} 至 {r.group(2)}")
+    d = RE_DEPT.search(sql);   d and parts.append(d.group(1))
+    c = RE_CAT.search(sql);    c and parts.append(c.group(1))
+    v = RE_VEND.search(sql);   v and parts.append(v.group(1))
+    return "、".join(parts)
+
+def make_explanation(sql: str, rows: List[Dict[str, Any]]) -> str:
+    if not sql:
+        return "未生成有效SQL。"
+    ctx = _ctx_from_sql(sql)
+
+    if not rows:
+        return (ctx + " 没有匹配到记录。").strip()
+
+    sl = sql.lower()
+
+    # 1) 单值聚合（SUM/AVG/COUNT… 或仅一行且存在数值列）
+    if " sum(" in sl or " avg(" in sl or " count(" in sl or " total(" in sl or len(rows) == 1:
+        row = rows[0]
+        k, v = _first_numeric(row)
+        if v is not None:
+            label_map = {
+                "total": "费用",
+                "SUM(amount)": "费用",
+                "actual": "实际费用",
+                "budget_amount": "预算",
+                "COUNT(*)": "条数",
+            }
+            label = label_map.get(k, "结果")
+            v = _fmt_num(v)
+            prefix = (ctx + " ") if ctx else ""
+            if label in ["费用", "实际费用", "预算", "结果"] and isinstance(v, (int, float)):
+                return f"{prefix}{label}为 {v} 元。"
+            return f"{prefix}{label}为 {v}。"
+
+    # 2) 分组聚合（GROUP BY）
+    if " group by " in sl and rows:
+        key = None
+        for cand in ["department", "category", "vendor", "month", "date"]:
+            if cand in rows[0]:
+                key = cand; break
+        if not key:
+            for k, v in rows[0].items():
+                if not isinstance(v, (int, float)):
+                    key = k; break
+        samples = []
+        for r in rows[:3]:
+            _, val = _first_numeric(r)
+            samples.append(f"{r.get(key)} { _fmt_num(val) }")
+        head = (ctx + " ") if ctx else ""
+        return f"{head}按 {key} 统计，共 {len(rows)} 行。示例：{ '，'.join(samples) }。"
+
+    # 3) 明细（尝试汇总金额）
+    if "amount" in rows[0]:
+        s = sum(float(r.get("amount", 0) or 0) for r in rows)
+        s = _fmt_num(s)
+        head = (ctx + " ") if ctx else ""
+        return f"{head}共 {len(rows)} 条，合计 {s} 元。"
+
+    # 兜底
+    return f"{(ctx + ' ') if ctx else ''}明细查询，返回 {len(rows)} 行。"
+
+# ====== 模型加载（启动时一次）======
+def load_model_once():
+    global tokenizer, model, MODE, model_ready
+    if model_ready:
+        return
+    logging.info("启动时加载分词器与模型（一次）...")
+    tok  = AutoTokenizer.from_pretrained(BASE_MODEL_PATH, use_fast=True)
+    base = AutoModelForCausalLM.from_pretrained(
+        BASE_MODEL_PATH, dtype=torch.float32, device_map=None
+    ).to("cpu")
+    if os.path.isdir(LORA_ADAPTER_PATH):
+        mdl = PeftModel.from_pretrained(base, LORA_ADAPTER_PATH).to("cpu")
+        MODE = "lora"
+    else:
+        mdl = base
+        MODE = "base"
+    if hasattr(mdl, "config"):
+        mdl.config.use_cache = False
+    tokenizer, model = tok, mdl
+    model_ready = True
+    logging.info(f"模型就绪，模式：{MODE}")
+
+# ====== FastAPI ======
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    load_model_once()
+    yield
+
+app = FastAPI(title="NL→SQL 中间件", lifespan=lifespan)
+
+# 静态页
+if os.path.isdir(UI_DIR):
+    app.mount("/ui", StaticFiles(directory=UI_DIR, html=True), name="ui")
+
+class AskIn(BaseModel):
+    query: Optional[str] = None
+    question: Optional[str] = None
+
+@app.get("/")
+def root():
+    if os.path.isfile(os.path.join(UI_DIR, "index.html")):
+        return RedirectResponse(url="/ui/")
+    return HTMLResponse("<p>未找到 ui/index.html，请检查 ui 目录。</p>")
+
+@app.get("/health")
+def health():
+    return {"db_exists": os.path.isfile(DB_PATH), "mode": MODE, "model_ready": model_ready}
+
+@app.post("/ask")
+def ask(body: AskIn):
+    if not model_ready:
+        return JSONResponse({"error": "服务初始化中，请稍后重试"}, status_code=503)
+    q = (body.query or body.question or "").strip()
+    if not q:
+        return JSONResponse({"error": "请输入问题"}, status_code=400)
+
+    t0 = time.time()
+    sql = generate_sql(q)
+    if not sql:
+        logging.warning(f"[BLOCKED] {q}")
+        return {
+            "query": q, "sql": sql, "rows": [],
+            "explanation": "未生成有效的SELECT或被安全策略拦截。"
+        }
+    try:
+        rows = run_sql(sql)
+        expl = make_explanation(sql, rows)
+        logging.info(f"[OK] {q} | {sql} | {len(rows)}行 | {round((time.time()-t0)*1000,1)}ms | {MODE}")
+        return {"query": q, "sql": sql, "rows": rows, "explanation": expl, "mode": MODE}
+    except Exception as e:
+        logging.exception(f"[ERROR] {q} | {sql} | {e}")
+        return JSONResponse({"query": q, "sql": sql, "error": str(e)}, status_code=500)
+
+# 直接运行
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run("middleware:app", host="127.0.0.1", port=8000, reload=False)

zhangxinyue2024103833/models/deepseek-sql-lora/README.md

@@ -0,0 +1,207 @@
+---
+base_model: D:/人大/结课作业/中间件/deepseek1.5B
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:D:/人大/结课作业/中间件/deepseek1.5B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.17.1

zhangxinyue2024103833/models/deepseek-sql-lora/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "D:/\u4eba\u5927/\u7ed3\u8bfe\u4f5c\u4e1a/\u4e2d\u95f4\u4ef6/deepseek1.5B",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 8,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 4,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "q_proj",
+    "o_proj",
+    "gate_proj",
+    "k_proj",
+    "down_proj",
+    "up_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

zhangxinyue2024103833/models/deepseek-sql-lora/chat_template.jinja

@@ -0,0 +1 @@
+{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜><think>\n'}}{% endif %}

zhangxinyue2024103833/models/deepseek-sql-lora/checkpoint-200/README.md

@@ -0,0 +1,207 @@
+---
+base_model: D:/人大/结课作业/中间件/deepseek1.5B
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:D:/人大/结课作业/中间件/deepseek1.5B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.17.1

zhangxinyue2024103833/models/deepseek-sql-lora/checkpoint-200/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "D:/\u4eba\u5927/\u7ed3\u8bfe\u4f5c\u4e1a/\u4e2d\u95f4\u4ef6/deepseek1.5B",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 8,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 4,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "q_proj",
+    "o_proj",
+    "gate_proj",
+    "k_proj",
+    "down_proj",
+    "up_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

zhangxinyue2024103833/models/deepseek-sql-lora/checkpoint-200/chat_template.jinja

@@ -0,0 +1 @@
+{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜><think>\n'}}{% endif %}

zhangxinyue2024103833/models/deepseek-sql-lora/checkpoint-200/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<｜begin▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

zhangxinyue2024103833/models/deepseek-sql-lora/checkpoint-200/tokenizer_config.json

@@ -0,0 +1,194 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<｜end▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<｜User｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151645": {
+      "content": "<｜Assistant｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151646": {
+      "content": "<｜begin▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|EOT|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151648": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151649": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "extra_special_tokens": {},
+  "legacy": true,
+  "model_max_length": 16384,
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizerFast",
+  "unk_token": null,
+  "use_default_system_prompt": false
+}

zhangxinyue2024103833/models/deepseek-sql-lora/checkpoint-200/trainer_state.json

@@ -0,0 +1,174 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.142857142857143,
+  "eval_steps": 500,
+  "global_step": 200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.35714285714285715,
+      "grad_norm": 1.7812830209732056,
+      "learning_rate": 9e-05,
+      "loss": 3.0968,
+      "step": 10
+    },
+    {
+      "epoch": 0.7142857142857143,
+      "grad_norm": 1.735021710395813,
+      "learning_rate": 9.944739353007344e-05,
+      "loss": 2.3665,
+      "step": 20
+    },
+    {
+      "epoch": 1.0714285714285714,
+      "grad_norm": 1.8056219816207886,
+      "learning_rate": 9.755282581475769e-05,
+      "loss": 1.6148,
+      "step": 30
+    },
+    {
+      "epoch": 1.4285714285714286,
+      "grad_norm": 1.0633488893508911,
+      "learning_rate": 9.43611409721806e-05,
+      "loss": 1.0062,
+      "step": 40
+    },
+    {
+      "epoch": 1.7857142857142856,
+      "grad_norm": 1.0730570554733276,
+      "learning_rate": 8.995939984474624e-05,
+      "loss": 0.7507,
+      "step": 50
+    },
+    {
+      "epoch": 2.142857142857143,
+      "grad_norm": 1.1507912874221802,
+      "learning_rate": 8.44676704559283e-05,
+      "loss": 0.5951,
+      "step": 60
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 1.0625510215759277,
+      "learning_rate": 7.803575286758364e-05,
+      "loss": 0.5993,
+      "step": 70
+    },
+    {
+      "epoch": 2.857142857142857,
+      "grad_norm": 1.0893510580062866,
+      "learning_rate": 7.083909302476453e-05,
+      "loss": 0.479,
+      "step": 80
+    },
+    {
+      "epoch": 3.2142857142857144,
+      "grad_norm": 1.101180911064148,
+      "learning_rate": 6.307399704769099e-05,
+      "loss": 0.4169,
+      "step": 90
+    },
+    {
+      "epoch": 3.571428571428571,
+      "grad_norm": 1.3419430255889893,
+      "learning_rate": 5.495227651252315e-05,
+      "loss": 0.3729,
+      "step": 100
+    },
+    {
+      "epoch": 3.928571428571429,
+      "grad_norm": 1.4156843423843384,
+      "learning_rate": 4.669547078371504e-05,
+      "loss": 0.3188,
+      "step": 110
+    },
+    {
+      "epoch": 4.285714285714286,
+      "grad_norm": 1.3531630039215088,
+      "learning_rate": 3.852880399766243e-05,
+      "loss": 0.2944,
+      "step": 120
+    },
+    {
+      "epoch": 4.642857142857143,
+      "grad_norm": 1.1633448600769043,
+      "learning_rate": 3.0675041535377405e-05,
+      "loss": 0.2422,
+      "step": 130
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 1.4601385593414307,
+      "learning_rate": 2.3348413563600325e-05,
+      "loss": 0.2885,
+      "step": 140
+    },
+    {
+      "epoch": 5.357142857142857,
+      "grad_norm": 0.9815771579742432,
+      "learning_rate": 1.6748771394307585e-05,
+      "loss": 0.2463,
+      "step": 150
+    },
+    {
+      "epoch": 5.714285714285714,
+      "grad_norm": 1.2873295545578003,
+      "learning_rate": 1.1056136061894384e-05,
+      "loss": 0.1919,
+      "step": 160
+    },
+    {
+      "epoch": 6.071428571428571,
+      "grad_norm": 1.52511727809906,
+      "learning_rate": 6.425787818636131e-06,
+      "loss": 0.2503,
+      "step": 170
+    },
+    {
+      "epoch": 6.428571428571429,
+      "grad_norm": 1.0356954336166382,
+      "learning_rate": 2.9840304941919415e-06,
+      "loss": 0.2002,
+      "step": 180
+    },
+    {
+      "epoch": 6.785714285714286,
+      "grad_norm": 1.298083782196045,
+      "learning_rate": 8.247462563808817e-07,
+      "loss": 0.2331,
+      "step": 190
+    },
+    {
+      "epoch": 7.142857142857143,
+      "grad_norm": 1.026187777519226,
+      "learning_rate": 6.834750376549792e-09,
+      "loss": 0.1929,
+      "step": 200
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 200,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 8,
+  "save_steps": 200,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 255102932520960.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

zhangxinyue2024103833/models/deepseek-sql-lora/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<｜begin▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

zhangxinyue2024103833/models/deepseek-sql-lora/tokenizer_config.json

@@ -0,0 +1,194 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<｜end▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<｜User｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151645": {
+      "content": "<｜Assistant｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151646": {
+      "content": "<｜begin▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|EOT|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151648": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151649": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "extra_special_tokens": {},
+  "legacy": true,
+  "model_max_length": 16384,
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizerFast",
+  "unk_token": null,
+  "use_default_system_prompt": false
+}

zhangxinyue2024103833/readme.md

@@ -0,0 +1,74 @@
+# 财务问答中间件（自然语言 → SQL → 结果）
+
+一句话说明：这是一个**把中文问题自动翻译成 SQL，查询 SQLite，再用人话解释结果**的小系统，“中间件 + 大模型”的结合。
+
+---
+
+## 🧠 原理简述
+1. 输入中文问题（例：8 月份市场部的费用是多少？）  
+2. 大模型生成对应的 SQL  
+3. SQLite 执行 SQL，得到结果  
+4. 后端把结果转成通俗解释  
+5. 网页展示：问题 → 解释 → SQL → 结果表  
+
+---
+
+## 📂 目录结构
+<pre>
+中间件/
+├── train.jsonl                微调样本（每行：问题→SQL 的 JSON）
+├── finetune.py                微调脚本：用 train.jsonl 对基础模型做 LoRA 微调
+├── middleware.py              后端中间件（FastAPI）：问句→SQL→查询→解释→返回给前端
+├── init_db.py                 初始化示例库：创建/写入演示用财务数据
+├── finance_demo.db            SQLite 数据库（运行 init_db.py 生成）
+│
+├── deepseek1.5B/              基础模型目录（你本地已有）
+├── models/
+│   └── deepseek-sql-lora/     微调产物（LoRA 适配器）
+│
+├── ui/                        前端网页
+│   ├── index.html             主页面（输入→显示解释/SQL/结果）
+│   ├── screenshot.png         演示截图（下方“演示案例”会引用）
+│   └── ...
+│
+└── logs/
+    └── middleware.log         运行日志
+    └── finetune.log           微调日志
+</pre>
+
+---
+
+## 🚀 使用步骤（通俗说明）
+
+### 1）初始化数据库  
+运行 `python init_db.py`  
+👉 生成或覆盖 `finance_demo.db`，写入示例财务数据（费用、部门、日期、预算等）。  
+
+### 2）微调小模型（可选）  
+运行 `python finetune.py`  
+👉 读取 `train.jsonl`，在 `deepseek1.5B/` 上做 LoRA 微调，结果存到 `models/deepseek-sql-lora/`。  
+
+### 3）启动后端中间件  
+运行 `python middleware.py`  
+👉 启动 FastAPI 服务，默认访问地址是 [http://127.0.0.1:8000/ui/](http://127.0.0.1:8000/ui/)。  
+
+### 4）打开前端页面演示  
+用浏览器访问 `http://127.0.0.1:8000/ui/`，输入问题或点击示例按钮，页面会显示：解释 → SQL → 查询结果。  
+推荐测试问题：  
+- 8月份市场部的费用是多少？  
+- 2024年8月各部门费用从高到低排序 
+
+---
+
+## 🎬 演示案例
+
+
+<img src="ui/screenshot.png" alt="演示页面" width="40%">
+
+---
+
+## ✨ 项目亮点
+- **大模型驱动**：利用大模型的强大语义理解和生成能力，把自然语言问题准确转化为 SQL，降低非技术人员使用数据库的门槛。  
+- **结构完整**：自然语言 → SQL → 数据库 → 解释，完整演示中间件与大模型结合；  
+- **可用与可维护**：前端可视化输出让业务人员能直接理解查询结果，后端日志记录则方便技术人员排查问题、进行审计，保证系统可靠运行。
+

zhangxinyue2024103833/requirements.txt

@@ -0,0 +1,9 @@
+torch>=2.2.0
+transformers>=4.44.0
+datasets>=2.20.0
+accelerate>=0.33.0
+peft>=0.11.1
+bitsandbytes>=0.43.1
+fastapi>=0.110.0
+uvicorn>=0.27.0
+pydantic>=2.0.0

zhangxinyue2024103833/train.jsonl

@@ -0,0 +1,32 @@
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"8月份市场部的费用是多少？","output":"SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE department='市场部' AND date LIKE '2024-08%';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年第三季度研发部的总支出","output":"SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE department='研发部' AND date BETWEEN '2024-07-01' AND '2024-09-30';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"列出2024年7月市场部的费用明细","output":"SELECT date,category,vendor,amount FROM expenses WHERE department='市场部' AND date LIKE '2024-07%';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"各部门在2024年8月分别花了多少钱，按从高到低排序","output":"SELECT department, IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%' GROUP BY department ORDER BY total DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年截至9月的累计广告费用","output":"SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE category='广告' AND date BETWEEN '2024-01-01' AND '2024-09-30';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"行政部2024年上半年的平均单笔报销金额","output":"SELECT IFNULL(AVG(amount),0) AS avg_amount FROM expenses WHERE department='行政部' AND date BETWEEN '2024-01-01' AND '2024-06-30';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"找出2024年8月金额超过2万的差旅费用","output":"SELECT date,department,vendor,amount FROM expenses WHERE category='差旅' AND date LIKE '2024-08%' AND amount > 20000 ORDER BY amount DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年7月到8月，财务部在办公类支出总额","output":"SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE department='财务部' AND category='办公' AND date BETWEEN '2024-07-01' AND '2024-08-31';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"按月份统计2024年研发部的总费用","output":"SELECT substr(date,1,7) AS month, IFNULL(SUM(amount),0) AS total FROM expenses WHERE department='研发部' AND date BETWEEN '2024-01-01' AND '2024-12-31' GROUP BY month ORDER BY month;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"市场部哪家供应商在2024年收款最多？返回前3名","output":"SELECT vendor, IFNULL(SUM(amount),0) AS total FROM expenses WHERE department='市场部' AND date BETWEEN '2024-01-01' AND '2024-12-31' GROUP BY vendor ORDER BY total DESC LIMIT 3;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年8月相较7月，公司整体费用环比","output":"SELECT '2024-08' AS month, IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%' UNION ALL SELECT '2024-07' AS month, IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-07%';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年每个部门的广告费用占比（总额和占比）","output":"WITH dept AS (SELECT department, IFNULL(SUM(amount),0) AS total FROM expenses WHERE category='广告' AND date BETWEEN '2024-01-01' AND '2024-12-31' GROUP BY department), allsum AS (SELECT IFNULL(SUM(total),0) AS s FROM dept) SELECT d.department, d.total, ROUND(d.total*1.0/a.s,4) AS ratio FROM dept d, allsum a ORDER BY d.total DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年9月销售部是否超预算？","output":"SELECT b.month, b.department, b.budget_amount, IFNULL(e.actual,0) AS actual, CASE WHEN IFNULL(e.actual,0) > b.budget_amount THEN 1 ELSE 0 END AS is_over FROM budgets b LEFT JOIN (SELECT substr(date,1,7) AS month, department, SUM(amount) AS actual FROM expenses WHERE date LIKE '2024-09%' GROUP BY department) e ON b.month=e.month AND b.department=e.department WHERE b.month='2024-09' AND b.department='销售部';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"列出2024年8月所有部门的预算与实际对比","output":"SELECT b.department, b.budget_amount, IFNULL(e.actual,0) AS actual, (IFNULL(e.actual,0)-b.budget_amount) AS diff FROM budgets b LEFT JOIN (SELECT department, SUM(amount) AS actual FROM expenses WHERE date LIKE '2024-08%' GROUP BY department) e ON b.department=e.department WHERE b.month='2024-08' ORDER BY diff DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"请给出2024年每月的总费用（时间升序）","output":"SELECT substr(date,1,7) AS month, IFNULL(SUM(amount),0) AS total FROM expenses WHERE date BETWEEN '2024-01-01' AND '2024-12-31' GROUP BY month ORDER BY month;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"找出2024-10-01到2024-10-07期间的支出","output":"SELECT date,department,category,vendor,amount FROM expenses WHERE date BETWEEN '2024-10-01' AND '2024-10-07' ORDER BY amount DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年培训类费用最高的部门是哪个？","output":"SELECT department FROM expenses WHERE category='培训' AND date BETWEEN '2024-01-01' AND '2024-12-31' GROUP BY department ORDER BY SUM(amount) DESC LIMIT 1;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"过去三个月（截至2024-09-30）各部门总费用","output":"SELECT department, IFNULL(SUM(amount),0) AS total FROM expenses WHERE date BETWEEN '2024-07-01' AND '2024-09-30' GROUP BY department ORDER BY total DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"列出2024年8月金额超过10000的所有记录","output":"SELECT date,department,category,vendor,amount FROM expenses WHERE date LIKE '2024-08%' AND amount > 10000 ORDER BY amount DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年8月各费用类别的支出占比","output":"WITH cat AS (SELECT category, IFNULL(SUM(amount),0) AS total FROM expenses WHERE date LIKE '2024-08%' GROUP BY category), s AS (SELECT IFNULL(SUM(total),0) AS allsum FROM cat) SELECT c.category, c.total, ROUND(c.total*1.0/s.allsum,4) AS ratio FROM cat c, s ORDER BY c.total DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"市场部2024年每月的平均单笔金额","output":"SELECT substr(date,1,7) AS month, IFNULL(AVG(amount),0) AS avg_amount FROM expenses WHERE department='市场部' AND date BETWEEN '2024-01-01' AND '2024-12-31' GROUP BY month ORDER BY month;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年上半年公司总费用、下半年公司总费用","output":"SELECT 'H1' AS half, IFNULL(SUM(amount),0) AS total FROM expenses WHERE date BETWEEN '2024-01-01' AND '2024-06-30' UNION ALL SELECT 'H2' AS half, IFNULL(SUM(amount),0) AS total FROM expenses WHERE date BETWEEN '2024-07-01' AND '2024-12-31';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年8月哪天的支出最高？","output":"SELECT date, SUM(amount) AS total FROM expenses WHERE date LIKE '2024-08%' GROUP BY date ORDER BY total DESC LIMIT 1;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"列出2024年每个部门的最高单笔支出","output":"SELECT department, MAX(amount) AS max_amount FROM expenses WHERE date BETWEEN '2024-01-01' AND '2024-12-31' GROUP BY department ORDER BY max_amount DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年8月人力费用总额和笔数","output":"SELECT IFNULL(SUM(amount),0) AS total, COUNT(*) AS cnt FROM expenses WHERE category='人力' AND date LIKE '2024-08%';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"供应商“XX传媒”在2024年收了多少钱？","output":"SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE vendor='XX传媒' AND date BETWEEN '2024-01-01' AND '2024-12-31';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年8月预算执行率（按部门）","output":"SELECT b.department, b.budget_amount, IFNULL(e.actual,0) AS actual, CASE WHEN b.budget_amount>0 THEN ROUND(IFNULL(e.actual,0)*1.0/b.budget_amount,4) ELSE NULL END AS exec_ratio FROM budgets b LEFT JOIN (SELECT department, SUM(amount) AS actual FROM expenses WHERE date LIKE '2024-08%' GROUP BY department) e ON b.department=e.department WHERE b.month='2024-08' ORDER BY exec_ratio DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年Q1公司总费用（1-3月）","output":"SELECT IFNULL(SUM(amount),0) AS total FROM expenses WHERE date BETWEEN '2024-01-01' AND '2024-03-31';"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年7月市场部与销售部费用对比","output":"SELECT department, IFNULL(SUM(amount),0) AS total FROM expenses WHERE department IN ('市场部','销售部') AND date LIKE '2024-07%' GROUP BY department ORDER BY total DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"最近三个月每月广告费用（截至2024-09）","output":"SELECT substr(date,1,7) AS month, IFNULL(SUM(amount),0) AS total FROM expenses WHERE category='广告' AND date BETWEEN '2024-07-01' AND '2024-09-30' GROUP BY month ORDER BY month;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年每个部门的人力费用，从高到低","output":"SELECT department, IFNULL(SUM(amount),0) AS total FROM expenses WHERE category='人力' AND date BETWEEN '2024-01-01' AND '2024-12-31' GROUP BY department ORDER BY total DESC;"}
+{"instruction":"请将自然语言问题转换为 SQLite 查询，并且只输出 SQL，不要其他内容。","input":"2024年8月各部门单笔均值、最大最小","output":"SELECT department, IFNULL(AVG(amount),0) AS avg_amount, MAX(amount) AS max_amount, MIN(amount) AS min_amount FROM expenses WHERE date LIKE '2024-08%' GROUP BY department ORDER BY avg_amount DESC;"}

zhangxinyue2024103833/ui/index.html

@@ -0,0 +1,193 @@
+<!doctype html>
+<html lang="zh">
+<head>
+<meta charset="utf-8" />
+<meta name="viewport" content="width=device-width, initial-scale=1" />
+<title>财务自然语言 → SQL 演示</title>
+<style>
+  :root{
+    --text:#0f172a; --muted:#64748b; --border:#e5e7eb;
+    --brand:#2563eb; --brand2:#7c3aed; --accent:#06b6d4;
+    --card:#ffffff; --shadow:0 14px 30px rgba(2,6,23,.08);
+    --mono: ui-monospace, SFMono-Regular, Menlo, Consolas, "Liberation Mono", monospace;
+  }
+  *{box-sizing:border-box}
+  body{
+    margin:0; color:var(--text);
+    font:14px/1.6 system-ui,-apple-system,"Segoe UI",Roboto,"PingFang SC","Hiragino Sans GB","Microsoft YaHei",sans-serif;
+    background:#f6f8fc;
+  }
+  .container{max-width:900px;margin:30px auto;padding:0 18px}
+
+  .header{display:flex;align-items:center;gap:14px;margin:0 0 20px}
+  .logo{width:44px;height:44px;border-radius:50%;
+    background:conic-gradient(from 210deg, var(--brand), var(--accent), var(--brand2), var(--brand));
+    box-shadow:var(--shadow)}
+  .title{font-weight:900;font-size:22px;}
+  .sub{color:var(--muted);font-size:13px}
+
+  .card{
+    background:var(--card); border:1px solid var(--border);
+    border-radius:20px; box-shadow:var(--shadow); overflow:hidden;
+    margin-bottom:18px;
+  }
+  .card-h{padding:14px 18px 10px; font-weight:800;}
+  .card-b{padding:16px 18px 18px}
+
+  /* 输入区 */
+  .input-wrap{position:relative}
+  .input{
+    width:100%; height:54px; padding:0 18px;
+    border:1px solid var(--border); border-radius:999px; outline:none; background:#fff; font-size:15px;
+  }
+  .actions{display:flex; gap:10px; margin-top:12px; flex-wrap:wrap}
+  .btn{
+    appearance:none; border:0; color:#fff; cursor:pointer;
+    background:linear-gradient(135deg, var(--brand), var(--brand2));
+    border-radius:999px; padding:10px 16px; font-weight:800;
+    box-shadow:0 8px 18px rgba(37,99,235,.18);
+  }
+  .btn-gray{background:#94a3b8; box-shadow:none}
+  .btn-round{width:42px;height:42px;display:flex;align-items:center;justify-content:center;
+    border-radius:50%; font-weight:900;}
+
+  .status{min-height:18px;color:var(--muted);font-size:13px;margin-top:6px}
+
+  /* 内容模块 */
+  .pill{
+    border:1px solid var(--border); border-radius:18px;
+    background:#f9fafb; padding:16px 20px;
+  }
+  .pill.mono{font-family:var(--mono); white-space:pre-wrap; word-break:break-all;}
+
+  /* 复制按钮 */
+  .copy-btn{
+    float:right; margin:-6px -6px 0 0;
+    border:none; border-radius:50%;
+    width:34px;height:34px; background:linear-gradient(135deg,var(--brand),var(--brand2));
+    color:#fff; font-weight:800; cursor:pointer;
+  }
+
+  /* 表格 */
+  .table-wrap{overflow:auto; border:1px solid var(--border); border-radius:16px}
+  table{border-collapse:collapse; width:100%}
+  th,td{border-bottom:1px solid var(--border); padding:10px; text-align:left}
+  th{background:#f8fafc}
+  tr:nth-child(even) td{background:#fbfdff}
+  .empty{color:var(--muted); padding:14px}
+</style>
+</head>
+<body>
+  <div class="container">
+    <div class="header">
+      <div class="logo"></div>
+      <div>
+        <div class="title">财务自然语言 → SQL 演示</div>
+        <div class="sub">输入问题 → 查看解释 → SQL → 结果</div>
+      </div>
+    </div>
+
+    <!-- 输入 -->
+    <div class="card">
+      <div class="card-h">输入问题</div>
+      <div class="card-b">
+        <div class="input-wrap">
+          <input id="q" class="input" placeholder="点击示例或输入…" />
+        </div>
+        <div class="actions">
+          <button id="btnAsk" class="btn">查询</button>
+          <button id="btnClear" class="btn btn-gray">清空</button>
+          <button class="btn btn-round" data-q="8月份市场部的费用是多少？">1</button>
+          <button class="btn btn-round" data-q="2024年8月各部门费用从高到低排序">2</button>
+        </div>
+        <div id="status" class="status"></div>
+      </div>
+    </div>
+
+    <!-- 解释 -->
+    <div class="card">
+      <div class="card-h">解释</div>
+      <div class="card-b">
+        <div class="pill" id="explain">（等待查询）</div>
+      </div>
+    </div>
+
+    <!-- SQL -->
+    <div class="card">
+      <div class="card-h">生成的 SQL 
+        <button id="btnCopy" class="copy-btn" title="复制 SQL">⧉</button>
+      </div>
+      <div class="card-b">
+        <div class="pill mono" id="sql">（等待查询）</div>
+      </div>
+    </div>
+
+    <!-- 结果 -->
+    <div class="card">
+      <div class="card-h">查询结果</div>
+      <div class="card-b">
+        <div id="result" class="empty">（暂无结果）</div>
+      </div>
+    </div>
+  </div>
+
+<script>
+  const $ = s=>document.querySelector(s);
+  const qInput=$('#q'), btnAsk=$('#btnAsk'), btnClear=$('#btnClear'), btnCopy=$('#btnCopy');
+  const explainBox=$('#explain'), sqlBox=$('#sql'), resultBox=$('#result'), status=$('#status');
+
+  document.querySelectorAll('[data-q]').forEach(el=>{
+    el.addEventListener('click', ()=>{ qInput.value = el.getAttribute('data-q'); });
+  });
+
+  function renderTable(rows){
+    if(!rows || !rows.length) return '<div class="empty">无结果</div>';
+    const cols=Object.keys(rows[0]);
+    const head='<thead><tr>'+cols.map(c=>`<th>${c}</th>`).join('')+'</tr></thead>';
+    const body='<tbody>'+rows.map(r=>'<tr>'+cols.map(c=>`<td>${r[c]}</td>`).join('')+'</tr>').join('')+'</tbody>';
+    return `<div class="table-wrap"><table>${head}${body}</table></div>`;
+  }
+
+  function askBackend(query){
+    const url = location.protocol==='file:' ? 'http://127.0.0.1:8000/ask' : '/ask';
+    return fetch(url,{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({query})})
+      .then(r=>r.json());
+  }
+
+  btnAsk.addEventListener('click', async ()=>{
+    const q=qInput.value.trim();
+    if(!q){ status.textContent='请输入问题'; return; }
+    status.textContent='查询中…'; btnAsk.disabled=true;
+
+    explainBox.textContent='（生成中…）';
+    sqlBox.textContent='（生成中…）';
+    resultBox.innerHTML='<div class="empty">生成中…</div>';
+
+    try{
+      const data=await askBackend(q);
+      explainBox.textContent=data.explanation||'（无解释）';
+      sqlBox.textContent=data.sql||'（无SQL）';
+      resultBox.innerHTML=renderTable(data.rows||[]);
+      status.textContent='完成';
+    }catch(e){
+      status.textContent='请求失败'; status.style.color='#ef4444';
+      console.error(e);
+    }finally{btnAsk.disabled=false;}
+  });
+
+  btnClear.addEventListener('click', ()=>{
+    qInput.value=''; status.textContent='';
+    explainBox.textContent='（等待查询）';
+    sqlBox.textContent='（等待查询）';
+    resultBox.innerHTML='（暂无结果）';
+  });
+
+  btnCopy.addEventListener('click', async ()=>{
+    const t=sqlBox.textContent.trim();
+    if(!t||t.startsWith('（')) return;
+    try{await navigator.clipboard.writeText(t); btnCopy.textContent='✓'; setTimeout(()=>btnCopy.textContent='⧉',1200);}
+    catch{btnCopy.textContent='!'; setTimeout(()=>btnCopy.textContent='⧉',1200);}
+  });
+</script>
+</body>
+</html>