Feat/lmy/pdf2vqa to main

.gitignore

@@ -5,6 +5,7 @@
 *.vscode
 *.txt
 *.__pycache__
+.venv/
 
 # compilation and distribution
 __pycache__

README.md

@@ -101,6 +101,7 @@ uniparser parse report.pdf
 
 ```python
 import os
+
 parser = UniParserClient(
     host="https://uniparser.dp.tech/",
     api_key=os.getenv("UNIPARSER_API_KEY"),
@@ -152,13 +153,10 @@ import os
 from uniparser_tools.api.clients import UniParserClient
 
 # 设置 API 密钥
-api_key = os.getenv('UNIPARSER_API_KEY')
+api_key = os.getenv("UNIPARSER_API_KEY")
 
 # 初始化客户端
-parser = UniParserClient(
-    host="https://uniparser.dp.tech/",
-    api_key=api_key
-)
+parser = UniParserClient(host="https://uniparser.dp.tech/", api_key=api_key)
 ```
 
 ### 2. 解析 PDF 文件（科学文献推荐默认）
@@ -170,12 +168,12 @@ from uniparser_tools.common.constant import ParseMode, ParseModeTextual
 result = parser.trigger_file(
     pdf_path="./example.pdf",
     textual=ParseModeTextual.OCRHighQuality,  # high quality
-    equation=ParseMode.OCRHighQuality,        # high quality
-    table=ParseMode.OCRHighQuality,           # high quality
-    chart=ParseMode.DumpBase64,               # original image base64
-    figure=ParseMode.DumpBase64,              # original image base64
-    expression=ParseMode.DumpBase64,          # original image base64
-    molecule=ParseMode.OCRFast,               # fast
+    equation=ParseMode.OCRHighQuality,  # high quality
+    table=ParseMode.OCRHighQuality,  # high quality
+    chart=ParseMode.DumpBase64,  # original image base64
+    figure=ParseMode.DumpBase64,  # original image base64
+    expression=ParseMode.DumpBase64,  # original image base64
+    molecule=ParseMode.OCRFast,  # fast
 )
 
 if result["status"] == "success":
@@ -327,9 +325,9 @@ from uniparser_tools.common.constant import FormatFlag
 result = parser.get_formatted(
     token,
     content=True,
-    textual=FormatFlag.Markdown,    # 文本使用 Markdown
-    table=FormatFlag.Html,          # 表格使用 HTML
-    equation=FormatFlag.Latex,       # 公式使用 LaTeX
+    textual=FormatFlag.Markdown,  # 文本使用 Markdown
+    table=FormatFlag.Html,  # 表格使用 HTML
+    equation=FormatFlag.Latex,  # 公式使用 LaTeX
 )
 
 if result["status"] == "success":

skills/UniParser-Tools/references/api-reference.md

@@ -15,9 +15,9 @@
 ```python
 result = parser.trigger_file(
     file_path="./document.pdf",
-    sync=False,                    # Required for async mode
-    callback_url="https://...",    # Your callback endpoint
-    callback_secret="your-secret", # For signature verification
+    sync=False,  # Required for async mode
+    callback_url="https://...",  # Your callback endpoint
+    callback_secret="your-secret",  # For signature verification
     # ... other parse mode parameters
 )
 ```

skills/UniParser-Tools/references/data-classes.md

@@ -54,6 +54,6 @@ for page in pages_tree:
         print(f"LaTeX: {item.latex}")
 
         # For tables, access DataFrame
-        if hasattr(item, 'df'):
+        if hasattr(item, "df"):
             print(f"DataFrame:\n{item.df}")
 ```

skills/UniParser-Tools/references/notes.md

@@ -14,6 +14,7 @@
    import hmac
    import hashlib
 
+
    def verify_callback(content, checksum, secret):
        expected = hmac.new(secret.encode(), content.encode(), hashlib.sha256).hexdigest()
        return hmac.compare_digest(expected, checksum)

skills/UniParser-Tools/references/patterns.md

@@ -77,26 +77,23 @@ from flask import Flask, request
 app = Flask(__name__)
 CALLBACK_SECRET = "your-secret-key"
 
-@app.route('/callback', methods=['POST'])
+
+@app.route("/callback", methods=["POST"])
 def handle_callback():
     data = request.json
-    content = data['content']
-    received_checksum = data['checksum']
-    
+    content = data["content"]
+    received_checksum = data["checksum"]
+
     # Verify signature
-    expected = hmac.new(
-        CALLBACK_SECRET.encode(),
-        json.dumps(content).encode(),
-        hashlib.sha256
-    ).hexdigest()
-
+    expected = hmac.new(CALLBACK_SECRET.encode(), json.dumps(content).encode(), hashlib.sha256).hexdigest()
+
     if not hmac.compare_digest(received_checksum, expected):
-        return {'error': 'Invalid signature'}, 401
-    
+        return {"error": "Invalid signature"}, 401
+
     # Process the result
-    token = data['token']
+    token = data["token"]
     print(f"Task {token} completed!")
-    return {'status': 'ok'}
+    return {"status": "ok"}
 ```
 
 ## Pattern 4: Mixed Format Output
@@ -105,10 +102,10 @@ def handle_callback():
 result = parser.get_formatted(
     token,
     content=True,
-    textual=FormatFlag.Markdown,   # Text as Markdown
-    table=FormatFlag.Html,         # Tables as HTML
-    equation=FormatFlag.Latex,     # Equations as LaTeX
-    figure=FormatFlag.Markdown,    # Figures as Markdown img
+    textual=FormatFlag.Markdown,  # Text as Markdown
+    table=FormatFlag.Html,  # Tables as HTML
+    equation=FormatFlag.Latex,  # Equations as LaTeX
+    figure=FormatFlag.Markdown,  # Figures as Markdown img
 )
 ```
 

skills/UniParser-Tools/references/utilities.md

@@ -25,22 +25,22 @@ clean_text = clean_scientific_text(text, strict=False)
 from uniparser_tools.utils.bbox import BBox, Point
 
 # BBox properties
-bbox.area        # Area
-bbox.width       # Width
-bbox.height      # Height
-bbox.tl          # Top-left point
-bbox.br          # Bottom-right point
-bbox.ctr         # Center point
-bbox.xyxy        # (x1, y1, x2, y2) tuple
-bbox.xywh        # (x, y, w, h) tuple
+bbox.area  # Area
+bbox.width  # Width
+bbox.height  # Height
+bbox.tl  # Top-left point
+bbox.br  # Bottom-right point
+bbox.ctr  # Center point
+bbox.xyxy  # (x1, y1, x2, y2) tuple
+bbox.xywh  # (x, y, w, h) tuple
 
 # BBox operations
-bbox.iou(other)           # Intersection over Union
-bbox.iof(other)           # Intersection over Foreground
+bbox.iou(other)  # Intersection over Union
+bbox.iof(other)  # Intersection over Foreground
 bbox.intersection(other)  # Intersection box
-bbox.union(other)         # Union box
-bbox.expand(pix, wh)      # Expand by pixels
-bbox.shrink(pix, wh)      # Shrink by pixels
+bbox.union(other)  # Union box
+bbox.expand(pix, wh)  # Expand by pixels
+bbox.shrink(pix, wh)  # Shrink by pixels
 ```
 
 ## Text Processing

uniparser_agent/README.md

@@ -0,0 +1,11 @@
+# UniParser Agent
+
+UniParser Agent 是基于 UniParser 的文档处理包，提供统一的命令行入口：
+
+- `parse`：将 PDF、图片或公开 PDF URL 解析为结构化文档。
+- `vqa`：从习题、试卷、题册或答案册中提取题目、答案、解析和相关图片，并生成结构化 VQA 数据。
+
+本包包含文档解析、LLM 调用和 pdf2vqa 流程所需的公共模块。
+
+安装方式、配置说明、命令参数、处理流程和输出格式，请参阅
+[pdf2vqa 完整文档](pdf2vqa/README.md)。

uniparser_agent/__init__.py

@@ -0,0 +1,3 @@
+"""UniParser document parsing and exam VQA extraction."""
+
+__version__ = "0.1.0"

uniparser_agent/cli.py

@@ -0,0 +1,144 @@
+from __future__ import annotations
+
+import json
+from typing import Optional
+
+import typer
+
+from uniparser_agent.llm import LLMConfig, resolve_llm_config
+from uniparser_agent.parse.service import parse_document
+from uniparser_agent.pdf2vqa.pipeline import run_vqa_pipeline
+
+
+app = typer.Typer(
+    name="uniparser-agent",
+    help="UniParser document parsing and exam VQA extraction.",
+    no_args_is_help=True,
+)
+
+
+@app.command("parse")
+def parse_cmd(
+    input_path: str = typer.Argument(..., help="Local PDF/image path or public PDF URL."),
+    output_dir: Optional[str] = typer.Option(None, "-o", "--output-dir", help="Output directory."),
+    overwrite: bool = typer.Option(False, "--overwrite", help="Replace output directory if it exists."),
+    json_output: bool = typer.Option(False, "--json", help="Print machine-readable JSON."),
+) -> None:
+    """Parse a document with UniParser scientific-paper defaults."""
+    result = parse_document(input_path, output_dir=output_dir, overwrite=overwrite)
+    if json_output:
+        typer.echo(json.dumps(result, ensure_ascii=False, indent=2))
+        return
+    typer.echo(f"Token: {result.get('token', '')}")
+    typer.echo(f"Pages tree: {result['pages_tree_path']}")
+    typer.echo(f"Markdown: {result['markdown_path']}")
+    typer.echo(f"Output directory: {result['output_dir']}")
+
+
+@app.command("vqa")
+def vqa_cmd(
+    input_path: Optional[str] = typer.Argument(
+        None,
+        help="Local PDF/image path or public PDF URL. Omit when using --pages-tree.",
+    ),
+    output_dir: Optional[str] = typer.Option(None, "-o", "--output-dir", help="VQA output directory."),
+    answer_pdf: Optional[str] = typer.Option(
+        None,
+        "--answer-pdf",
+        help="Answer booklet PDF. Merged after the question booklet (local PDFs only).",
+    ),
+    pages_tree: Optional[str] = typer.Option(
+        None,
+        "--pages-tree",
+        help="Skip UniParser parse and use an existing pages_tree.json.",
+    ),
+    overwrite: bool = typer.Option(False, "--overwrite", help="Replace output directory if it exists."),
+    api_key: Optional[str] = typer.Option(
+        None,
+        "--api-key",
+        help="LLM API key (overrides OPENAI_API_KEY).",
+        envvar=[],
+    ),
+    base_url: Optional[str] = typer.Option(
+        None,
+        "--base-url",
+        help="LLM base URL (overrides OPENAI_BASE_URL).",
+    ),
+    model: Optional[str] = typer.Option(
+        None,
+        "--model",
+        help="LLM model name (overrides OPENAI_MODEL).",
+    ),
+    enable_thinking: bool = typer.Option(
+        False,
+        "--enable-thinking/--no-enable-thinking",
+        help="Pass chat_template_kwargs.enable_thinking for Qwen-compatible servers.",
+    ),
+    json_output: bool = typer.Option(False, "--json", help="Print machine-readable JSON."),
+) -> None:
+    """Parse with UniParser (unless --pages-tree), then extract VQA pairs via LLM."""
+    if answer_pdf and pages_tree:
+        raise typer.BadParameter("Use either --answer-pdf or --pages-tree, not both.")
+    if answer_pdf and not input_path:
+        raise typer.BadParameter("--answer-pdf requires the question booklet as INPUT.")
+    if not input_path and not pages_tree:
+        raise typer.BadParameter("Provide INPUT (pdf/url/image) or --pages-tree.")
+    if input_path and pages_tree:
+        raise typer.BadParameter("Use either INPUT or --pages-tree, not both.")
+
+    llm_config = _build_llm_config(
+        api_key=api_key,
+        base_url=base_url,
+        model=model,
+        enable_thinking=enable_thinking,
+    )
+    result = run_vqa_pipeline(
+        input_path=input_path,
+        answer_pdf=answer_pdf,
+        pages_tree_path=pages_tree,
+        output_dir=output_dir,
+        overwrite=overwrite,
+        llm_config=llm_config,
+    )
+    if json_output:
+        typer.echo(json.dumps(result, ensure_ascii=False, indent=2))
+        return
+
+    paths = result["paths"]
+    if paths.get("merged_pdf"):
+        typer.echo(f"Merged PDF: {paths['merged_pdf']}")
+    typer.echo(f"Pages tree: {paths['pages_tree']}")
+    typer.echo(f"Content list items: {result['n_content_items']}")
+    typer.echo(f"VQA images: {result.get('n_vqa_images', 0)} -> {paths.get('vqa_images', '')}")
+    typer.echo(f"Merged VQA pairs: {result['n_merged_vqa']}")
+    typer.echo(f"JSONL: {paths['merged_vqa_pairs_jsonl']}")
+    typer.echo(f"Markdown: {paths['merged_vqa_pairs_md']}")
+    if paths.get("vqa_sharegpt"):
+        typer.echo(f"ShareGPT: {paths['vqa_sharegpt']}")
+    typer.echo(f"Output directory: {paths['output_dir']}")
+
+
+def _build_llm_config(
+    *,
+    api_key: Optional[str],
+    base_url: Optional[str],
+    model: Optional[str],
+    enable_thinking: bool,
+) -> LLMConfig:
+    try:
+        return resolve_llm_config(
+            api_key=api_key,
+            base_url=base_url,
+            model=model,
+            enable_thinking=enable_thinking,
+        )
+    except ValueError as exc:
+        raise typer.BadParameter(str(exc)) from exc
+
+
+def main() -> None:
+    app()
+
+
+if __name__ == "__main__":
+    main()

uniparser_agent/llm/__init__.py

@@ -0,0 +1,11 @@
+"""Shared OpenAI-compatible LLM helpers for uniparser_agent."""
+
+from uniparser_agent.llm.client import OpenAICompatLLM
+from uniparser_agent.llm.config import LLMConfig, resolve_llm_config
+
+
+__all__ = [
+    "LLMConfig",
+    "OpenAICompatLLM",
+    "resolve_llm_config",
+]