paper2torch 0.1.0__tar.gz

paper2torch-0.1.0/PKG-INFO

@@ -0,0 +1,26 @@
+Metadata-Version: 2.4
+Name: paper2torch
+Version: 0.1.0
+Summary: Convert research papers to PyTorch code using LLMs
+Home-page: https://github.com/karmagodjs/paper2torch
+Author: Dhruv Kumar
+Author-email: rafftarsingh7982@gmail.com
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+Requires-Dist: groq
+Requires-Dist: pymupdf
+Requires-Dist: click
+Requires-Dist: rich
+Requires-Dist: python-dotenv
+Dynamic: author
+Dynamic: author-email
+Dynamic: classifier
+Dynamic: description-content-type
+Dynamic: home-page
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary

paper2torch-0.1.0/paper2torch/__init__.py

@@ -0,0 +1,3 @@
+__version__ = "0.1.0"
+__author__ = "Dhruv Kumar"
+__description__ = "Convert research papers to PyTorch code"

paper2torch-0.1.0/paper2torch/cli.py

@@ -0,0 +1,104 @@
+import click
+import os
+import json
+from rich.console import Console
+from rich.panel import Panel
+from rich.progress import Progress, SpinnerColumn, TextColumn
+
+# naya
+from paper2torch.parser import extract_text_from_pdf, extract_relevant_sections
+from paper2torch.extractor import extract_sections, get_core_content
+from paper2torch.generator import generate_pytorch_code
+from paper2torch.validator import validate_code
+
+console = Console()
+
+
+@click.command()
+@click.argument('pdf_path')
+@click.option('--output', '-o', default='./output', help='Output folder path')
+@click.option('--title', '-t', default='Unknown Paper', help='Paper title')
+def main(pdf_path, output, title):
+    """
+    paper2torch — Convert research papers to PyTorch code.
+    
+    Usage: python cli.py paper.pdf --output ./generated
+    """
+    
+    console.print(Panel.fit(
+        "[bold red]paper2torch[/bold red] — Research Paper to PyTorch",
+        border_style="red"
+    ))
+    
+    # Step 1: PDF parse karo
+    with Progress(
+        SpinnerColumn(),
+        TextColumn("[progress.description]{task.description}"),
+        transient=True
+    ) as progress:
+        
+        task = progress.add_task("Parsing PDF...", total=None)
+        
+        try:
+            pdf_data = extract_text_from_pdf(pdf_path)
+            console.print(f"[green]✓[/green] PDF parsed — {pdf_data['total_pages']} pages found")
+        except FileNotFoundError:
+            console.print(f"[red]✗ PDF not found: {pdf_path}[/red]")
+            return
+        
+        # Step 2: Sections extract karo
+        progress.update(task, description="Extracting sections...")
+        sections = extract_sections(pdf_data['full_text'])
+        core_content = get_core_content(sections)
+        console.print(f"[green]✓[/green] Sections extracted — {len(core_content)} chars")
+        
+        # Step 3: PyTorch code generate karo
+        progress.update(task, description="Generating PyTorch code with Gemini...")
+        result = generate_pytorch_code(core_content, title)
+        console.print(f"[green]✓[/green] PyTorch code generated")
+        
+        # Step 4: Validate karo
+        progress.update(task, description="Validating code...")
+        validation = validate_code(result['model_code'])
+        
+        if validation['passed']:
+            console.print(f"[green]✓[/green] Validation passed")
+        else:
+            console.print(f"[yellow]⚠[/yellow] Validation: {validation['summary']}")
+    
+    # Step 5: Output save karo
+    os.makedirs(output, exist_ok=True)
+    
+    # model.py save karo
+    model_path = os.path.join(output, 'model.py')
+    with open(model_path, 'w', encoding='utf-8') as f:
+        f.write(result['model_code'])
+    
+    # config.py save karo
+    config_path = os.path.join(output, 'config.py')
+    with open(config_path, 'w', encoding='utf-8') as f:
+        f.write(result['config_code'])
+    
+    # README save karo
+    readme_path = os.path.join(output, 'README_generated.md')
+    with open(readme_path, 'w', encoding='utf-8') as f:
+        f.write(f"# {title}\n\n")
+        f.write("## Generated by paper2torch\n\n")
+        f.write("### Architecture Info\n\n")
+        f.write(f"```json\n{result['architecture_info']}\n```\n\n")
+        f.write("### Validation\n\n")
+        f.write(f"- Status: {'✓ Passed' if validation['passed'] else '⚠ Issues found'}\n")
+        f.write(f"- Summary: {validation['summary']}\n")
+    
+    console.print(Panel.fit(
+        f"[bold green]Done![/bold green]\n\n"
+        f"Files saved to [cyan]{output}[/cyan]\n"
+        f"  • model.py\n"
+        f"  • config.py\n"
+        f"  • README_generated.md",
+        border_style="green"
+    ))
+
+
+if __name__ == "__main__":
+    main()

paper2torch-0.1.0/paper2torch/extractor.py

@@ -0,0 +1,54 @@
+import re
+
+def extract_sections(full_text: str) -> dict:
+    sections = {
+        "abstract": "",
+        "introduction": "",
+        "methodology": "",
+        "architecture": "",
+        "experiments": "",
+        "conclusion": ""
+    }
+    
+    section_patterns = {
+        "abstract": r"abstract",
+        "introduction": r"introduction",
+        "methodology": r"(method|methodology|approach|proposed method)",
+        "architecture": r"(architecture|model architecture|network architecture)",
+        "experiments": r"(experiment|evaluation|results)",
+        "conclusion": r"(conclusion|conclusions)"
+    }
+    
+    lines = full_text.split('\n')
+    current_section = None
+    
+    for line in lines:
+        line_lower = line.lower().strip()
+        
+        # Check if line is a section header
+        if len(line.strip()) < 60:
+            for section, pattern in section_patterns.items():
+                if re.search(pattern, line_lower):
+                    current_section = section
+                    break
+        
+        if current_section:
+            sections[current_section] += line + "\n"
+    
+    return sections
+
+
+def get_core_content(sections: dict) -> str:
+    """
+    LLM ko feed karne ke liye core content nikalta hai
+    """
+    priority = ["abstract", "methodology", "architecture", "introduction"]
+    
+    core = ""
+    for section in priority:
+        if sections.get(section):
+            core += f"\n=== {section.upper()} ===\n"
+            core += sections[section]
+    
+    # Limit to 8000 chars — Groq/Gemini context ke liye
+    return core[:8000]

paper2torch-0.1.0/paper2torch/generator.py

@@ -0,0 +1,94 @@
+import os
+from groq import Groq
+from dotenv import load_dotenv
+
+load_dotenv()
+
+client = Groq(api_key=os.getenv("GROQ_API_KEY"))
+
+
+def generate_pytorch_code(core_content: str, paper_title: str = "Unknown") -> dict:
+
+    # Step 1: Architecture extract karo
+    arch_prompt = f"""
+You are an expert ML engineer. Read this research paper content and extract the model architecture.
+
+Paper content:
+{core_content}
+
+Return ONLY a JSON with these fields:
+{{
+    "model_name": "name of the model",
+    "architecture_type": "transformer/cnn/rnn/mlp/other",
+    "key_components": ["list", "of", "components"],
+    "hyperparameters": {{"param_name": "value"}},
+    "input_format": "description of input",
+    "output_format": "description of output"
+}}
+Return only JSON, no explanation.
+"""
+
+    arch_response = client.chat.completions.create(
+        model="llama-3.3-70b-versatile",
+        messages=[{"role": "user", "content": arch_prompt}],
+        max_tokens=1000
+    )
+    arch_text = arch_response.choices[0].message.content.strip()
+
+    # Step 2: PyTorch code generate karo
+    code_prompt = f"""
+You are an expert PyTorch developer. Based on this research paper content, write a complete PyTorch implementation.
+
+Paper content:
+{core_content}
+
+Rules:
+1. Write a complete nn.Module class
+2. Include __init__ and forward methods
+3. Add clear comments explaining each component
+4. Include a config dataclass at the top
+5. Add a simple test at the bottom inside if __name__ == "__main__"
+6. Use only standard PyTorch — no external libraries
+7. Make it runnable as-is
+
+Return ONLY Python code, no explanation, no markdown backticks.
+"""
+
+    code_response = client.chat.completions.create(
+        model="llama-3.3-70b-versatile",
+        messages=[{"role": "user", "content": code_prompt}],
+        max_tokens=4000
+    )
+    code_text = code_response.choices[0].message.content.strip()
+
+    if code_text.startswith("```"):
+        lines = code_text.split('\n')
+        code_text = '\n'.join(lines[1:-1])
+
+    # Step 3: Config generate karo
+    config_prompt = f"""
+Based on this research paper, extract all hyperparameters and write a Python config dataclass.
+
+Paper content:
+{core_content[:2000]}
+
+Return ONLY a Python dataclass with all hyperparameters found. Use @dataclass decorator.
+No explanation, no markdown backticks.
+"""
+
+    config_response = client.chat.completions.create(
+        model="llama-3.3-70b-versatile",
+        messages=[{"role": "user", "content": config_prompt}],
+        max_tokens=1000
+    )
+    config_text = config_response.choices[0].message.content.strip()
+
+    if config_text.startswith("```"):
+        lines = config_text.split('\n')
+        config_text = '\n'.join(lines[1:-1])
+
+    return {
+        "architecture_info": arch_text,
+        "model_code": code_text,
+        "config_code": config_text
+    }

paper2torch-0.1.0/paper2torch/parser.py

@@ -0,0 +1,50 @@
+import fitz
+import os
+
+def extract_text_from_pdf(pdf_path: str) -> dict:
+    if not os.path.exists(pdf_path):
+        raise FileNotFoundError(f"PDF not found: {pdf_path}")
+    
+    doc = fitz.open(pdf_path)
+    
+    full_text = ""
+    pages = []
+    
+    for page_num, page in enumerate(doc):
+        text = page.get_text()
+        pages.append({
+            "page": page_num + 1,
+            "text": text
+        })
+        full_text += text + "\n"
+    
+    doc.close()
+    
+    return {
+        "full_text": full_text,
+        "pages": pages,
+        "total_pages": len(pages)
+    }
+
+
+def extract_relevant_sections(full_text: str) -> str:
+    keywords = [
+        "abstract", "introduction", "method", "methodology",
+        "architecture", "model", "approach", "proposed",
+        "network", "layer", "attention", "encoder", "decoder"
+    ]
+    
+    lines = full_text.split('\n')
+    relevant_lines = []
+    capture = False
+    
+    for line in lines:
+        line_lower = line.lower().strip()
+        
+        if any(kw in line_lower for kw in keywords) and len(line.strip()) < 60:
+            capture = True
+        
+        if capture:
+            relevant_lines.append(line)
+    
+    return '\n'.join(relevant_lines)

paper2torch-0.1.0/paper2torch/validator.py

@@ -0,0 +1,76 @@
+import ast
+import torch
+import sys
+
+
+def validate_syntax(code: str) -> dict:
+    """
+    Python syntax check karta hai
+    """
+    try:
+        ast.parse(code)
+        return {
+            "valid": True,
+            "error": None
+        }
+    except SyntaxError as e:
+        return {
+            "valid": False,
+            "error": f"Syntax error at line {e.lineno}: {e.msg}"
+        }
+
+
+def validate_pytorch(code: str) -> dict:
+    """
+    PyTorch import aur basic structure check karta hai
+    """
+    issues = []
+    
+    # torch import check
+    if "import torch" not in code:
+        issues.append("torch not imported")
+    
+    # nn.Module check
+    if "nn.Module" not in code:
+        issues.append("No nn.Module class found")
+    
+    # forward method check
+    if "def forward" not in code:
+        issues.append("No forward method found")
+    
+    # __init__ check
+    if "def __init__" not in code:
+        issues.append("No __init__ method found")
+    
+    return {
+        "valid": len(issues) == 0,
+        "issues": issues
+    }
+
+
+def validate_code(code: str) -> dict:
+    """
+    Full validation — syntax + pytorch structure
+    """
+    syntax_result = validate_syntax(code)
+    
+    if not syntax_result["valid"]:
+        return {
+            "passed": False,
+            "syntax": syntax_result,
+            "pytorch": None,
+            "summary": f"Syntax error: {syntax_result['error']}"
+        }
+    
+    pytorch_result = validate_pytorch(code)
+    
+    passed = pytorch_result["valid"]
+    
+    summary = "All checks passed!" if passed else f"Issues: {', '.join(pytorch_result['issues'])}"
+    
+    return {
+        "passed": passed,
+        "syntax": syntax_result,
+        "pytorch": pytorch_result,
+        "summary": summary
+    }

paper2torch-0.1.0/paper2torch.egg-info/PKG-INFO

@@ -0,0 +1,26 @@
+Metadata-Version: 2.4
+Name: paper2torch
+Version: 0.1.0
+Summary: Convert research papers to PyTorch code using LLMs
+Home-page: https://github.com/karmagodjs/paper2torch
+Author: Dhruv Kumar
+Author-email: rafftarsingh7982@gmail.com
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+Requires-Dist: groq
+Requires-Dist: pymupdf
+Requires-Dist: click
+Requires-Dist: rich
+Requires-Dist: python-dotenv
+Dynamic: author
+Dynamic: author-email
+Dynamic: classifier
+Dynamic: description-content-type
+Dynamic: home-page
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary

paper2torch-0.1.0/paper2torch.egg-info/SOURCES.txt

@@ -0,0 +1,14 @@
+README.md
+setup.py
+paper2torch/__init__.py
+paper2torch/cli.py
+paper2torch/extractor.py
+paper2torch/generator.py
+paper2torch/parser.py
+paper2torch/validator.py
+paper2torch.egg-info/PKG-INFO
+paper2torch.egg-info/SOURCES.txt
+paper2torch.egg-info/dependency_links.txt
+paper2torch.egg-info/entry_points.txt
+paper2torch.egg-info/requires.txt
+paper2torch.egg-info/top_level.txt

paper2torch-0.1.0/paper2torch.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

paper2torch-0.1.0/paper2torch.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+paper2torch = paper2torch.cli:main

paper2torch-0.1.0/paper2torch.egg-info/requires.txt

@@ -0,0 +1,5 @@
+groq
+pymupdf
+click
+rich
+python-dotenv

paper2torch-0.1.0/paper2torch.egg-info/top_level.txt

@@ -0,0 +1 @@
+paper2torch

paper2torch-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

paper2torch-0.1.0/setup.py

@@ -0,0 +1,35 @@
+from setuptools import setup, find_packages
+
+with open("README.md", "r", encoding="utf-8") as f:
+    long_description = f.read()
+
+setup(
+    name="paper2torch",
+    version="0.1.0",
+    author="Dhruv Kumar",
+    author_email="rafftarsingh7982@gmail.com",
+    description="Convert research papers to PyTorch code using LLMs",
+    long_description=long_description,
+    long_description_content_type="text/markdown",
+    url="https://github.com/karmagodjs/paper2torch",
+    packages=find_packages(),
+    install_requires=[
+        "groq",
+        "pymupdf",
+        "click",
+        "rich",
+        "python-dotenv",
+    ],
+    entry_points={
+        "console_scripts": [
+            "paper2torch=paper2torch.cli:main",
+        ],
+    },
+    python_requires=">=3.8",
+    classifiers=[
+        "Programming Language :: Python :: 3",
+        "License :: OSI Approved :: MIT License",
+        "Operating System :: OS Independent",
+        "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    ],
+)