add transformers

README.md

@@ -14,6 +14,7 @@
 - [Jax](https://jax.readthedocs.io/en/latest/jax.html)
 - [Ray](https://docs.ray.io/en/latest/genindex.html)
 - [Langchain](https://api.python.langchain.com/en/latest/langchain_api_reference.html)
+- [Hugging Face Transformers](https://huggingface.co/docs/transformers/index)
 
 ## Install
 Checkout the [Releases](https://github.com/lsgrep/mldocs/releases), download the latest `mldocs.alfredworkflow`,

crawler/generate_ml_keywords.py

@@ -1,9 +1,11 @@
 import json
 import re
 from pathlib import Path
+from typing import Dict, Any
 
 import requests
 import yaml
+from bs4 import BeautifulSoup
 
 data_dir = f'{str(Path(__file__).resolve().parent.parent)}/data'
 
@@ -73,23 +75,287 @@ def parse_generated_docs(link, pattern=None):
                 doc_url = f'{base_url}/{href}'
                 metadata = {'url': doc_url}
                 data[k] = metadata
+    return data
+
+
+def parse_huggingface_main_classes() -> Dict[str, Any]:
+    """Parse Hugging Face Transformers main API classes documentation.
+
+    Returns:
+        Dictionary mapping class names to their documentation URLs
+    """
+    data = {}
+    base_domain = 'https://huggingface.co'
+
+    # Get main class pages from the API reference page
+    print('\nFetching main classes from API reference...')
+    resp = requests.get(f'{base_domain}/docs/transformers/main_classes/configuration')
+    soup = BeautifulSoup(resp.text, 'html.parser')
+
+    # Find the sidebar navigation
+    main_class_pages = []
+    for link in soup.select('a'):
+        href = link.get('href', '')
+        if '/main_classes/' in href:
+            page_name = href.split('/')[-1]
+            if page_name and page_name not in main_class_pages:
+                main_class_pages.append(page_name)
+
+    # Remove duplicates and sort
+    main_class_pages = sorted(set(main_class_pages))
+
+    print(f'\nFound main class pages: {main_class_pages}')
+    print(f'Processing {len(main_class_pages)} main class pages...')
+
+    # Process each main class page
+    for page in main_class_pages:
+        page_url = f'{base_domain}/docs/transformers/main_classes/{page}'
+
+        try:
+            resp = requests.get(page_url)
+            soup = BeautifulSoup(resp.text, 'html.parser')
+
+            # Find all class sections (h2 and h3 headers)
+            class_sections = soup.find_all(['h2', 'h3'])
+            current_class = None
+
+            # Track links for summary
+            page_links = set()
+            page_classes = set()
+
+            # Process all links first
+            for a in soup.find_all('a'):
+                href = a.get('href', '')
+                if href and '#transformers.' in href:
+                    page_links.add(href)
+
+            # Process class sections
+            for section in class_sections:
+                # Look for class name in the text
+                class_match = re.search(r'class transformers\.(\w+)', section.text)
+                if class_match:
+                    current_class = class_match.group(1)
+                    class_key = f'transformers.{current_class}'
+                    page_classes.add(class_key)
+
+                    data[class_key] = {
+                        'url': f'{page_url}#transformers.{current_class}'
+                    }
+                    continue
+
+                section_id = section.get('id', '')
+                if not section_id or section_id.startswith('_') or section_id.endswith('_'):
+                    continue
+
+                # If we're in a class context, this might be a method
+                if current_class:
+                    method_key = f'transformers.{current_class}.{section_id}'
+                    data[method_key] = {
+                        'url': f'{page_url}#transformers.{current_class}.{section_id}'
+                    }
+
+            # Print summary for this page
+            print(f'\n{page.upper()}:')
+            if page_classes:
+                print(f'  Classes ({len(page_classes)}): {sorted(page_classes)}')
+            print(f'  Total links: {len(page_links)}')
+
+        except Exception as e:
+            print(f'Error processing {page_url}: {str(e)}')
+            continue
+
+    return data
 
+def parse_huggingface_docs(base_url: str, test_mode: bool = False) -> Dict[str, Any]:
+    """Parse Hugging Face Transformers documentation.
+
+    Args:
+        base_url: The base URL of the Hugging Face documentation
+
+    Returns:
+        Dictionary mapping function/class names to their documentation URLs
+    """
+    data = {}
+    base_domain = 'https://huggingface.co'
+
+    # Get the main page content
+    resp = requests.get(base_url)
+    soup = BeautifulSoup(resp.text, 'html.parser')
+
+    # Find all model documentation links
+    model_links = []
+    for link in soup.find_all('a'):
+        href = link.get('href')
+        if href and 'model_doc' in href:
+            # Handle both absolute and relative URLs
+            if href.startswith('http'):
+                full_url = href
+            elif href.startswith('//'):
+                full_url = f'https:{href}'
+            else:
+                # Handle relative paths
+                if href.startswith('/'):
+                    full_url = f'{base_domain}{href}'
+                else:
+                    # Construct URL relative to the docs base path
+                    docs_base = '/'.join(base_url.split('/')[:-1])
+                    full_url = f'{docs_base}/{href}'
+            model_links.append(full_url)
+
+    # For testing, only process ZoeDepth
+    if test_mode:
+        model_links = ['https://huggingface.co/docs/transformers/model_doc/zoedepth']
+
+    # Process each model's documentation
+    for model_url in model_links:
+        try:
+            print(f'Processing {model_url}')
+            print('Found model links:', len(model_links))
+            model_resp = requests.get(model_url)
+            model_soup = BeautifulSoup(model_resp.text, 'html.parser')
+
+            # Extract model name from URL
+            model_name = model_url.split('/')[-1].replace('-', '_')
+
+            # Add the main model entry
+            model_key = f'transformers.{model_name}'
+            print(f'Adding model entry: {model_key}')
+            data[model_key] = {
+                'url': model_url
+            }
+
+            # Find all class definitions (h2 and h3 headers)
+            class_sections = model_soup.find_all(['h2', 'h3'])
+            current_class = None
+
+            for section in class_sections:
+                # Look for the class name in the text
+                class_match = re.search(r'class transformers\.([\w]+)', section.text)
+                if class_match:
+                    current_class = class_match.group(1)
+                    class_key = f'transformers.{model_name}.{current_class}'
+                    print(f'Found class: {class_key}')
+
+                    data[class_key] = {
+                        'url': f'{model_url}#transformers.{current_class.split(".")[-1]}'
+                    }
+                    continue
+
+                section_id = section.get('id', '')
+                if not section_id or section_id.startswith('_') or section_id.endswith('_'):
+                    continue
+
+                # If we're in a class context, this might be a method
+                if current_class:
+                    current_class = section_id
+                    class_key = f'transformers.{model_name}.{current_class}'
+                    print(f'Found class: {class_key}')
+
+                    # Get class description
+                    desc = ''
+                    next_p = section.find_next('p')
+                    if next_p:
+                        desc = next_p.text
+
+                    data[class_key] = {
+                        'url': f'{model_url}#{section_id}',
+                        'desc': desc
+                    }
+
+                    # Find and add all methods in this class
+                    method_sections = section.find_next(['h4', 'h5'])
+                    while method_sections and method_sections.find_previous(['h2', 'h3']) == section:
+                        method_id = method_sections.get('id', '')
+                        if method_id and not method_id.startswith('_'):
+                            method_key = f'{class_key}.{method_id}'
+
+                            # Get method description and parameters
+                            desc = ''
+                            params = []
+                            next_elem = method_sections.find_next(['p', 'ul'])
+                            while next_elem and next_elem.name in ['p', 'ul']:
+                                if next_elem.name == 'p':
+                                    desc += next_elem.text + ' '
+                                elif next_elem.name == 'ul':
+                                    for li in next_elem.find_all('li'):
+                                        params.append(li.text)
+                                next_elem = next_elem.find_next(['p', 'ul'])
+
+                            data[method_key] = {
+                                'url': f'{model_url}#{method_id}',
+                                'desc': desc.strip(),
+                                'params': params
+                            }
+
+                        method_sections = method_sections.find_next(['h4', 'h5'])
+
+                # If we're in a class context, this might be a method
+                elif current_class:
+                    method_key = f'transformers.{model_name}.{current_class}.{section_id}'
+
+                    data[method_key] = {
+                        'url': f'{model_url}#transformers.{current_class.split(".")[-1]}.{section_id}'
+                    }
+
+        except Exception as e:
+            print(f'Error processing {model_url}: {str(e)}')
+            continue
+
     return data
 
 
 if __name__ == '__main__':
-    data = prepare_base_keywords()
+    # Load existing data if it exists
+    doc_file = f'{data_dir}/ml.json'
+    try:
+        with open(doc_file, 'r') as f:
+            data = json.load(f)
+        print(f'Loaded {len(data)} existing entries')
+    except FileNotFoundError:
+        data = {}
+
+    # Add base keywords if not present
+    base_data = prepare_base_keywords()
+    for key, value in base_data.items():
+        if key not in data:
+            data[key] = value
+
     seed_file = f'{data_dir}/seed.yaml'
     seed = load_seed_file(seed_file)
+
+    # Process TensorFlow docs
     for tensorflow_doc in seed['tensorflow']:
         print(f'processing: {tensorflow_doc["name"]}')
         crawled = parse_tf_docs(tensorflow_doc['url'], tensorflow_doc['prefix'])
         data.update(crawled)
+
+    # Process generated docs
     for api_doc in seed['generated']:
         print(f'processing: {api_doc["name"]}')
         doc_url = api_doc['url']
-        data.update(parse_generated_docs(doc_url))
+
+        # Special handling for Hugging Face Transformers
+        if api_doc['name'] == 'transformers':
+            print('Processing Hugging Face Transformers documentation...')
+            crawled = parse_huggingface_docs(doc_url, test_mode=False)
+            print(f'Crawled model data keys: {list(crawled.keys())}')
+            # Preserve Hugging Face model entries
+            for key, value in crawled.items():
+                data[key] = value
+
+            print('\nProcessing Hugging Face main API classes...')
+            main_classes = parse_huggingface_main_classes()
+            print('\nMain API Classes found:')
+            for key, value in main_classes.items():
+                print(f'  {key} -> {value["url"]}')
+            # Add main API class entries
+            for key, value in main_classes.items():
+                data[key] = value
+        else:
+            data.update(parse_generated_docs(doc_url))
 
-    doc_file = f'{data_dir}/ml.json'
+    print(f'Writing {len(data)} entries to {doc_file}')
+    print('Sample keys:', list(data.keys())[:5], '...')
+    print('Hugging Face keys:', [k for k in data.keys() if k.startswith('transformers.')])
     with open(doc_file, 'w') as f:
-        json.dump(data, f)
+        json.dump(data, f, indent=2)

data/seed.yaml

@@ -36,4 +36,6 @@ generated: # API docs those are generated by tools
   - name: 'xgboost'
     url: 'https://xgboost.readthedocs.io/en/stable/genindex.html'
   - name: 'langchain'
-    url: 'https://api.python.langchain.com/en/latest/langchain_api_reference.html'
+    url: 'https://api.python.langchain.com/en/latest/langchain_api_reference.html'
+  - name: 'transformers'
+    url: 'https://huggingface.co/docs/transformers/index'