SLM-RAG-Arena / utils /context_processor.py
oliver-aizip's picture
update data pipeline
347797e
import re
import html
import json
def clean_text(text):
"""Clean text with common issues like HTML entities and escaped quotes."""
if not text or not isinstance(text, str):
return text
# Fix incomplete HTML entities
incomplete_entities = {'&#x27': ''', '&quot': '"', '&lt': '<', '&gt': '>', '&amp': '&'}
for incomplete, complete in incomplete_entities.items():
text = re.sub(f"{re.escape(incomplete)}(?!;)", complete, text)
# Convert HTML entities to characters
try:
text = html.unescape(text)
except Exception:
pass
# Handle escaped quotes and other special characters
replacements = {
r'\"': '"', r"\'": "'", r"\n": "\n", r"\t": "\t", r"\\": "\\",
# Also normalize fancy quotes
'"': '"', '"': '"', ''': "'", ''': "'", '`': "'", '´': "'"
}
for pattern, replacement in replacements.items():
text = text.replace(pattern, replacement)
# Remove trailing backslash if present
if text.rstrip().endswith('\\'):
text = text.rstrip().rstrip('\\')
return text
def get_context_html(example, show_full=False):
"""Format context chunks into HTML for display."""
html_output = ""
# Process insufficient context warning if needed
if example.get("insufficient", False):
insufficient_reason = example.get("insufficient_reason", "")
reason_html = f"<p>{insufficient_reason}</p>" if insufficient_reason else "<p>The context may not contain enough information to fully answer the question, or the question might be ambiguous. Models should ideally indicate this limitation or refuse to answer.</p>"
html_output += f"""
<div class="insufficient-alert">
<strong>
<svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" style="vertical-align: middle; margin-right: 5px;">
<path d="m21.73 18-8-14a2 2 0 0 0-3.48 0l-8 14A2 2 0 0 0 4 21h16a2 2 0 0 0 1.73-3Z"></path>
<line x1="12" y1="9" x2="12" y2="13"></line>
<line x1="12" y1="17" x2="12.01" y2="17"></line>
</svg>
Insufficient Context
</strong>
{reason_html}
</div>
"""
html_output += '<div class="context-items-container">'
# Display full contexts or highlighted contexts based on toggle
if show_full:
# Show full context - directly use the strings from the list in full_contexts
if "full_contexts" in example and example["full_contexts"]:
for context_item in example["full_contexts"]:
if isinstance(context_item, dict) and 'content' in context_item:
content = context_item.get('content', '')
elif isinstance(context_item, str):
content = context_item
else:
content = str(context_item)
# Escape HTML entities for safe display
escaped_content = html.escape(content)
# Create the context item box - no headers
html_output += f'<div class="context-item">{escaped_content}</div>'
else:
# Show highlighted contexts
if "contexts" in example and example["contexts"]:
for context_item in example["contexts"]:
if isinstance(context_item, dict):
content = context_item.get('content', '')
is_primary = context_item.get('is_primary', False)
# Extra class for primary context styling
extra_class = " primary-context" if is_primary else ""
# Use content directly as it already has HTML highlighting
html_output += f'<div class="context-item{extra_class}">{content}</div>'
elif isinstance(context_item, str):
# For direct string contexts
html_output += f'<div class="context-item">{context_item}</div>'
else:
html_output += '<div class="context-item">No context available. Try toggling to full context view.</div>'
html_output += '</div>'
return html_output