#!/usr/bin/env python3
# HTML report generator for visualizing exact kept spans on the original context.
# Redesigned with a Modern/Clean aesthetic.
import json
import html
import argparse
from pathlib import Path
from difflib import SequenceMatcher
# --- Data Loading Utilities ---
def read_jsonl(path: Path):
rows = []
with path.open("r", encoding="utf-8") as f:
for line in f:
line = line.strip()
if not line:
continue
rows.append(json.loads(line))
return rows
def read_json(path: Path):
with path.open("r", encoding="utf-8") as f:
return json.load(f)
# --- Interval Logic ---
def merge_intervals(intervals):
if not intervals:
return []
intervals = sorted(intervals)
merged = [list(intervals[0])]
for s, e in intervals[1:]:
if s <= merged[-1][1]:
merged[-1][1] = max(merged[-1][1], e)
else:
merged.append([s, e])
return [(s, e) for s, e in merged]
def normalize_kept_char_spans(kept_char_spans, original_len: int):
"""Validate and merge kept character spans on the original context."""
if kept_char_spans is None:
return None
if not isinstance(kept_char_spans, (list, tuple)):
raise ValueError(f"kept_char_spans must be a list, got {type(kept_char_spans)}")
spans = []
for pair in kept_char_spans:
if not isinstance(pair, (list, tuple)) or len(pair) != 2:
raise ValueError(f"Invalid span entry: {pair!r}")
s, e = int(pair[0]), int(pair[1])
if e <= s:
continue
if s < 0 or e < 0 or s > original_len or e > original_len:
pass
# Clamp to bounds to prevent crashes
s = max(0, min(s, original_len))
e = max(0, min(e, original_len))
if s < e:
spans.append((s, e))
if not spans:
return []
return merge_intervals(spans)
def compute_kept_intervals(original: str, compressed: str, min_match_chars: int = 8):
"""Approximate kept spans via difflib matching (fallback only)."""
if not original or not compressed:
return []
sm = SequenceMatcher(a=original, b=compressed, autojunk=False)
intervals = []
for a0, b0, size in sm.get_matching_blocks():
if size >= min_match_chars:
intervals.append((a0, a0 + size))
return merge_intervals(intervals)
# --- HTML Rendering Logic ---
def render_highlight_html(original: str, kept_intervals):
"""Render the original text as HTML with kept spans highlighted."""
parts = []
cur = 0
for s, e in kept_intervals:
# Text before the kept span (Dropped)
if cur < s:
dropped = html.escape(original[cur:s])
if dropped:
parts.append(f'{dropped}')
# The kept span (Kept)
kept = html.escape(original[s:e])
if kept:
parts.append(f'{kept}')
cur = e
# Tail (Dropped)
if cur < len(original):
dropped = html.escape(original[cur:])
if dropped:
parts.append(f'{dropped}')
return "".join(parts)
def build_report_item(idx: int, qa_row: dict, res_row: dict, min_match_chars: int):
original = str(qa_row.get("input", ""))
inst = str(qa_row.get("instruction", ""))
comp = str(res_row.get("compressed_context", ""))
ratio = res_row.get("compression_ratio", None)
kept_char_spans = res_row.get("kept_char_spans", None)
kept_intervals = normalize_kept_char_spans(kept_char_spans, original_len=len(original))
if kept_intervals is None:
if not res_row.get("_allow_approx", False):
raise RuntimeError(
f"Sample #{idx}: Missing `kept_char_spans`. "
"Pass --allow_approx to use diff-based fallback."
)
kept_intervals = compute_kept_intervals(original, comp, min_match_chars=min_match_chars)
highlighted_original = render_highlight_html(original, kept_intervals)
# Statistics
kept_chars = sum(e - s for s, e in kept_intervals)
total_chars = len(original)
kept_pct = (kept_chars / total_chars * 100.0) if total_chars else 0.0
kept_spans_n = len(kept_intervals)
original_len = res_row.get("original_len", total_chars)
compressed_len = res_row.get("compressed_len", 0)
# HTML Components
inst_html = html.escape(inst)
# Logic for Badge Colors based on ratio
ratio_color_class = "neutral"
if ratio:
if ratio < 0.3: ratio_color_class = "success"
elif ratio > 0.8: ratio_color_class = "warning"
stats_html = []
if ratio is not None:
stats_html.append(f'''
Compression
{ratio:.2f}x
''')
if original_len is not None and compressed_len is not None:
stats_html.append(f'''
Tokens
{original_len} → {compressed_len}
''')
stats_html.append(f'''
Kept
{kept_pct:.1f}% ({kept_spans_n} spans)
''')
stats_block = "".join(stats_html)
item_html = f"""
Instruction / Query
{inst_html}
"""
return item_html
def build_full_html(items_html: str, title: str):
return f"""
{html.escape(title)}
{items_html}
Generated by Beaver Visualizer