Upload clean_kawmuthu.py with huggingface_hub

clean_kawmuthu.py

@@ -0,0 +1,94 @@
+"""
+Cleanup script for kawmuthu_training.jsonl
+Fixes:
+1. Strips dates (-2002, --1998, --April 2010, etc.)
+2. Removes event/institutional notes (Freshers Nite, Faculty of Engineering, SLBC, etc.)
+3. Collapses excessive whitespace/padding
+4. Fixes mismatched theme instructions
+
+Usage:
+    python clean_kawmuthu.py
+Input:  kawmuthu_training.jsonl
+Output: kawmuthu_training_clean.jsonl
+"""
+
+import json
+import re
+
+def clean_poem_content(text):
+    # Remove date patterns like -2002, --1998, --April 2010, -2021.09.17
+    text = re.sub(r'\n[-–]+\s*\d{4}[\.\d]*.*', '', text)
+    text = re.sub(r'\n[-–]+\s*[A-Za-z]+\s*\d{4}.*', '', text)
+
+    # Remove event/institutional notes
+    # Patterns like "2005 Freshers Nite souvenir(held by Batch 04,..."
+    text = re.sub(r'\n?\d{4}\s+Freshers.*?(?=\n\n|\Z)', '', text, flags=re.DOTALL)
+    # "a music teacher at Sujatha Vidyalaya..." type lines
+    text = re.sub(r'\n?,?\s*a music teacher at.*?(?=\n\n|\Z)', '', text, flags=re.DOTALL)
+    # Any remaining SLBC/UOM/Faculty references
+    text = re.sub(r'\n.*?(Faculty of Engineering|University of Moratuwa|UOM|SLBC|Ruhunu Sevaya|Sujatha Vidyalaya).*', '', text)
+    # Remove ©Author and @Author lines
+    text = re.sub(r'\n?[@©][A-Za-z\s\.]+', '', text)
+
+    # Collapse excessive whitespace (4+ spaces in a row = alignment padding in blog)
+    text = re.sub(r'[ \t]{4,}', ' ', text)
+
+    # Collapse 3+ blank lines into max 2
+    text = re.sub(r'\n{3,}', '\n\n', text)
+
+    return text.strip()
+
+def fix_theme_instruction(instruction, content):
+    """Fix cases where theme instruction doesn't match poem content."""
+    # 'සිංහල ගැන සිංහල කවියක් ලියන්න' means 'write a poem about Sinhala language'
+    # but was used for all poems regardless of theme - replace with generic
+    if instruction == 'සිංහල ගැන සිංහල කවියක් ලියන්න':
+        return 'සිංහල කවියක් ලියන්න'
+    return instruction
+
+def clean_example(example):
+    content = clean_poem_content(example['output'])
+    instruction = fix_theme_instruction(example['instruction'], content)
+
+    # Rebuild the text field from scratch
+    text = (
+        'Below is an instruction that describes a task. '
+        'Write a response that appropriately completes the request.\n\n'
+        f'### Instruction:\n{instruction}\n\n'
+        f'### Response:\n{content}'
+    )
+
+    example['output'] = content
+    example['instruction'] = instruction
+    example['text'] = text
+    return example
+
+# ── Main ────────────────────────────────────────────────────
+
+input_file  = 'kawmuthu_poems_training.jsonl'
+output_file = 'kawmuthu_poems_training_clean.jsonl'
+
+examples = []
+with open(input_file, encoding='utf-8') as f:
+    for line in f:
+        line = line.strip()
+        if line:
+            examples.append(json.loads(line))
+
+print(f'Loaded {len(examples)} examples')
+
+cleaned = [clean_example(ex) for ex in examples]
+
+# Show a before/after sample
+print('\n--- BEFORE ---')
+print(examples[0]['output'][:300])
+print('\n--- AFTER ---')
+print(cleaned[0]['output'][:300])
+
+with open(output_file, 'w', encoding='utf-8') as f:
+    for ex in cleaned:
+        f.write(json.dumps(ex, ensure_ascii=False) + '\n')
+
+print(f'\nSaved {len(cleaned)} cleaned examples to {output_file}')
+print('\nNext step:')
+print('  cat combined_training.jsonl kawmuthu_training_clean.jsonl > final_training.jsonl')