Spaces:
Sleeping
Sleeping
| import streamlit as st | |
| import base64 | |
| from reportlab.lib.pagesizes import A4 | |
| from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer, Table, TableStyle | |
| from reportlab.lib.styles import getSampleStyleSheet, ParagraphStyle | |
| from reportlab.lib import colors | |
| import io | |
| import re | |
| import fitz # PyMuPDF | |
| from PIL import Image | |
| # Set page config for wide layout and collapsed sidebar | |
| st.set_page_config(layout="wide", initial_sidebar_state="collapsed") | |
| # Initial markdown content | |
| default_markdown = """# Cutting-Edge ML Outline | |
| ## Core ML Techniques | |
| 1. 🌟 **Mixture of Experts (MoE)** | |
| - Conditional computation techniques | |
| - Sparse gating mechanisms | |
| - Training specialized sub-models | |
| 2. 🔥 **Supervised Fine-Tuning (SFT) using PyTorch** | |
| - Loss function customization | |
| - Gradient accumulation strategies | |
| - Learning rate schedulers | |
| 3. 🤖 **Large Language Models (LLM) using Transformers** | |
| - Attention mechanisms | |
| - Tokenization strategies | |
| - Position encodings | |
| ## Training Methods | |
| 4. 📊 **Self-Rewarding Learning using NPS 0-10 and Verbatims** | |
| - Custom reward functions | |
| - Feedback categorization | |
| - Signal extraction from text | |
| 5. 👍 **Reinforcement Learning from Human Feedback (RLHF)** | |
| - Preference datasets | |
| - PPO implementation | |
| - KL divergence constraints | |
| 6. 🔗 **MergeKit: Merging Models to Same Embedding Space** | |
| - TIES merging | |
| - Task arithmetic | |
| - SLERP interpolation | |
| ## Optimization & Deployment | |
| 7. 📏 **DistillKit: Model Size Reduction with Spectrum Analysis** | |
| - Knowledge distillation | |
| - Quantization techniques | |
| - Model pruning strategies | |
| 8. 🧠 **Agentic RAG Agents using Document Inputs** | |
| - Vector database integration | |
| - Query planning | |
| - Self-reflection mechanisms | |
| 9. ⏳ **Longitudinal Data Summarization from Multiple Docs** | |
| - Multi-document compression | |
| - Timeline extraction | |
| - Entity tracking | |
| ## Knowledge Representation | |
| 10. 📑 **Knowledge Extraction using Markdown Knowledge Graphs** | |
| - Entity recognition | |
| - Relationship mapping | |
| - Hierarchical structuring | |
| 11. 🗺️ **Knowledge Mapping with Mermaid Diagrams** | |
| - Flowchart generation | |
| - Sequence diagram creation | |
| - State diagrams | |
| 12. 💻 **ML Code Generation with Streamlit/Gradio/HTML5+JS** | |
| - Code completion | |
| - Unit test generation | |
| - Documentation synthesis | |
| """ | |
| # Process multilevel markdown for PDF output | |
| def markdown_to_pdf_content(markdown_text): | |
| lines = markdown_text.strip().split('\n') | |
| pdf_content = [] | |
| in_list_item = False | |
| current_item = None | |
| sub_items = [] | |
| for line in lines: | |
| line = line.strip() | |
| if not line: | |
| continue | |
| if line.startswith('# '): | |
| pass | |
| elif line.startswith('## '): | |
| if current_item and sub_items: | |
| pdf_content.append([current_item, sub_items]) | |
| sub_items = [] | |
| current_item = None | |
| section = line.replace('## ', '').strip() | |
| pdf_content.append(f"<b>{section}</b>") | |
| in_list_item = False | |
| elif re.match(r'^\d+\.', line): | |
| if current_item and sub_items: | |
| pdf_content.append([current_item, sub_items]) | |
| sub_items = [] | |
| current_item = line.strip() | |
| in_list_item = True | |
| elif line.startswith('- ') and in_list_item: | |
| sub_items.append(line.strip()) | |
| else: | |
| if not in_list_item: | |
| pdf_content.append(line.strip()) | |
| if current_item and sub_items: | |
| pdf_content.append([current_item, sub_items]) | |
| mid_point = len(pdf_content) // 2 | |
| left_column = pdf_content[:mid_point] | |
| right_column = pdf_content[mid_point:] | |
| return left_column, right_column | |
| # Main PDF creation with parameterized text sizes | |
| def create_main_pdf(markdown_text, base_font_size=10, auto_size=False): | |
| buffer = io.BytesIO() | |
| doc = SimpleDocTemplate( | |
| buffer, | |
| pagesize=(A4[1], A4[0]), | |
| leftMargin=36, | |
| rightMargin=36, | |
| topMargin=36, | |
| bottomMargin=36 | |
| ) | |
| styles = getSampleStyleSheet() | |
| story = [] | |
| page_height = A4[0] - 72 | |
| title_height = 20 | |
| spacer_height = 10 | |
| left_column, right_column = markdown_to_pdf_content(markdown_text) | |
| total_items = 0 | |
| for col in (left_column, right_column): | |
| for item in col: | |
| if isinstance(item, list): | |
| main_item, sub_items = item | |
| total_items += 1 + len(sub_items) | |
| else: | |
| total_items += 1 | |
| # 🔧 Adjust this multiplier to control autosizing sensitivity | |
| if auto_size: | |
| base_font_size = max(6, min(12, 200 / total_items)) | |
| # 🔧 Font size parameters - tweak these ratios as needed | |
| item_font_size = base_font_size | |
| subitem_font_size = base_font_size * 0.9 | |
| section_font_size = base_font_size * 1.2 | |
| title_font_size = min(16, base_font_size * 1.5) | |
| title_style = styles['Heading1'] | |
| title_style.textColor = colors.darkblue | |
| title_style.alignment = 1 | |
| title_style.fontSize = title_font_size | |
| section_style = ParagraphStyle( | |
| 'SectionStyle', | |
| parent=styles['Heading2'], | |
| textColor=colors.darkblue, | |
| fontSize=section_font_size, | |
| leading=section_font_size * 1.2, | |
| spaceAfter=2 | |
| ) | |
| item_style = ParagraphStyle( | |
| 'ItemStyle', | |
| parent=styles['Normal'], | |
| fontSize=item_font_size, | |
| leading=item_font_size * 1.2, | |
| fontName='Helvetica-Bold', | |
| spaceAfter=1 | |
| ) | |
| subitem_style = ParagraphStyle( | |
| 'SubItemStyle', | |
| parent=styles['Normal'], | |
| fontSize=subitem_font_size, | |
| leading=subitem_font_size * 1.2, | |
| leftIndent=10, | |
| spaceAfter=1 | |
| ) | |
| story.append(Paragraph("Cutting-Edge ML Outline (ReportLab)", title_style)) | |
| story.append(Spacer(1, spacer_height)) | |
| left_cells = [] | |
| for item in left_column: | |
| if isinstance(item, str) and item.startswith('<b>'): | |
| text = item.replace('<b>', '').replace('</b>', '') | |
| left_cells.append(Paragraph(text, section_style)) | |
| elif isinstance(item, list): | |
| main_item, sub_items = item | |
| left_cells.append(Paragraph(main_item, item_style)) | |
| for sub_item in sub_items: | |
| left_cells.append(Paragraph(sub_item, subitem_style)) | |
| else: | |
| left_cells.append(Paragraph(item, item_style)) | |
| right_cells = [] | |
| for item in right_column: | |
| if isinstance(item, str) and item.startswith('<b>'): | |
| text = item.replace('<b>', '').replace('</b>', '') | |
| right_cells.append(Paragraph(text, section_style)) | |
| elif isinstance(item, list): | |
| main_item, sub_items = item | |
| right_cells.append(Paragraph(main_item, item_style)) | |
| for sub_item in sub_items: | |
| right_cells.append(Paragraph(sub_item, subitem_style)) | |
| else: | |
| right_cells.append(Paragraph(item, item_style)) | |
| max_cells = max(len(left_cells), len(right_cells)) | |
| left_cells.extend([""] * (max_cells - len(left_cells))) | |
| right_cells.extend([""] * (max_cells - len(right_cells))) | |
| table_data = list(zip(left_cells, right_cells)) | |
| col_width = (A4[1] - 72) / 2.0 | |
| table = Table(table_data, colWidths=[col_width, col_width], hAlign='CENTER') | |
| table.setStyle(TableStyle([ | |
| ('VALIGN', (0, 0), (-1, -1), 'TOP'), | |
| ('ALIGN', (0, 0), (-1, -1), 'LEFT'), | |
| ('BACKGROUND', (0, 0), (-1, -1), colors.white), | |
| ('GRID', (0, 0), (-1, -1), 0, colors.white), | |
| ('LINEAFTER', (0, 0), (0, -1), 0.5, colors.grey), | |
| ('LEFTPADDING', (0, 0), (-1, -1), 2), | |
| ('RIGHTPADDING', (0, 0), (-1, -1), 2), | |
| ('TOPPADDING', (0, 0), (-1, -1), 1), | |
| ('BOTTOMPADDING', (0, 0), (-1, -1), 1), | |
| ])) | |
| story.append(table) | |
| doc.build(story) | |
| buffer.seek(0) | |
| return buffer.getvalue() | |
| # Function to convert PDF bytes to image using fitz | |
| def pdf_to_image(pdf_bytes): | |
| try: | |
| doc = fitz.open(stream=pdf_bytes, filetype="pdf") | |
| page = doc[0] | |
| pix = page.get_pixmap(matrix=fitz.Matrix(2.0, 2.0)) # 2x zoom | |
| img = Image.frombytes("RGB", [pix.width, pix.height], pix.samples) | |
| doc.close() | |
| return img | |
| except Exception as e: | |
| st.error(f"Failed to render PDF preview: {e}") | |
| return None | |
| # Sidebar for settings (collapsed by default) | |
| with st.sidebar: | |
| auto_size = st.checkbox("Auto-size text", value=True) | |
| if not auto_size: | |
| base_font_size = st.slider("Base Font Size (points)", min_value=6, max_value=16, value=10, step=1) | |
| else: | |
| base_font_size = 10 | |
| st.info("Font size will auto-adjust between 6-12 points based on content length.") | |
| # Use session state to persist markdown content | |
| if 'markdown_content' not in st.session_state: | |
| st.session_state.markdown_content = default_markdown | |
| # Generate PDF | |
| with st.spinner("Generating PDF..."): | |
| pdf_bytes = create_main_pdf(st.session_state.markdown_content, base_font_size, auto_size) | |
| # Display PDF preview in a full-width container | |
| with st.container(): | |
| pdf_image = pdf_to_image(pdf_bytes) | |
| if pdf_image: | |
| st.image(pdf_image, use_container_width=True) | |
| else: | |
| st.info("Download the PDF to view it locally.") | |
| # Download button | |
| st.download_button( | |
| label="Download PDF", | |
| data=pdf_bytes, | |
| file_name="ml_outline.pdf", | |
| mime="application/pdf" | |
| ) | |
| # Markdown editor | |
| edited_markdown = st.text_area( | |
| "Modify the markdown content below:", | |
| value=st.session_state.markdown_content, | |
| height=300 | |
| ) | |
| # Update markdown and regenerate PDF on change | |
| if st.button("Update PDF"): | |
| st.session_state.markdown_content = edited_markdown | |
| st.rerun() | |
| # Save markdown option | |
| st.download_button( | |
| label="Save Markdown", | |
| data=st.session_state.markdown_content, | |
| file_name="ml_outline.md", | |
| mime="text/markdown" | |
| ) |