Spaces:
Sleeping
Sleeping
Commit History
-
updated readme, ruff formatted all files
OliverPerrindf3ebbd -
Merge pull request #1 from Advancement
Oliver PerrinUnverified signatureAdvancement: Attention Pooling, Temperature Sampling, Comprehensive Evaluation & Code Cleanupc472a19 -
Cleaned up code, added multiseed training wrapper, PyTorch profiler training option, updated gradio demo, made changes to research paper to match new changes and new training results from adding new training techniques, architecture.md now explains all designs and decisions
OliverPerrin90a2698
-
Added many new improvements based on feedback from others
OliverPerrin0d858b5 -
Updated gradio version to fix HfFolder import error
OliverPerrin35856cb -
Updated Research Paper, README, and old gradio about info, along with other docs.
OliverPerrin8f5fea2
-
Added separate academic and research versions of the research paper
OliverPerrin4bc92d5
-
Full training results & evaluation with BERTScore
OliverPerrinTraining: - Completed 7 epochs (~6 hours) with early stopping - Optimized config: batch=10, grad_accum=4, lr=3e-5, topic_weight=0.3 - Froze bottom 4 encoder layers for stable transfer learning Evaluation Results: - Summarization: ROUGE-1=0.3064, BERTScore F1=0.8300 - Topic Classification: 85.2% accuracy (7 classes) - Emotion Detection: F1=0.1987 (28 classes, multi-label) Paper (docs/paper.tex): - Added comprehensive experimental results section - Added training dynamics figures - Fixed Unicode characters for LaTeX compatibility - Updated methodology with actual datasets used New scripts: - scripts/evaluate.py: Comprehensive evaluation with BERTScore - Rebuilt discovery dataset with 1000 samples for demo1e95f87
-
Add evaluation_report.json for metrics display
OliverPerrin7e802ad -
Add metrics tab with evaluation report, show all labels
OliverPerrin18e0afe -
Force HF Space rebuild with cache bust
OliverPerrin32b685d -
Updated Gradio Demo to match new approac for book descritpions and shown metrics
OliverPerrinf1ab267 -
Medium training run
OliverPerrinb93250a -
Fix literary summaries: exclude plays/epics, use human summaries from BookSum
OliverPerrin- Expanded EXCLUDED_TITLES to include French plays (Tartuffe, Phaedra, Cyrano), Greek/Roman plays, and poetic epics (Idylls of the King, Paradise Lost, etc.) - Demo now shows 'Human Summary (from BookSum)' vs 'AI-Generated Summary' - Literary content uses reference_summary from BookSum instead of model inference - Academic papers still use AI-generated summaries (model performs well on these)f1cb860 -
Fix literary summaries: use only BookSum data (not Gutenberg excerpts)
OliverPerrinThe model was hallucinating garbage ('slender, snarling, smoky...') for Gutenberg excerpts because: - Gutenberg data was only used for language modeling training - Model was NEVER trained to summarize random Gutenberg paragraphs - Only BookSum literary content has proper chapter→summary training Changes: - Removed load_books() (Gutenberg) from discovery dataset - Use only load_literary() (BookSum) + load_academic_papers() (arXiv) - Increased both to 300 samples each (600 total) - Updated demo description to reflect actual sourcesd57b866 -
Medium training run
OliverPerrin277b4d1
-
Fix regression: restore balanced training settings, add technical manual filter
OliverPerrinTraining configs: - Restored LR: 3e-5 (medium), 2e-5 (full) - Restored weight decay: 0.01 - Restored label smoothing: 0.1 - Restored dropout: 0.1 - Restored task weights: emotion=1.5, topic=0.5 - Restored epochs: 5 (medium), 8 (full) - Restored early stopping patience: 3 Discovery dataset: - Added technical manual filter (pottery recipes, camera specs, etc.) - These non-narrative texts produce garbage summaries9710200 -
Medium training run
OliverPerrin8a54fec -
Major improvements: real titles, anti-overfitting, better demo UX
OliverPerrinTraining configs (medium.yaml, full.yaml): - Reduced epochs: 5→4, 8→6 (stop before overfitting) - Lower LR: 3e-5→2e-5, 2e-5→1.5e-5 - Increased weight decay: 0.01→0.02 - Increased label smoothing: 0.1→0.15 - Boost emotion weight: 1.5→2.0 (tends to underfit) - Reduce topic weight: 0.5→0.3 (overfits quickly) - Earlier early stopping: patience 3→2 - Increased dropout: 0.1→0.15 Data download (download_data.py): - BookSum: Extract real book titles and chapter names - arXiv: Extract paper titles from abstract first sentence - Gutenberg: Parse JSON metadata for real titles and authors Discovery dataset (build_discovery_dataset.py): - Use real titles from source data - Increased default sizes: 300 books, 200 papers, 100 literary - Better fallback for missing titles - Improved error handling Demo (demo_gradio.py): - Cleaner card format with prominent titles - Confidence badges (green/yellow/red) - Dynamic counts in header - Better table layout for topic/emotion0fe274c -
Merge remote-tracking branch 'refs/remotes/origin/main'
OliverPerrinff3a7e9 -
Medium training run
OliverPerrine12987d
-
Changed license
Oliver PerrinUnverified signatureffbcac3 -
New trianing run on english only data
OliverPerrin40cadca -
Add English language filter to all data downloads
OliverPerrin- Filter BookSum, arXiv, and Gutenberg for English only - Common English word ratio check (min 8%) - Non-English pattern detection (French, German, Spanish, Italian, Latin) - Skip non-English content at download time, not discovery time - Reports skipped count for each source8573220 -
Updated Training run, fixed dataset langauge issue
OliverPerrine3422d2 -
Improve discovery dataset quality
OliverPerrin- Add text quality filters (English only, no metadata/errata) - Add keyword-based emotion fallback for undertrained model - Increase sample size to 300 (150 books, 100 papers, 50 literary) - Now has diverse emotions: sadness, joy, anger, fear, etc. - Topic distribution: Science (107), Philosophy (100), History (46), Fiction (31)5c41b92 -
Demo requirements bug fix
OliverPerrin0eab5ef -
Fix type hints in demo_gradio.py for Pylance
OliverPerrin6d135aa -
Trigger rebuild
OliverPerrin6f4d4de -
Use HF Datasets for discovery demo
OliverPerrin- Created build_discovery_dataset.py to generate curated sample - Uploaded 200 items to OliverPerrin/LexiMind-Discovery dataset - Rewrote demo_gradio.py to load from HF Hub instead of local file - Removed catalog.json (no longer needed) - Demo now works on HF Space without path issues218e2b1 -
Fix catalog.json path resolution for HF Space
OliverPerrina484623 -
Redesign Gradio demo as book/paper discovery tool
OliverPerrin- Browse by Topic: Fiction, Science, Technology, Philosophy, History, Business, Arts - Browse by Emotion: Filter by emotional tone (fear, love, curiosity, etc.) - Pre-analyzed catalog with summaries, topics, and emotions - Clean, simple interface focused on discovery - Remove unnecessary training visualization complexity061c1e7 -
Fix Gradio demo metrics display and visualization script MLflow URI
OliverPerrin- Update demo to show training metrics from history JSON - Add text-based loss/accuracy progress display - Fix MLflow tracking URI to use SQLite database - Update Architecture tab with correct training data info - Add dev checkpoint note to demo header4bda87e -
Updated the model with new bug fixes
OliverPerrin5c8762c -
Fixed compiling issue, added legnth penalty, and atttempting freezing encoder layers 0-5 to lower parameters and preserve T5's langauge understanding.
OliverPerrinbaf3026
-
Optimize Docker for CPU-only (10x faster builds)
OliverPerrin4667b2a -
Remove custom theme - use Gradio default for compatibility
OliverPerrin13f7ac3 -
Ignore mlruns.db binary file
OliverPerrinc5c58be -
Minor fixes to Gradio and visualization scripts
OliverPerrin2ce1629 -
Improve emotion task architecture and fix visualizations
OliverPerrinArchitecture improvements: - Added hidden_dim option to ClassificationHead for 2-layer MLP - Emotion head now uses 768→384→28 MLP for better multi-label capacity - Updated task weights: emotion=1.5, summarization=1.0, topic=0.8 Visualization fixes: - LR schedule now generates theoretical curve when no metrics logged - Added learning_rate logging to MLflow in trainer - Log warmup_steps and scheduler_type to MLflow params Gradio UI modernization: - New yellow/black research theme (lexitheme) - Removed emojis for cleaner academic presentation - Progress bar style emotion scores - Updated tab names and descriptions6a7a381 -
Clean up codebase and fix training bugs
OliverPerrin- Fixed total_loss tracking for validation (early stopping works now) - Lowered emotion threshold from 0.5 to 0.3 for multi-label - Optimized full.yaml config for faster training (50k samples cap) - Consolidated utils into core.py - Removed unused modules and scripts - Fixed Gutenberg download key format issue - Updated visualize_training default model name - Switch MLflow to SQLite backend1601799
-
Fix Gradio demo: whole-book summaries, emotion sorting, styling
OliverPerrinebf2964 -
Add demo data for HF Space (books + news samples)
OliverPerrin22d8c0b -
Redesign Gradio demo with book/news browsing, update .gitignore
OliverPerrinfc64ea0
-
Update Gradio demo, inference factory, and evaluation results
OliverPerrin38eb401 -
Update Gradio demo, inference factory, and evaluation results
OliverPerrine3d3925
-
Update LexiMind: improved training, model architecture, and evaluation
OliverPerrin076bc18
-
Improve summarization output quality
OliverPerrin- Add _format_summary() for proper capitalization and punctuation - Add repetition_penalty (1.2) to reduce repetitive outputs - Fix period spacing and sentence capitalization - Remove leading special characters from generated text69b8f98 -
Expand datasets: GoEmotions (28 emotions), Yahoo Answers (10 topics), BookSum
OliverPerrin- Updated download_data.py with new dataset downloaders - Updated preprocess_data.py to handle JSONL format - Updated labels.json with 28 emotions and 10 topics - Updated datasets.yaml with new dataset configs - Updated full.yaml training config (warmup_steps: 1000) - Training results: 74% topic accuracy, ROUGE 0.2849becd3c -
Fix: Regenerate tokenizer without add_prefix_space to fix CI
OliverPerrin97abce2