Spaces:
Running
Running
Commit History
-
Tutorial, feedback, and github buttons added
e1f91c2
-
Fixed slider loading bug
4416500
-
Updated and partitioned glossary
e2d10e5 -
Moved glossary button to header instead of hidden in sidebar
42ae447
-
Added simpler ablation experiment buttons to add full categories
3c5f2fd -
Added pre-built prompts for app and chatbot
018a9f2
-
Bug fix: type issue with models leading to gibberish outputs
5528a77 -
Replace token references with word in output selection stage
Updates scrubber labels, chart titles, disclaimers, and static fallback text to use "word" instead of "token" for consistency with jargon reduction. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>ac0335e
-
Rename L#-H# notation to L#-D# for detector consistency
Updates the head identifier labels from H (head) to D (detector) in ablation chips, results display, and pipeline detector labels to match the jargon-reduction renaming. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>deb2071
-
Reduce jargon across UI for younger/less-technical audiences
Replace technical ML terminology with plain-English equivalents while preserving technical terms as parentheticals for educational value: - Tokenization → Text Splitting - Embedding/vector → Meaning Encoding/list of numbers - MLP (Feed-Forward) → Knowledge Retrieval - Ablation → Test by Removing - Token Attribution → Word Influence - Attention heads → detectors (with "heads" as technical note) - Activation score → activity level - Probability shift → confidence change Updated glossary entries to lead with friendly names. Chatbot suggestions also updated for consistency. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>abf6a1c
-
Fixed ablation graphs first token population, still need to fix original graphs when no beam selected
670d882
-
Fixed duplicate activation store bug
78cc5b4 -
added color swatches for easy mapping from attention categories to BertViz
96e2bc7
-
display attention category scores in order
c629c1f -
completed the rest of the explanation improvements from JARVIS audit
ff1ba74 -
New models chosen and rag docs updated
f478beb
-
Attention refactor, better categorization and explanation
7fa8fb4
-
Fixed glossary UI to be cleaner
1cce926 -
Added glossary embeddings and explanations
2800225
-
ablation experiment fixes, better UI and full sequence comparison
a938eeb
-
feat(output): Add token scrubber with per-position top-5 next-token probabilities
Co-authored-by: Cursor <cursoragent@cursor.com>d60cfe2 -
Add drag-to-resize on chat window left edge
Co-authored-by: Cursor <cursoragent@cursor.com>718dfcb
-
Added message greeting to chatbot
b6266bb -
Reset chat history on browser refresh to show greeting
Co-authored-by: Cursor <cursoragent@cursor.com>cbe8dbf -
Add greeting message to chatbot on first open
Co-authored-by: Cursor <cursoragent@cursor.com>678a73d
-
Resolve merge conflicts in README.md and todo.md
Co-authored-by: Cursor <cursoragent@cursor.com>b4d0227 -
feat: Add AI chatbot with Gemini integration and RAG support
Co-authored-by: Cursor <cursoragent@cursor.com>62884e7 -
cleanup for merge
b194610 -
chore: remove unused code, imports, and deprecated functions
- Delete components/tokenization_panel.py (superseded by pipeline.py) - Remove 6 unused imports from app.py - Remove deprecated _get_top_attended_tokens() and references - Remove unused create_stage_summary() from pipeline.py - Remove 7 unused utility functions from model_patterns.py and beam_search.py - Update utils/__init__.py exports and README.md Total: ~1,087 lines removed. All 81 tests pass. Co-authored-by: Cursor <cursoragent@cursor.com>2965a7d -
conductor setup
aabd66e
-
conductor(checkpoint): Checkpoint end of Phase 2: Frontend Control Panel
24bd049
-
Fix ablation head selection: handle old data format, prevent spurious selections on render
bb577e6 -
Fix 5 UI issues: comparison button, tokenization layout, expandable attention categories, BertViz instructions, multi-layer ablation
0f46e9c -
rules change for simple git commit messages
d90df7f -
Refactor analysis scope and enhance attention UI (Agents F and G)
8a1fc4b -
Add educational explanations to pipeline stages (Agent C): embedding lookup table, MLP knowledge storage, attention viz guide
3e6c5b6 -
Improve UI clarity: rename generation settings labels and remove score display
d134e9a
-
Add comprehensive test suite for utility modules (73 tests)
e13b1ba
-
Refactor: Replace heatmap with pipeline explanation visualization
3850656 -
heatmap implementation
ac745db
-
Ablation updated for full sequence, needs refactor for front-end and workflow
2ad1c2e
-
Fixed errors for beam search refactor
c19d5a8
-
Switched to beam search instead of single token
4c3d673 -
Fix storage quota error by using memory storage and add type safety checks
cc96787
-
Update layer highlighting to 100% threshold for actual output token only
- Changed detect_significant_probability_increases to only check actual output token instead of all top 5 tokens - Updated threshold from 50% to 100% (doubled probability) for highlighting layers - Modified compute_layer_wise_summaries to accept activation_data and extract actual output token - Updated all calls to compute_layer_wise_summaries in app.py to pass activation_data - Added _create_actual_output_display helper function to show actual model output - Display actual output token below final output graphs (normal mode, ablation mode, comparison mode) - Added tooltip explaining why actual output may differ from highest probability token in final layer due to residual connections - Updated graph tooltip text to reflect 100% threshold change9d22c3c
-
Improve explanations: optimize head categorization and comparison thresholds, enhance ablation explanation, rewrite all tooltips for high school level
b828e58 -
Clean up: fix significant layer threshold to 50%, remove unused logit lens parameter, remove certainty calculations
a4e6de8
-
Add ablation experiment before/after comparison with Reset button
- Added session-activation-store-original to preserve pre-ablation data - Modified run_head_ablation callback to save original data before ablation - Updated create_layer_accordions to detect ablation mode and show before/after sections for all layers - Modified _create_single_prompt_chart and _create_token_probability_delta_chart to accept title suffixes - Added before/after comparison for line graphs showing top 5 token probabilities across layers - Added Reset Ablation button in sidebar (visible only during ablation) - Implemented reset_ablation callback to restore original data and clear ablation - Added visual polish with section headers, background colors (blue for before, orange for after), and explanatory tooltips - All changes maintain backward compatibility with non-ablation moded701d46
-
Fix token predictions not being collected - changed condition to check for block_outputs and norm_parameters instead of logit_lens_parameter
d6f21db
-
Implement PHASE 3: Tokenization Explanation Section - Created components/tokenization_panel.py with three-column layout - Added HTML/CSS-based static tokenization diagram - Integrated tokenization panel into main_panel.py - Added update_tokenization_display() callback - Added CSS styling with hover effects and connecting lines - Implemented comparison mode support with vertical stacking
f0bb174