Added numbers to BertViz visualization, easier to map attention head numbers

added color swatches for easy mapping from attention categories to BertViz

Fixed model-switching bug with activation store clearing

New models chosen and rag docs updated

Attention refactor, better categorization and explanation

Chatbot improvements: UI cleaning, chat streaming, typing animation, and better markdown formatting

ablation experiment fixes, better UI and full sequence comparison

feat(output): Add token scrubber with per-position top-5 next-token probabilities

Switch to cost-optimized paid models: Gemini 2.5 Flash Lite + text-embedding-3-small

Switch to free models: Qwen3-Next chat + local embeddings

Migrate from Gemini to OpenRouter API

Migrate from google-generativeai to google-genai SDK

Resolve merge conflicts in README.md and todo.md

feat: Add AI chatbot with Gemini integration and RAG support

cleanup for merge

chore: remove unused code, imports, and deprecated functions

conductor setup

conductor(checkpoint): Checkpoint end of Phase 3: Visualization & Feedback Loop

conductor(checkpoint): Checkpoint end of Phase 2: Frontend Control Panel

fix(utils): Export generate_bertviz_model_view_html from package

conductor(checkpoint): Checkpoint end of Phase 1: Backend Support for Ablation

feat(pipeline): Add ablation support to beam search and verify pipeline dispatch

feat(ablation): Unify ablation execution into execute_forward_pass

Fix multi-layer ablation to ablate all heads simultaneously

rules change for simple git commit messages

Switch attention viz to head_view and add categorization (Agent D)

Add comprehensive test suite for utility modules (73 tests)

Refactor: Replace heatmap with pipeline explanation visualization

heatmap implementation

Ablation updated for full sequence, needs refactor for front-end and workflow

Fixed errors for beam search refactor

Switched to beam search instead of single token

Fix storage quota error by using memory storage and add type safety checks

Better explanations, moved transformer layer diagram to be more visible

Fix meta tensor error in safe_to_serializable

Update layer highlighting to 100% threshold for actual output token only

Improve explanations: optimize head categorization and comparison thresholds, enhance ablation explanation, rewrite all tooltips for high school level

Clean up: fix significant layer threshold to 50%, remove unused logit lens parameter, remove certainty calculations

Add ablation experiment before/after comparison with Reset button

Fix ablation: add attention_modules to session store so ablation can find attention layers

Add comprehensive debugging to ablation data flow

Fix ablation: capture ablated attention module output to populate layer data

Refine layer organization: add title to diagram, move graph below layers, give Residual a box, reorder Explore button

Reorganize Layer-by-Layer Predictions: move diagram to main container, reorder sections, update arrows, increase threshold to 75%

Fix norm_selection type mismatch and ensure norm_parameters are always selected - Changed auto-selection to return list instead of string - Updated run_analysis to handle norm_patterns as list - Added norm_parameters requirement to Run Analysis button

Clean up temporary documentation

Fix token predictions not being collected - changed condition to check for block_outputs and norm_parameters instead of logit_lens_parameter

feat: implement PHASE 6 (BertViz button relocation) and PHASE 7 (CSS styling)

Phase 1: Backend infrastructure for token merging, layer-wise tracking, and head ablation

merge for ablation experiments