Spaces:

cdpearlman
/

LLMVis

Running

App Files Files Community

LLMVis / utils

Commit History

Mar 22

00:46
Fixed ablation experiments, drilled down on which pre-loaded ones to run

cdpearlman

3d71d04

Mar 19

22:42
Bug fix: type issue with models leading to gibberish outputs

cdpearlman

5528a77

Mar 17

15:24
Rename L#-H# notation to L#-D# for detector consistency⁠

cdpearlman +1: Claude Opus 4.6

deb2071

Mar 6

20:45
Added numbers to BertViz visualization, easier to map attention head numbers

cdpearlman

67f5be3

Mar 5

01:50
added color swatches for easy mapping from attention categories to BertViz

cdpearlman

96e2bc7
01:30
Fixed model-switching bug with activation store clearing

cdpearlman

7ad3e7b

Feb 27

01:38
New models chosen and rag docs updated

cdpearlman

f478beb

Feb 26

21:08
Attention refactor, better categorization and explanation

cdpearlman

7fa8fb4

Feb 22

02:49
Chatbot improvements: UI cleaning, chat streaming, typing animation, and better markdown formatting

cdpearlman

a684a8d
02:21
ablation experiment fixes, better UI and full sequence comparison

cdpearlman

a938eeb

Feb 16

21:10
feat(output): Add token scrubber with per-position top-5 next-token probabilities⁠

cdpearlman +1: Cursor

d60cfe2

Feb 3

20:56
Switch to cost-optimized paid models: Gemini 2.5 Flash Lite + text-embedding-3-small⁠

cdpearlman +1: Cursor

129c11e
20:37
Switch to free models: Qwen3-Next chat + local embeddings⁠

cdpearlman +1: Cursor

94b077a
20:20
Migrate from Gemini to OpenRouter API⁠

cdpearlman +1: Cursor

ef3e36a
20:06
Migrate from google-generativeai to google-genai SDK⁠

cdpearlman +1: Cursor

ddd91a5
19:34
Resolve merge conflicts in README.md and todo.md⁠

cdpearlman +1: Cursor

b4d0227
18:32
feat: Add AI chatbot with Gemini integration and RAG support⁠

cdpearlman +1: Cursor

62884e7
18:18
cleanup for merge

cdpearlman

b194610
18:14
chore: remove unused code, imports, and deprecated functions⁠

cdpearlman +1: Cursor

2965a7d
17:51
conductor setup

cdpearlman

aabd66e

Jan 29

20:24
conductor(checkpoint): Checkpoint end of Phase 3: Visualization & Feedback Loop

cdpearlman

72f82a4
20:19
conductor(checkpoint): Checkpoint end of Phase 2: Frontend Control Panel

cdpearlman

24bd049
20:19
fix(utils): Export generate_bertviz_model_view_html from package

cdpearlman

5cad664
20:06
conductor(checkpoint): Checkpoint end of Phase 1: Backend Support for Ablation

cdpearlman

5fc7374
20:01
feat(pipeline): Add ablation support to beam search and verify pipeline dispatch

cdpearlman

d2ea949
19:52
feat(ablation): Unify ablation execution into execute_forward_pass

cdpearlman

2f4e4f3
04:41
Fix multi-layer ablation to ablate all heads simultaneously

cdpearlman

ae08976

Jan 26

00:13
rules change for simple git commit messages

cdpearlman

d90df7f
00:03
Switch attention viz to head_view and add categorization (Agent D)

cdpearlman

ef749ae

Jan 22

21:03
Add comprehensive test suite for utility modules (73 tests)

cdpearlman

e13b1ba

Jan 14

23:01
Refactor: Replace heatmap with pipeline explanation visualization

cdpearlman

3850656
21:30
heatmap implementation

cdpearlman

ac745db

Dec 31, 2025

05:01
Ablation updated for full sequence, needs refactor for front-end and workflow

cdpearlman

2ad1c2e

Dec 23, 2025

02:34
Fixed errors for beam search refactor

cdpearlman

c19d5a8

Dec 22, 2025

23:25
Switched to beam search instead of single token

cdpearlman

4c3d673
23:24
Fix storage quota error by using memory storage and add type safety checks

cdpearlman

cc96787

Nov 12, 2025

17:50
Better explanations, moved transformer layer diagram to be more visible

cdpearlman

080932d

Nov 11, 2025

06:05
Fix meta tensor error in safe_to_serializable⁠

cdpearlman

1f5efac
05:58
Update layer highlighting to 100% threshold for actual output token only⁠

cdpearlman

9d22c3c

Nov 10, 2025

20:44
Improve explanations: optimize head categorization and comparison thresholds, enhance ablation explanation, rewrite all tooltips for high school level

cdpearlman

b828e58
20:41
Clean up: fix significant layer threshold to 50%, remove unused logit lens parameter, remove certainty calculations

cdpearlman

a4e6de8

Nov 7, 2025

04:47
Add ablation experiment before/after comparison with Reset button⁠

cdpearlman

d701d46

Nov 5, 2025

00:12
Fix ablation: add attention_modules to session store so ablation can find attention layers

cdpearlman

efbe4ad

Nov 4, 2025

22:05
Add comprehensive debugging to ablation data flow

cdpearlman

a1ae11e
21:00
Fix ablation: capture ablated attention module output to populate layer data

cdpearlman

455f0c2
17:55
Refine layer organization: add title to diagram, move graph below layers, give Residual a box, reorder Explore button

cdpearlman

34fa23c
17:48
Reorganize Layer-by-Layer Predictions: move diagram to main container, reorder sections, update arrows, increase threshold to 75%

cdpearlman

c127d17
17:04
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

cdpearlman

b98d237
16:45
Clean up temporary documentation

cdpearlman

f64fe64
16:43
Fix token predictions not being collected - changed condition to check for block_outputs and norm_parameters instead of logit_lens_parameter

cdpearlman

d6f21db

Commit History

Mar 22

Fixed ablation experiments, drilled down on which pre-loaded ones to run

Mar 19

Bug fix: type issue with models leading to gibberish outputs

Mar 17

Rename L#-H# notation to L#-D# for detector consistency⁠

Mar 6

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

Mar 5

added color swatches for easy mapping from attention categories to BertViz

Fixed model-switching bug with activation store clearing

Feb 27

New models chosen and rag docs updated

Feb 26

Attention refactor, better categorization and explanation

Feb 22

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

ablation experiment fixes, better UI and full sequence comparison

Feb 16

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

Feb 3

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

Jan 29

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

Jan 26

rules change for simple git commit messages

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

Jan 22

Add comprehensive test suite for utility modules (73 tests)

Jan 14

Refactor: Replace heatmap with pipeline explanation visualization

heatmap implementation

Dec 31, 2025

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

Dec 23, 2025

Fixed errors for beam search refactor

Dec 22, 2025

Switched to beam search instead of single token

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

Nov 12, 2025

Better explanations, moved transformer layer diagram to be more visible

Nov 11, 2025

Fix meta tensor error in safe_to_serializable⁠

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

Nov 10, 2025

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

Nov 7, 2025

Add ablation experiment before/after comparison with Reset button⁠

Nov 5, 2025

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

Nov 4, 2025

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