SkinDeepRESEARCHSteve Seguin

Browser test results

Recorded browser comparisons

Saved BERT comparisons and the standalone tiny-classifier check.

These are saved runs of the BERT browser demo: 100 previously inspected ToxicChat messages, two reversed-order passes per comparison. Time includes input preparation and inference, but excludes loading and warmup. Each row belongs to its own paired run.

These are exploratory measurements on one computer, not fresh validation or a promise of the same speed on your device.

Fixed depth
PathCorrect / 100Time / 100Mean blocks runChanged answers
All 4 layers831.82 s4.000
Stop after layer 2821.01 s2.005

Run this comparison · Related Qwen study

Detailed evidence

Every decision, timing and model setting · Run protocol and source hashes · Implementation and limitations

Check before continuing
PathCorrect / 100Time / 100Mean blocks runChanged answers
All 4 layers831.90 s4.000
Check after layer 2831.22 s2.560

Run this comparison · Related Qwen study

Detailed evidence

Every decision, timing and model setting · Run protocol and source hashes · Implementation and limitations

Small model, then larger model
PathCorrect / 100Time / 100Mean blocks runChanged answers
All 4 layers831.87 s4.000
Tiny BERT, then larger BERT820.77 s2.843

The fallback runs two independent BERT models. Their blocks have different widths; the sum is not a compute-saving percentage. The 0.05/0.95 gate was not validated for this pairing.

Run this comparison · Related Qwen study

Detailed evidence

Every decision, timing and model setting · Run protocol and source hashes · Implementation and limitations

Compare trained weights
PathCorrect / 100Time / 100Mean blocks runChanged answers
All 4 layers831.98 s4.000
Updated training811.88 s4.004

Run this comparison · Related Qwen study

Detailed evidence

Every decision, timing and model setting · Run protocol and source hashes · Implementation and limitations

Process a queue together
PathCorrect / 100Time / 100Mean blocks runChanged answers
All 4 layers832.09 s4.000
Groups of four834.21 s4.000

Run this comparison · Related Qwen study

Detailed evidence

Every decision, timing and model setting · Run protocol and source hashes · Implementation and limitations

Float32 versus INT8
PathCorrect / 100Time / 100Mean blocks runChanged answers
All 4 layers832.02 s4.000
INT8 weights831.47 s4.000

The browser INT8 run matched these float32 decisions. Python INT8 differed on one message; this is not proof of lossless conversion across runtimes.

Run this comparison · Related Qwen study

Detailed evidence

Every decision, timing and model setting · Run protocol and source hashes · Implementation and limitations

Standalone tiny classifier: separate export check

The two-layer classifier in the small-model demo got 82/100 correct, missed 13 toxic messages and falsely blocked 5 benign messages. The browser preserved all Python decisions. This is a separate single-pass functional check, not one of the paired comparisons above.

Both layers run. This later specialist uses a 0.4 BLOCK threshold; it is not the earlier 0.3-threshold model in the original Qwen cascade study.

Run this tiny classifier

Detailed evidence

Model, training and export checks · Every browser decision and timing