arm stringclasses 2
values | tokenizer_revision stringclasses 2
values | method stringclasses 7
values | K_subset_25q float64 0.55 0.7 | K_star_subset_25q float64 0 0.04 | n_questions int64 24 24 |
|---|---|---|---|---|---|
new_tokenizer | current (11c9b309, Aug 27 2024) | log_p_a_q | 0.546204 | 0 | 24 |
new_tokenizer | current (11c9b309, Aug 27 2024) | log_p_norm_a_q | 0.551 | 0 | 24 |
new_tokenizer | current (11c9b309, Aug 27 2024) | p_true | 0.688781 | 0.041667 | 24 |
new_tokenizer | current (11c9b309, Aug 27 2024) | log_p_a_q_eot | 0.56592 | 0 | 24 |
new_tokenizer | current (11c9b309, Aug 27 2024) | log_p_norm_a_q_eot | 0.562155 | 0 | 24 |
new_tokenizer | current (11c9b309, Aug 27 2024) | verif_v1_truefalse_score | 0.697365 | 0.041667 | 24 |
new_tokenizer | current (11c9b309, Aug 27 2024) | verif_v2_yesno_score | 0.692193 | 0.041667 | 24 |
old_tokenizer | pre-PR#36 (1937c70277fc, Jul 10 2024) | log_p_a_q | 0.546204 | 0 | 24 |
old_tokenizer | pre-PR#36 (1937c70277fc, Jul 10 2024) | log_p_norm_a_q | 0.551 | 0 | 24 |
old_tokenizer | pre-PR#36 (1937c70277fc, Jul 10 2024) | p_true | 0.688781 | 0.041667 | 24 |
old_tokenizer | pre-PR#36 (1937c70277fc, Jul 10 2024) | log_p_a_q_eot | 0.56592 | 0 | 24 |
old_tokenizer | pre-PR#36 (1937c70277fc, Jul 10 2024) | log_p_norm_a_q_eot | 0.562155 | 0 | 24 |
old_tokenizer | pre-PR#36 (1937c70277fc, Jul 10 2024) | verif_v1_truefalse_score | 0.697365 | 0.041667 | 24 |
old_tokenizer | pre-PR#36 (1937c70277fc, Jul 10 2024) | verif_v2_yesno_score | 0.692193 | 0.041667 | 24 |
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Inside-Out Replication V2 — Gemma OLD-tokenizer canary
Negative-result canary on google/gemma-2-9b-it / P26 / test (25-qid
subset), testing whether the HF tokenizer-files update (commit 4efc01a1,
"Update tokenizer files (#36)", 2024-08-07) is a candidate cause of our
Gemma P(True) anomaly (~0.64 vs paper-implied ~0.47).
Two arms — same model weights, same scoring code, same data, same attn
implementation (sdpa) and query_pre_attn_scalar=256. Only the tokenizer
revision differs:
new_tokenizer:11c9b309(current latest, Aug 27 2024)old_tokenizer:1937c70277fc(Jul 10 2024, immediately before PR #36)
Headline result
| metric | new (Aug) | old (Jul) | Δ |
|---|---|---|---|
| log_p_a_q | 0.5462 | 0.5462 | 0.0000 |
| log_p_norm_a_q | 0.5510 | 0.5510 | 0.0000 |
| p_true | 0.6888 | 0.6888 | 0.0000 |
| log_p_a_q_eot | 0.5659 | 0.5659 | 0.0000 |
| log_p_norm_a_q_eot | 0.5622 | 0.5622 | 0.0000 |
| verif_v1_truefalse | 0.6974 | 0.6974 | 0.0000 |
| verif_v2_yesno | 0.6922 | 0.6922 | 0.0000 |
Δ = 0.0000 on every metric. The tokenizer update didn't change tokenization
of the entity answers we score on P26 (likely a tokenizer.json fast-tokenizer
addition or special-token metadata change that doesn't affect typical entity
strings).
Tokenizer revision is ruled out as a candidate cause of the Gemma anomaly.
- Downloads last month
- 8