suchirsalhan commited on
Commit
6e129aa
·
verified ·
1 Parent(s): 096fe32

refresh 2026-08-26T19:45:01Z (48 files)

Browse files
README.md CHANGED
@@ -52,8 +52,8 @@ configs:
52
  all 20 matched checkpoints, the endpoint-scaled benchmark mean rises +0.219 (18/20 positive,
53
  Wilcoxon p = 9.5e-6) (§6 T0).
54
 
55
- **Coverage: mechanism 20/20 matched pairs, property panel 0/20.** Regenerated
56
- 2026-08-26 19:32 UTC.
57
  <!-- AIM-HEADLINE:END -->
58
  Published at
59
  [`Mergeability-2/aim-activation-informed-merging`](https://huggingface.co/datasets/Mergeability-2/aim-activation-informed-merging),
@@ -97,7 +97,7 @@ Three things follow immediately, and they frame everything below.
97
  - **`model.embed_tokens` is exempt** (the `'embed' not in name` guard), so the merged embedding
98
  table survives untouched. `lm_head` is *not* exempt and is shrunk.
99
 
100
- ## 2. Does AIM do what it claims? (mechanism, complete)
101
 
102
  ### 2.1 The published checkpoints are exactly the closed form — recovered from public artefacts
103
 
@@ -108,14 +108,17 @@ elementwise ratios: the checkpoints are stored in bf16, so the deltas carry ~2^-
108
  quantisation noise that makes elementwise ratios meaningless while leaving the projection well
109
  determined.)
110
 
111
- On `TaskArithmetic / Code-Math`:
112
 
113
- | quantity | predicted by the stated rule | recovered |
114
- |---|---|---|
115
- | min over channels of `r_j` | ω = 0.4 exactly | **0.396** (median over tensors) |
116
- | max over channels of `r_j` | 1.0 exactly | **0.998** |
117
- | variance of `W_AIM - W_base` explained by a per-channel gain | 1.0 | **R² = 0.9989** |
118
- | `embed_tokens` Frobenius ratio | 1.0 (exempt) | **1.0000** |
 
 
 
119
 
120
  ### 2.2 The shrinkage is keyed to the *base model's* activations — confirmed independently
121
 
@@ -124,10 +127,10 @@ base model, on pile-val (135 blocks × 512 tokens), and compares the salience **
124
  published checkpoint pair**, `â_j = (1 - r_j)/(1 - ω)`, against the salience **measured from the
125
  base model**.
126
 
127
- Over 281 (tensor, module) pairs:
128
 
129
- - median Pearson **r = 0.9921** (5th percentile 0.952, minimum 0.891)
130
- - median absolute error **0.026** on a 0–1 scale, median fitted slope 1.04
131
 
132
  So AIM's claim — that it preserves the weights the base model's activations single out — is
133
  **true as stated, and verifiable without running the merge, the calibration, or a benchmark**.
@@ -136,9 +139,10 @@ So AIM's claim — that it preserves the weights the base model's activations si
136
 
137
  This is where the interesting part is.
138
 
139
- - **Total weight change removed: 7.3%.** Summed over every non-embedding tensor,
140
- `||ΔW_AIM||_F / ||ΔW_merge||_F = 0.927`. AIM keeps 93% of the merge's task vector.
141
- Cosine between the AIM delta and the baseline delta is **0.999** (median per tensor).
 
142
  - **The protection lands on almost nothing.** Because `a_j` is normalised by its *maximum* and
143
  LLaMA's residual stream has massive activation outliers, the salience vector is extremely
144
  peaked. Median fraction of input channels with `a_j > 0.5` (i.e. delta cut by more than 30%):
@@ -151,17 +155,21 @@ This is where the interesting part is.
151
  | `mlp.down_proj` | 0.272 | 0.0005 |
152
  | `lm_head` | 0.012 | 0.0008 |
153
 
154
- Across all modules, **0.2% of input channels** carry `a_j > 0.5`.
 
 
155
 
156
- - **A single scalar per tensor reproduces AIM to 99.85%.** Fitting one activation-*agnostic*
157
- scalar per tensor instead of the full per-channel profile explains R² = 0.99846 of
158
- `W_AIM - W_base`, against R² = 0.99885 for the full profile. The activation-informed targeting
159
- accounts for a quarter of an already-tiny residual.
 
160
 
161
- That last line is the reason section 3's control exists. If a magnitude-matched *uniform* shrink
162
- reproduces 99.85% of what AIM does to the weights, then the burden is on the activation-informed
163
  part to show it does something in *activation* space that the uniform control does not — which is
164
  exactly AIM's claim, and exactly what a with/without benchmark comparison cannot separate.
 
165
 
166
  ### 2.4 Where the protection lands: AIM is mostly an MLP-input intervention
167
 
@@ -193,9 +201,13 @@ blocks, and the top-8 sets of two randomly chosen blocks overlap 62% on average.
193
 
194
  ## 3. The activation-space test, and the property panel
195
 
196
- Running (`scripts/aim/run_aim.sh`, two shards on GPUs 6 and 7, one matched pair on disk at a
197
- time, resumable ledger at `results/aim/ledger_s*.json`). `scripts/aim/autopilot_aim.sh` appends
198
- the numbers to this file when both shards stop, so the run finishes unattended. What it computes:
 
 
 
 
199
 
200
  ### 3.1 The test AIM's claim actually needs — and its control
201
 
@@ -288,6 +300,25 @@ pullback cheap in activation space (that is the result above, and it is unanimou
288
  checkpoints what varies most with the benchmark gain is simply how much of the merge got pulled
289
  back. On 20 points none of these correlations is individually distinguishable from zero, so this
290
  is a direction to look, not a claim.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
291
 
292
  ## 4. What did not run, and why
293
 
@@ -428,31 +459,31 @@ Per benchmark, paired across the 20 matched pairs:
428
 
429
  ### T2 — Does AIM's effect differ by merge operator?
430
 
431
- | operator | n_metrics | n_moved_q<0.05 | mean_|diff| (SD units) | largest movers |
432
- |:-------------------|------------:|-----------------:|-------------------------:|:--------------------------------------------------------------------------------|
433
- | ALL | 5 | 5 | 0.3679 | cka_to_base (+0.58), act_deviation_from_base (-0.51), beh_kl_to_base (-0.28) |
434
- | DARETaskArithmetic | 5 | 0 | 0.446 | cka_to_base (+0.83), act_deviation_from_base (-0.60), beh_kl_to_base (-0.29) |
435
- | DARETies | 5 | 0 | 0.4776 | cka_to_base (+0.89), act_deviation_from_base (-0.64), beh_kl_to_base (-0.35) |
436
- | TaskArithmetic | 5 | 0 | 0.4246 | cka_to_base (+0.70), act_deviation_from_base (-0.57), beh_kl_to_base (-0.35) |
437
- | Ties | 5 | 0 | 0.2288 | act_deviation_from_base (-0.39), cka_to_base (+0.27), qmd_raw_from_base (-0.19) |
438
- | WIDEN | 5 | 0 | 0.2624 | act_deviation_from_base (-0.35), beh_js_to_base (-0.30), beh_kl_to_base (-0.28) |
439
 
440
 
441
- Full table: `results/aim/tables/T2_per_operator.csv` (30 rows: every metric x every level, with the paired mean difference in SD units, the count of positive differences, Wilcoxon/sign p and BH q).
442
 
443
 
444
  ### T3 — Does it differ by which domains were merged?
445
 
446
- | combo | n_metrics | n_moved_q<0.05 | mean_|diff| (SD units) | largest movers |
447
- |:----------------------------|------------:|-----------------:|-------------------------:|:--------------------------------------------------------------------------------|
448
- | ALL | 5 | 5 | 0.3679 | cka_to_base (+0.58), act_deviation_from_base (-0.51), beh_kl_to_base (-0.28) |
449
- | Code-Instruction_Tuned | 5 | 0 | 0.3761 | cka_to_base (+0.63), act_deviation_from_base (-0.53), qmd_raw_from_base (-0.26) |
450
- | Code-Math | 5 | 0 | 0.1684 | act_deviation_from_base (-0.28), beh_js_to_base (-0.17), beh_kl_to_base (-0.16) |
451
- | Code-Math-Instruction_Tuned | 5 | 0 | 0.4938 | cka_to_base (+0.93), act_deviation_from_base (-0.65), beh_kl_to_base (-0.34) |
452
- | Math-Instruction_Tuned | 5 | 0 | 0.4332 | cka_to_base (+0.63), act_deviation_from_base (-0.59), beh_kl_to_base (-0.38) |
453
 
454
 
455
- Full table: `results/aim/tables/T3_per_combo.csv` (25 rows: every metric x every level, with the paired mean difference in SD units, the count of positive differences, Wilcoxon/sign p and BH q).
456
 
457
 
458
  ### T1 — Paired per-checkpoint table
@@ -460,17 +491,59 @@ Full table: `results/aim/tables/T3_per_combo.csv` (25 rows: every metric x every
460
  One row per (operator, task combination, property): the with-AIM value, its baseline twin's value, the raw difference and the difference in SD units, joined to the published outcome for that pair. `results/aim/tables/T1_paired_per_checkpoint.csv`. This is the core artefact of the matched design; it is too large to render inline (20 pairs x ~70 properties).
461
 
462
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
463
  ### T5 — Coverage and provenance
464
 
465
  0/20 matched pairs complete. Published benchmark scores cover 20/20.
466
 
467
  | operator_label | combo | ledger_status | weight_mechanism | activation_mechanism | property_panel_both_arms | published_benchmarks |
468
  |:-----------------|:----------------------------|:----------------|:-------------------|:-----------------------|:---------------------------|:-----------------------|
469
- | Task Arithmetic | Code-Math | not-run | True | True | False | True |
470
  | Task Arithmetic | Code-Instruction_Tuned | not-run | True | True | False | True |
471
  | Task Arithmetic | Math-Instruction_Tuned | not-run | True | True | False | True |
472
  | Task Arithmetic | Code-Math-Instruction_Tuned | not-run | True | True | False | True |
473
- | TIES | Code-Math | not-run | True | True | False | True |
474
  | TIES | Code-Instruction_Tuned | not-run | True | True | False | True |
475
  | TIES | Math-Instruction_Tuned | not-run | True | True | False | True |
476
  | TIES | Code-Math-Instruction_Tuned | not-run | True | True | False | True |
@@ -492,29 +565,29 @@ F8 property columns, and the reason for any that are empty. NaN-by-construction
492
 
493
  | column | family | in_F8 | populated | reason_if_not |
494
  |:--------------------------|:---------------|:--------|:------------|:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
495
- | weight_cosine | weight space | True | False | sweep incomplete |
496
  | qmd_raw | weight space | True | False | sweep incomplete |
497
  | coord_fraction | weight space | True | False | NaN by construction: coordinate_gap / qmd_raw with a zero numerator. |
498
- | subspace_overlap | weight space | True | False | sweep incomplete |
499
- | spectral_overcounting | weight space | True | False | sweep incomplete |
500
- | geo_cka_mean | representation | True | False | sweep incomplete |
501
- | geo_procrustes_mean | representation | True | False | sweep incomplete |
502
- | geo_subspace_overlap_mean | representation | True | False | sweep incomplete |
503
- | geo_svcca_mean | representation | True | False | sweep incomplete |
504
- | geo_cka_late_minus_early | representation | True | False | sweep incomplete |
505
- | ret_identity_p_at_1 | retrieval | True | False | sweep incomplete |
506
- | ret_procrustes_p_at_1 | retrieval | True | False | sweep incomplete |
507
- | ret_ridge_p_at_1 | retrieval | True | False | sweep incomplete |
508
- | ret_gain_over_identity | retrieval | True | False | sweep incomplete |
509
- | grad_cosine | gradient | True | False | sweep incomplete |
510
- | grad_l2 | gradient | True | False | sweep incomplete |
511
- | grad_norm_ratio | gradient | True | False | sweep incomplete |
512
- | grad_cosine_layer_min | gradient | True | False | sweep incomplete |
513
- | beh_js | behaviour | True | False | sweep incomplete |
514
- | beh_logit_cosine | behaviour | True | False | sweep incomplete |
515
- | beh_topk_overlap | behaviour | True | False | sweep incomplete |
516
- | beh_rank_corr | behaviour | True | False | sweep incomplete |
517
- | beh_entropy_gap | behaviour | True | False | sweep incomplete |
518
  | geoX_* | representation | False | False | extended 512-sentence probe not run: it is the robustness check on the n/d problem, not the figure, and the 41-layer Procrustes/SVCCA geometry at 512x5120 dominated the sweep's cost on a contended box. |
519
  | retX_* | retrieval | False | False | extended 512-sentence probe not run: it is the robustness check on the n/d problem, not the figure, and the 41-layer Procrustes/SVCCA geometry at 512x5120 dominated the sweep's cost on a contended box. |
520
 
 
52
  all 20 matched checkpoints, the endpoint-scaled benchmark mean rises +0.219 (18/20 positive,
53
  Wilcoxon p = 9.5e-6) (§6 T0).
54
 
55
+ **Coverage: mechanism 20/20 matched pairs, property panel 2/20.** Regenerated
56
+ 2026-08-26 19:44 UTC.
57
  <!-- AIM-HEADLINE:END -->
58
  Published at
59
  [`Mergeability-2/aim-activation-informed-merging`](https://huggingface.co/datasets/Mergeability-2/aim-activation-informed-merging),
 
97
  - **`model.embed_tokens` is exempt** (the `'embed' not in name` guard), so the merged embedding
98
  table survives untouched. `lm_head` is *not* exempt and is shrunk.
99
 
100
+ ## 2. What AIM does to the weights (mechanism, 20/20 pairs, complete)
101
 
102
  ### 2.1 The published checkpoints are exactly the closed form — recovered from public artefacts
103
 
 
108
  quantisation noise that makes elementwise ratios meaningless while leaving the projection well
109
  determined.)
110
 
111
+ Across **all 20 matched pairs** (median over the 20, and over the ~360 tensors of each):
112
 
113
+ | quantity | predicted by the stated rule | recovered | range over the 20 pairs |
114
+ |---|---|---|---|
115
+ | min over channels of `r_j` | ω = 0.4 exactly | **0.3998** | 0.372 0.400 |
116
+ | max over channels of `r_j` | 1.0 exactly | **0.9953** | 0.993 – 1.002 |
117
+ | variance of `W_AIM W_base` explained by a per-input-channel gain | 1.0 | **R² = 0.9992** | 0.990 – 0.9995 |
118
+ | `model.embed_tokens` Frobenius ratio | 1.0 (exempt by the `'embed' not in name` guard) | **1.0000** | 1.0000 – 1.0019 |
119
+
120
+ Nothing about ω was assumed: 0.3998 is *measured* from the checkpoints and lands on the paper's
121
+ stated 0.4. The residual 0.08% is bf16 quantisation, not model error.
122
 
123
  ### 2.2 The shrinkage is keyed to the *base model's* activations — confirmed independently
124
 
 
127
  published checkpoint pair**, `â_j = (1 - r_j)/(1 - ω)`, against the salience **measured from the
128
  base model**.
129
 
130
+ Over **5620 (checkpoint, module) pairs** — all 20 matched pairs × ~280 modules each:
131
 
132
+ - median Pearson **r = 0.9956** (5th percentile 0.952, minimum 0.684)
133
+ - median absolute error **0.0088** on a 0–1 scale, median fitted slope 1.04
134
 
135
  So AIM's claim — that it preserves the weights the base model's activations single out — is
136
  **true as stated, and verifiable without running the merge, the calibration, or a benchmark**.
 
139
 
140
  This is where the interesting part is.
141
 
142
+ - **Total weight change removed: 8.0%.** Summed over every non-embedding tensor,
143
+ `ΔW_AIM_F / ΔW_merge_F = 0.920` (median over the 20 pairs; range 0.917–0.935). AIM keeps
144
+ 92% of the merge's task vector. Cosine between the AIM delta and the baseline delta is
145
+ **0.9994** (median per tensor).
146
  - **The protection lands on almost nothing.** Because `a_j` is normalised by its *maximum* and
147
  LLaMA's residual stream has massive activation outliers, the salience vector is extremely
148
  peaked. Median fraction of input channels with `a_j > 0.5` (i.e. delta cut by more than 30%):
 
155
  | `mlp.down_proj` | 0.272 | 0.0005 |
156
  | `lm_head` | 0.012 | 0.0008 |
157
 
158
+ Across all modules, **0.2% of input channels** carry `a_j > 0.5`, and only **0.84%** are
159
+ protected enough to lose more than a tenth of their delta (`r_j < 0.9`; range 0.66–0.86% over
160
+ the 20 pairs).
161
 
162
+ - **A single scalar per tensor reproduces AIM's weights to 99.88%.** Fitting one
163
+ activation-*agnostic* scalar per tensor instead of the full per-channel profile explains
164
+ R² = 0.99882 of `W_AIM W_base`, against R² = 0.99918 for the full profile. In weight space the
165
+ activation-informed targeting accounts for barely a third of an already-tiny residual — which is
166
+ exactly why §3.2's control is the test that matters, and why its result is a surprise.
167
 
168
+ That last line is the reason §3's control exists. If a magnitude-matched *uniform* shrink
169
+ reproduces 99.88% of what AIM does to the weights, then the burden is on the activation-informed
170
  part to show it does something in *activation* space that the uniform control does not — which is
171
  exactly AIM's claim, and exactly what a with/without benchmark comparison cannot separate.
172
+ **It does: see §3.2.**
173
 
174
  ### 2.4 Where the protection lands: AIM is mostly an MLP-input intervention
175
 
 
201
 
202
  ## 3. The activation-space test, and the property panel
203
 
204
+ Two phases, chained per shard (`scripts/aim/chain_aim.sh`, GPUs 6 and 7, one matched pair on disk
205
+ at a time, resumable ledger at `results/aim/ledger_s*.json`). Phase `mech` — the test of AIM's own
206
+ claim, §3.1–3.2 ran first and is **complete, 20/20**; it costs no gradients and no property
207
+ columns and took ~85 s a pair. Phase `panel` — the F8 property suite, §3.3 — then covers as many
208
+ pairs as fit before its wall-clock deadline. `scripts/aim/autopilot_aim.sh` rebuilds every table
209
+ and figure and pushes to the Hub every 15 minutes, so partial coverage is always visible and
210
+ always labelled.
211
 
212
  ### 3.1 The test AIM's claim actually needs — and its control
213
 
 
300
  checkpoints what varies most with the benchmark gain is simply how much of the merge got pulled
301
  back. On 20 points none of these correlations is individually distinguishable from zero, so this
302
  is a direction to look, not a claim.
303
+ ### 3.3 The property panel
304
+
305
+ The F8 suite (`figures/final/F8_metric_families.png`'s 23 properties in five families, same
306
+ estimators, same 32-sentence probe) computed on **(merged model, base model)** rather than on a
307
+ parent pair. The parent set is constant within a task combination, so a pre-merge property cannot
308
+ vary across the 20 cells; the object that does vary is the merge's own relation to the base it was
309
+ built from.
310
+
311
+ Two caveats stated up front, because they decide how the figure may be read.
312
+
313
+ - **Four points per operator.** Each operator has four task combinations, so a per-operator
314
+ correlation has n = 4 and its exact permutation p cannot go below 1/12. Every per-operator cell
315
+ in `figures/aim/F8_AIM_metric_families.png` is struck through; that is the honest result, not a
316
+ bug. The bottom row pools all pairs after within-operator centring and uses an exact
317
+ combo-clustered permutation.
318
+ - **`F8b_AIM_property_deltas` is the panel with power**: the paired with-minus-without difference
319
+ in each property over the matched pairs, Wilcoxon signed rank on the pooled row. Note that a
320
+ shift toward the base model is expected there *by construction* (§3.1); what the panel adds is
321
+ **which families move more than the weight-space shrink alone would predict**.
322
 
323
  ## 4. What did not run, and why
324
 
 
459
 
460
  ### T2 — Does AIM's effect differ by merge operator?
461
 
462
+ | operator | n_metrics | n_moved_q<0.05 | mean_|diff| (SD units) | largest movers |
463
+ |:-------------------|------------:|-----------------:|-------------------------:|:---------------------------------------------------------------------------------------------------------|
464
+ | ALL | 49 | 5 | 0.8104 | ret_procrustes_p_at_5 (-1.73), grad_cosine_layer_min (-1.64), grad_cosine_layer_late_minus_early (+1.53) |
465
+ | DARETaskArithmetic | 5 | 0 | 0.446 | cka_to_base (+0.83), act_deviation_from_base (-0.60), beh_kl_to_base (-0.29) |
466
+ | DARETies | 5 | 0 | 0.4776 | cka_to_base (+0.89), act_deviation_from_base (-0.64), beh_kl_to_base (-0.35) |
467
+ | TaskArithmetic | 49 | 0 | 1.021 | subspace_overlap (+2.45), grad_cosine_layer_min (-2.13), spectral_overcounting (-2.00) |
468
+ | Ties | 49 | 0 | 0.6138 | ret_procrustes_p_at_5 (-1.73), grad_norm_a (+1.27), grad_norm_ratio (+1.27) |
469
+ | WIDEN | 5 | 0 | 0.2624 | act_deviation_from_base (-0.35), beh_js_to_base (-0.30), beh_kl_to_base (-0.28) |
470
 
471
 
472
+ Full table: `results/aim/tables/T2_per_operator.csv` (162 rows: every metric x every level, with the paired mean difference in SD units, the count of positive differences, Wilcoxon/sign p and BH q).
473
 
474
 
475
  ### T3 — Does it differ by which domains were merged?
476
 
477
+ | combo | n_metrics | n_moved_q<0.05 | mean_|diff| (SD units) | largest movers |
478
+ |:----------------------------|------------:|-----------------:|-------------------------:|:---------------------------------------------------------------------------------------------------------|
479
+ | ALL | 49 | 5 | 0.8104 | ret_procrustes_p_at_5 (-1.73), grad_cosine_layer_min (-1.64), grad_cosine_layer_late_minus_early (+1.53) |
480
+ | Code-Instruction_Tuned | 5 | 0 | 0.3761 | cka_to_base (+0.63), act_deviation_from_base (-0.53), qmd_raw_from_base (-0.26) |
481
+ | Code-Math | 49 | 0 | 0.7802 | ret_procrustes_p_at_5 (-1.73), grad_cosine_layer_min (-1.64), grad_cosine_layer_late_minus_early (+1.53) |
482
+ | Code-Math-Instruction_Tuned | 5 | 0 | 0.4938 | cka_to_base (+0.93), act_deviation_from_base (-0.65), beh_kl_to_base (-0.34) |
483
+ | Math-Instruction_Tuned | 5 | 0 | 0.4332 | cka_to_base (+0.63), act_deviation_from_base (-0.59), beh_kl_to_base (-0.38) |
484
 
485
 
486
+ Full table: `results/aim/tables/T3_per_combo.csv` (113 rows: every metric x every level, with the paired mean difference in SD units, the count of positive differences, Wilcoxon/sign p and BH q).
487
 
488
 
489
  ### T1 — Paired per-checkpoint table
 
491
  One row per (operator, task combination, property): the with-AIM value, its baseline twin's value, the raw difference and the difference in SD units, joined to the published outcome for that pair. `results/aim/tables/T1_paired_per_checkpoint.csv`. This is the core artefact of the matched design; it is too large to render inline (20 pairs x ~70 properties).
492
 
493
 
494
+ ### F8 / F8b — the property panels, pooled row
495
+
496
+ `figures/aim/F8_AIM_metric_families.png` panels (a)/(b)/(c) and `figures/aim/F8b_AIM_property_deltas.png`, bottom row (`all (within-op.)`, n = 1 pairs). Correlations are against the endpoint-scaled published benchmark mean.
497
+
498
+ | property | family | r with AIM | r baseline | delta r | paired delta property (SD) | q (BH) |
499
+ |:--------------------------|:---------------|:-------------|:-------------|:----------|:-----------------------------|:---------|
500
+ | weight_cosine | weight space | — | — | — | — | — |
501
+ | qmd_raw | weight space | — | — | — | — | — |
502
+ | coord_fraction | weight space | — | — | — | — | — |
503
+ | subspace_overlap | weight space | — | — | — | — | — |
504
+ | spectral_overcounting | weight space | — | — | — | — | — |
505
+ | geo_cka_mean | representation | — | — | — | — | — |
506
+ | geo_procrustes_mean | representation | — | — | — | — | — |
507
+ | geo_subspace_overlap_mean | representation | — | — | — | — | — |
508
+ | geo_svcca_mean | representation | — | — | — | — | — |
509
+ | geo_cka_late_minus_early | representation | — | — | — | — | — |
510
+ | ret_identity_p_at_1 | retrieval | — | — | — | — | — |
511
+ | ret_procrustes_p_at_1 | retrieval | — | — | — | — | — |
512
+ | ret_ridge_p_at_1 | retrieval | — | — | — | — | — |
513
+ | ret_gain_over_identity | retrieval | — | — | — | — | — |
514
+ | grad_cosine | gradient | — | — | — | — | — |
515
+ | grad_l2 | gradient | — | — | — | — | — |
516
+ | grad_norm_ratio | gradient | — | — | — | — | — |
517
+ | grad_cosine_layer_min | gradient | — | — | — | — | — |
518
+ | beh_js | behaviour | — | — | — | — | — |
519
+ | beh_logit_cosine | behaviour | — | — | — | — | — |
520
+ | beh_topk_overlap | behaviour | — | — | — | — | — |
521
+ | beh_rank_corr | behaviour | — | — | — | — | — |
522
+ | beh_entropy_gap | behaviour | — | — | — | — | — |
523
+
524
+
525
+ **Does AIM change which properties track the outcome?** The cross-arm agreement of the property ranking is compared against a within-arm split-half ceiling (the six published benchmarks split into two disjoint halves), because with four task combinations per operator an agreement near zero could equally mean 'the ranking changed' or 'four points cannot pin down a correlation'.
526
+
527
+ | quantity | Spearman over properties |
528
+ |:-------------------------------|:---------------------------|
529
+ | within_arm_split_half_baseline | — |
530
+ | within_arm_split_half_aim | — |
531
+ | cross_arm_full_outcome | — |
532
+ | cross_arm_h1 | — |
533
+ | cross_arm_h2 | — |
534
+
535
+
536
  ### T5 — Coverage and provenance
537
 
538
  0/20 matched pairs complete. Published benchmark scores cover 20/20.
539
 
540
  | operator_label | combo | ledger_status | weight_mechanism | activation_mechanism | property_panel_both_arms | published_benchmarks |
541
  |:-----------------|:----------------------------|:----------------|:-------------------|:-----------------------|:---------------------------|:-----------------------|
542
+ | Task Arithmetic | Code-Math | not-run | True | True | True | True |
543
  | Task Arithmetic | Code-Instruction_Tuned | not-run | True | True | False | True |
544
  | Task Arithmetic | Math-Instruction_Tuned | not-run | True | True | False | True |
545
  | Task Arithmetic | Code-Math-Instruction_Tuned | not-run | True | True | False | True |
546
+ | TIES | Code-Math | not-run | True | True | True | True |
547
  | TIES | Code-Instruction_Tuned | not-run | True | True | False | True |
548
  | TIES | Math-Instruction_Tuned | not-run | True | True | False | True |
549
  | TIES | Code-Math-Instruction_Tuned | not-run | True | True | False | True |
 
565
 
566
  | column | family | in_F8 | populated | reason_if_not |
567
  |:--------------------------|:---------------|:--------|:------------|:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
568
+ | weight_cosine | weight space | True | True | nan |
569
  | qmd_raw | weight space | True | False | sweep incomplete |
570
  | coord_fraction | weight space | True | False | NaN by construction: coordinate_gap / qmd_raw with a zero numerator. |
571
+ | subspace_overlap | weight space | True | True | nan |
572
+ | spectral_overcounting | weight space | True | True | nan |
573
+ | geo_cka_mean | representation | True | True | nan |
574
+ | geo_procrustes_mean | representation | True | True | nan |
575
+ | geo_subspace_overlap_mean | representation | True | True | nan |
576
+ | geo_svcca_mean | representation | True | True | nan |
577
+ | geo_cka_late_minus_early | representation | True | True | nan |
578
+ | ret_identity_p_at_1 | retrieval | True | True | nan |
579
+ | ret_procrustes_p_at_1 | retrieval | True | True | nan |
580
+ | ret_ridge_p_at_1 | retrieval | True | True | nan |
581
+ | ret_gain_over_identity | retrieval | True | True | nan |
582
+ | grad_cosine | gradient | True | True | nan |
583
+ | grad_l2 | gradient | True | True | nan |
584
+ | grad_norm_ratio | gradient | True | True | nan |
585
+ | grad_cosine_layer_min | gradient | True | True | nan |
586
+ | beh_js | behaviour | True | True | nan |
587
+ | beh_logit_cosine | behaviour | True | True | nan |
588
+ | beh_topk_overlap | behaviour | True | True | nan |
589
+ | beh_rank_corr | behaviour | True | True | nan |
590
+ | beh_entropy_gap | behaviour | True | True | nan |
591
  | geoX_* | representation | False | False | extended 512-sentence probe not run: it is the robustness check on the n/d problem, not the figure, and the 41-layer Procrustes/SVCCA geometry at 512x5120 dominated the sweep's cost on a contended box. |
592
  | retX_* | retrieval | False | False | extended 512-sentence probe not run: it is the robustness check on the n/d problem, not the figure, and the 41-layer Procrustes/SVCCA geometry at 512x5120 dominated the sweep's cost on a contended box. |
593
 
figures/F8_AIM_metric_families.pdf ADDED
Binary file (39.4 kB). View file
 
figures/F8_AIM_metric_families.png ADDED

Git LFS Details

  • SHA256: 12bf92353faa0dbc4ea1fbead1259e77d7b9b06101e40e569daefb270f24ae09
  • Pointer size: 131 Bytes
  • Size of remote file: 445 kB
figures/F8b_AIM_property_deltas.pdf ADDED
Binary file (36.6 kB). View file
 
figures/F8b_AIM_property_deltas.png ADDED

Git LFS Details

  • SHA256: 8a55778fed077808a271395ac006762db9d285ab6d9adae06fc8b24b79ef5d73
  • Pointer size: 131 Bytes
  • Size of remote file: 349 kB
figures/F9_AIM_mechanism.pdf CHANGED
Binary files a/figures/F9_AIM_mechanism.pdf and b/figures/F9_AIM_mechanism.pdf differ
 
results/ledger_s0.json CHANGED
@@ -28,6 +28,12 @@
28
  "phase": "mech",
29
  "status": "ok"
30
  },
 
 
 
 
 
 
31
  "TaskArithmetic__Math-Instruction_Tuned#mech": {
32
  "phase": "mech",
33
  "secs": 197,
 
28
  "phase": "mech",
29
  "status": "ok"
30
  },
31
+ "TaskArithmetic__Code-Math#panel": {
32
+ "phase": "panel",
33
+ "secs": 743,
34
+ "status": "ok",
35
+ "when": "2026-08-26T19:42:29"
36
+ },
37
  "TaskArithmetic__Math-Instruction_Tuned#mech": {
38
  "phase": "mech",
39
  "secs": 197,
results/ledger_s1.json CHANGED
@@ -39,6 +39,12 @@
39
  "phase": "mech",
40
  "status": "ok"
41
  },
 
 
 
 
 
 
42
  "Ties__Math-Instruction_Tuned#mech": {
43
  "phase": "mech",
44
  "secs": 196,
 
39
  "phase": "mech",
40
  "status": "ok"
41
  },
42
+ "Ties__Code-Math#panel": {
43
+ "phase": "panel",
44
+ "secs": 738,
45
+ "status": "ok",
46
+ "when": "2026-08-26T19:39:41"
47
+ },
48
  "Ties__Math-Instruction_Tuned#mech": {
49
  "phase": "mech",
50
  "secs": 196,
results/panel_rows_s0.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ operator,combo,aim,arm,repo,geo_cka_mean,geo_cka_late_minus_early,geo_cka_peak_depth,geo_rowwise_cosine_mean,geo_rowwise_cosine_late_minus_early,geo_rowwise_cosine_peak_depth,geo_procrustes_mean,geo_procrustes_late_minus_early,geo_procrustes_peak_depth,geo_subspace_overlap_mean,geo_subspace_overlap_late_minus_early,geo_subspace_overlap_peak_depth,geo_svcca_mean,geo_svcca_late_minus_early,geo_svcca_peak_depth,ret_identity_p_at_1,ret_identity_p_at_5,ret_identity_p_at_10,ret_procrustes_p_at_1,ret_procrustes_p_at_5,ret_procrustes_p_at_10,ret_ridge_p_at_1,ret_ridge_p_at_5,ret_ridge_p_at_10,ret_gain_over_identity,beh_logit_kl,beh_kl_sym,beh_js,beh_logit_cosine,beh_logit_l2,beh_entropy_gap,beh_topk_overlap,beh_rank_corr,grad_cosine,grad_l2,grad_norm_ratio,grad_norm_a,grad_norm_b,grad_cosine_layer_mean,grad_cosine_layer_late_minus_early,grad_cosine_layer_min,weight_cosine,subspace_overlap,spectral_overcounting
2
+ TaskArithmetic,Code-Math,0,merge,ahn1376/TaskArithmetic___Code-Math,0.9930885569927597,-0.07698849211054926,0.1794871794871795,0.9908710673451424,-0.08166122436523438,0.07692307692307693,0.002721947189754578,0.01404953890491245,1.0,0.7774382523120954,-0.20737463432520575,0.0,0.9891374250741723,0.00666258984990542,0.5641025641025641,1.0,1.0,1.0,0.15384615384615385,0.6153846153846154,0.8461538461538461,0.15384615384615385,0.5384615384615384,0.8461538461538461,-0.8461538461538461,0.30399625235676625,0.3235272971571356,0.07076306797142021,0.9613301176577806,162.13931274414062,-0.9580553334901714,0.75,0.9612488259798522,0.592158773428496,2306.759099282534,0.9110638542598302,2425.0298323354004,2661.756166702006,0.5155714649707079,0.2209797352552414,0.2175084501504898,0.999807596206665,0.99853515625,-0.9999980330467224
3
+ TaskArithmetic,Code-Math,1,aim,ahn1376/TaskArithmetic___Code-Math___AIM,0.9945506464794935,-0.055673375970133,0.1794871794871795,0.9926170140504837,-0.06605297327041626,0.07692307692307693,0.002190665566547556,0.010471952043090593,1.0,0.7996645381073592,-0.1796519648387671,0.0,0.990373600650402,0.005690394472177118,0.5641025641025641,1.0,1.0,1.0,0.15384615384615385,0.5384615384615384,0.8461538461538461,0.15384615384615385,0.5384615384615384,0.8461538461538461,-0.8461538461538461,0.2630344502284437,0.27374887236882367,0.06055055836978797,0.9688364844769239,145.14276123046875,-0.8577658281539922,0.759375,0.9668371863411456,0.5782269179049978,2370.2585704517037,0.93503639362456,2488.8388868209768,2661.756166702006,0.5337781719863415,0.3219577968120575,0.13126400113105774,0.999807596206665,0.999355137348175,-0.9999980926513672
results/panel_rows_s1.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ operator,combo,aim,arm,repo,geo_cka_mean,geo_cka_late_minus_early,geo_cka_peak_depth,geo_rowwise_cosine_mean,geo_rowwise_cosine_late_minus_early,geo_rowwise_cosine_peak_depth,geo_procrustes_mean,geo_procrustes_late_minus_early,geo_procrustes_peak_depth,geo_subspace_overlap_mean,geo_subspace_overlap_late_minus_early,geo_subspace_overlap_peak_depth,geo_svcca_mean,geo_svcca_late_minus_early,geo_svcca_peak_depth,ret_identity_p_at_1,ret_identity_p_at_5,ret_identity_p_at_10,ret_procrustes_p_at_1,ret_procrustes_p_at_5,ret_procrustes_p_at_10,ret_ridge_p_at_1,ret_ridge_p_at_5,ret_ridge_p_at_10,ret_gain_over_identity,beh_logit_kl,beh_kl_sym,beh_js,beh_logit_cosine,beh_logit_l2,beh_entropy_gap,beh_topk_overlap,beh_rank_corr,grad_cosine,grad_l2,grad_norm_ratio,grad_norm_a,grad_norm_b,grad_cosine_layer_mean,grad_cosine_layer_late_minus_early,grad_cosine_layer_min,weight_cosine,subspace_overlap,spectral_overcounting
2
+ Ties,Code-Math,0,merge,ahn1376/Ties___Code-Math,0.9950276483976795,-0.03982648848181902,0.1794871794871795,0.9932829976081848,-0.05726140737533569,0.07692307692307693,0.0019867146024871996,0.007681346688587776,1.0,0.8307646546247305,-0.15156557415598138,0.0,0.9918700300931974,0.005308828495821705,0.5641025641025641,1.0,1.0,1.0,0.15384615384615385,0.6153846153846154,0.8461538461538461,0.15384615384615385,0.5384615384615384,0.8461538461538461,-0.8461538461538461,0.21040047600780368,0.21566948027724875,0.04867244805998143,0.9735978525131941,134.2702178955078,-0.7119038706301577,0.790625,0.9714328588858698,0.5770794629628759,2248.9716232321766,0.7936686956273205,2112.5525449043575,2661.756166702006,0.5680897474288941,0.27943167090415955,0.18610823154449463,0.999807596206665,0.9989471435546875,-0.9999980926513672
3
+ Ties,Code-Math,1,aim,ahn1376/Ties___Code-Math___AIM,0.9959680857592679,-0.031412654834733655,0.1794871794871795,0.9945648342370987,-0.047279417514801025,0.07692307692307693,0.0016178933505263781,0.006126797202974266,1.0,0.8532723782550617,-0.12419270942044114,0.0,0.9923686480991026,0.005257749399884104,0.5641025641025641,1.0,1.0,1.0,0.15384615384615385,0.5384615384615384,0.8461538461538461,0.15384615384615385,0.5384615384615384,0.8461538461538461,-0.8461538461538461,0.18102952042879136,0.1838637848608179,0.04193555079188003,0.9786411840468645,120.259521484375,-0.6537038415514891,0.796875,0.9756381849494506,0.5798330660603503,2303.8239813727732,0.8719713407752943,2320.9750934960557,2661.756166702006,0.5935959197580815,0.3435364365577698,0.13996994495391846,0.999807596206665,0.998933732509613,-0.9999980926513672
results/panel_stats.json ADDED
@@ -0,0 +1,1270 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "n_checkpoints": 2,
3
+ "ranking_agreement": {},
4
+ "ranking_ceiling": {
5
+ "within_arm_split_half_baseline": NaN,
6
+ "within_arm_split_half_aim": NaN,
7
+ "cross_arm_full_outcome": NaN,
8
+ "cross_arm_h1": NaN,
9
+ "cross_arm_h2": NaN
10
+ },
11
+ "operators": [
12
+ "TaskArithmetic",
13
+ "Ties",
14
+ "DARETaskArithmetic",
15
+ "DARETies",
16
+ "WIDEN"
17
+ ],
18
+ "combos": [
19
+ "Code-Math"
20
+ ],
21
+ "r_aim": [
22
+ [
23
+ NaN,
24
+ NaN,
25
+ NaN,
26
+ NaN,
27
+ NaN,
28
+ NaN,
29
+ NaN,
30
+ NaN,
31
+ NaN,
32
+ NaN,
33
+ NaN,
34
+ NaN,
35
+ NaN,
36
+ NaN,
37
+ NaN,
38
+ NaN,
39
+ NaN,
40
+ NaN,
41
+ NaN,
42
+ NaN,
43
+ NaN,
44
+ NaN,
45
+ NaN
46
+ ],
47
+ [
48
+ NaN,
49
+ NaN,
50
+ NaN,
51
+ NaN,
52
+ NaN,
53
+ NaN,
54
+ NaN,
55
+ NaN,
56
+ NaN,
57
+ NaN,
58
+ NaN,
59
+ NaN,
60
+ NaN,
61
+ NaN,
62
+ NaN,
63
+ NaN,
64
+ NaN,
65
+ NaN,
66
+ NaN,
67
+ NaN,
68
+ NaN,
69
+ NaN,
70
+ NaN
71
+ ],
72
+ [
73
+ NaN,
74
+ NaN,
75
+ NaN,
76
+ NaN,
77
+ NaN,
78
+ NaN,
79
+ NaN,
80
+ NaN,
81
+ NaN,
82
+ NaN,
83
+ NaN,
84
+ NaN,
85
+ NaN,
86
+ NaN,
87
+ NaN,
88
+ NaN,
89
+ NaN,
90
+ NaN,
91
+ NaN,
92
+ NaN,
93
+ NaN,
94
+ NaN,
95
+ NaN
96
+ ],
97
+ [
98
+ NaN,
99
+ NaN,
100
+ NaN,
101
+ NaN,
102
+ NaN,
103
+ NaN,
104
+ NaN,
105
+ NaN,
106
+ NaN,
107
+ NaN,
108
+ NaN,
109
+ NaN,
110
+ NaN,
111
+ NaN,
112
+ NaN,
113
+ NaN,
114
+ NaN,
115
+ NaN,
116
+ NaN,
117
+ NaN,
118
+ NaN,
119
+ NaN,
120
+ NaN
121
+ ],
122
+ [
123
+ NaN,
124
+ NaN,
125
+ NaN,
126
+ NaN,
127
+ NaN,
128
+ NaN,
129
+ NaN,
130
+ NaN,
131
+ NaN,
132
+ NaN,
133
+ NaN,
134
+ NaN,
135
+ NaN,
136
+ NaN,
137
+ NaN,
138
+ NaN,
139
+ NaN,
140
+ NaN,
141
+ NaN,
142
+ NaN,
143
+ NaN,
144
+ NaN,
145
+ NaN
146
+ ],
147
+ [
148
+ NaN,
149
+ NaN,
150
+ NaN,
151
+ NaN,
152
+ NaN,
153
+ NaN,
154
+ NaN,
155
+ NaN,
156
+ NaN,
157
+ NaN,
158
+ NaN,
159
+ NaN,
160
+ NaN,
161
+ NaN,
162
+ NaN,
163
+ NaN,
164
+ NaN,
165
+ NaN,
166
+ NaN,
167
+ NaN,
168
+ NaN,
169
+ NaN,
170
+ NaN
171
+ ]
172
+ ],
173
+ "r_base": [
174
+ [
175
+ NaN,
176
+ NaN,
177
+ NaN,
178
+ NaN,
179
+ NaN,
180
+ NaN,
181
+ NaN,
182
+ NaN,
183
+ NaN,
184
+ NaN,
185
+ NaN,
186
+ NaN,
187
+ NaN,
188
+ NaN,
189
+ NaN,
190
+ NaN,
191
+ NaN,
192
+ NaN,
193
+ NaN,
194
+ NaN,
195
+ NaN,
196
+ NaN,
197
+ NaN
198
+ ],
199
+ [
200
+ NaN,
201
+ NaN,
202
+ NaN,
203
+ NaN,
204
+ NaN,
205
+ NaN,
206
+ NaN,
207
+ NaN,
208
+ NaN,
209
+ NaN,
210
+ NaN,
211
+ NaN,
212
+ NaN,
213
+ NaN,
214
+ NaN,
215
+ NaN,
216
+ NaN,
217
+ NaN,
218
+ NaN,
219
+ NaN,
220
+ NaN,
221
+ NaN,
222
+ NaN
223
+ ],
224
+ [
225
+ NaN,
226
+ NaN,
227
+ NaN,
228
+ NaN,
229
+ NaN,
230
+ NaN,
231
+ NaN,
232
+ NaN,
233
+ NaN,
234
+ NaN,
235
+ NaN,
236
+ NaN,
237
+ NaN,
238
+ NaN,
239
+ NaN,
240
+ NaN,
241
+ NaN,
242
+ NaN,
243
+ NaN,
244
+ NaN,
245
+ NaN,
246
+ NaN,
247
+ NaN
248
+ ],
249
+ [
250
+ NaN,
251
+ NaN,
252
+ NaN,
253
+ NaN,
254
+ NaN,
255
+ NaN,
256
+ NaN,
257
+ NaN,
258
+ NaN,
259
+ NaN,
260
+ NaN,
261
+ NaN,
262
+ NaN,
263
+ NaN,
264
+ NaN,
265
+ NaN,
266
+ NaN,
267
+ NaN,
268
+ NaN,
269
+ NaN,
270
+ NaN,
271
+ NaN,
272
+ NaN
273
+ ],
274
+ [
275
+ NaN,
276
+ NaN,
277
+ NaN,
278
+ NaN,
279
+ NaN,
280
+ NaN,
281
+ NaN,
282
+ NaN,
283
+ NaN,
284
+ NaN,
285
+ NaN,
286
+ NaN,
287
+ NaN,
288
+ NaN,
289
+ NaN,
290
+ NaN,
291
+ NaN,
292
+ NaN,
293
+ NaN,
294
+ NaN,
295
+ NaN,
296
+ NaN,
297
+ NaN
298
+ ],
299
+ [
300
+ NaN,
301
+ NaN,
302
+ NaN,
303
+ NaN,
304
+ NaN,
305
+ NaN,
306
+ NaN,
307
+ NaN,
308
+ NaN,
309
+ NaN,
310
+ NaN,
311
+ NaN,
312
+ NaN,
313
+ NaN,
314
+ NaN,
315
+ NaN,
316
+ NaN,
317
+ NaN,
318
+ NaN,
319
+ NaN,
320
+ NaN,
321
+ NaN,
322
+ NaN
323
+ ]
324
+ ],
325
+ "delta_r": [
326
+ [
327
+ NaN,
328
+ NaN,
329
+ NaN,
330
+ NaN,
331
+ NaN,
332
+ NaN,
333
+ NaN,
334
+ NaN,
335
+ NaN,
336
+ NaN,
337
+ NaN,
338
+ NaN,
339
+ NaN,
340
+ NaN,
341
+ NaN,
342
+ NaN,
343
+ NaN,
344
+ NaN,
345
+ NaN,
346
+ NaN,
347
+ NaN,
348
+ NaN,
349
+ NaN
350
+ ],
351
+ [
352
+ NaN,
353
+ NaN,
354
+ NaN,
355
+ NaN,
356
+ NaN,
357
+ NaN,
358
+ NaN,
359
+ NaN,
360
+ NaN,
361
+ NaN,
362
+ NaN,
363
+ NaN,
364
+ NaN,
365
+ NaN,
366
+ NaN,
367
+ NaN,
368
+ NaN,
369
+ NaN,
370
+ NaN,
371
+ NaN,
372
+ NaN,
373
+ NaN,
374
+ NaN
375
+ ],
376
+ [
377
+ NaN,
378
+ NaN,
379
+ NaN,
380
+ NaN,
381
+ NaN,
382
+ NaN,
383
+ NaN,
384
+ NaN,
385
+ NaN,
386
+ NaN,
387
+ NaN,
388
+ NaN,
389
+ NaN,
390
+ NaN,
391
+ NaN,
392
+ NaN,
393
+ NaN,
394
+ NaN,
395
+ NaN,
396
+ NaN,
397
+ NaN,
398
+ NaN,
399
+ NaN
400
+ ],
401
+ [
402
+ NaN,
403
+ NaN,
404
+ NaN,
405
+ NaN,
406
+ NaN,
407
+ NaN,
408
+ NaN,
409
+ NaN,
410
+ NaN,
411
+ NaN,
412
+ NaN,
413
+ NaN,
414
+ NaN,
415
+ NaN,
416
+ NaN,
417
+ NaN,
418
+ NaN,
419
+ NaN,
420
+ NaN,
421
+ NaN,
422
+ NaN,
423
+ NaN,
424
+ NaN
425
+ ],
426
+ [
427
+ NaN,
428
+ NaN,
429
+ NaN,
430
+ NaN,
431
+ NaN,
432
+ NaN,
433
+ NaN,
434
+ NaN,
435
+ NaN,
436
+ NaN,
437
+ NaN,
438
+ NaN,
439
+ NaN,
440
+ NaN,
441
+ NaN,
442
+ NaN,
443
+ NaN,
444
+ NaN,
445
+ NaN,
446
+ NaN,
447
+ NaN,
448
+ NaN,
449
+ NaN
450
+ ],
451
+ [
452
+ NaN,
453
+ NaN,
454
+ NaN,
455
+ NaN,
456
+ NaN,
457
+ NaN,
458
+ NaN,
459
+ NaN,
460
+ NaN,
461
+ NaN,
462
+ NaN,
463
+ NaN,
464
+ NaN,
465
+ NaN,
466
+ NaN,
467
+ NaN,
468
+ NaN,
469
+ NaN,
470
+ NaN,
471
+ NaN,
472
+ NaN,
473
+ NaN,
474
+ NaN
475
+ ]
476
+ ],
477
+ "q_aim": [
478
+ [
479
+ NaN,
480
+ NaN,
481
+ NaN,
482
+ NaN,
483
+ NaN,
484
+ NaN,
485
+ NaN,
486
+ NaN,
487
+ NaN,
488
+ NaN,
489
+ NaN,
490
+ NaN,
491
+ NaN,
492
+ NaN,
493
+ NaN,
494
+ NaN,
495
+ NaN,
496
+ NaN,
497
+ NaN,
498
+ NaN,
499
+ NaN,
500
+ NaN,
501
+ NaN
502
+ ],
503
+ [
504
+ NaN,
505
+ NaN,
506
+ NaN,
507
+ NaN,
508
+ NaN,
509
+ NaN,
510
+ NaN,
511
+ NaN,
512
+ NaN,
513
+ NaN,
514
+ NaN,
515
+ NaN,
516
+ NaN,
517
+ NaN,
518
+ NaN,
519
+ NaN,
520
+ NaN,
521
+ NaN,
522
+ NaN,
523
+ NaN,
524
+ NaN,
525
+ NaN,
526
+ NaN
527
+ ],
528
+ [
529
+ NaN,
530
+ NaN,
531
+ NaN,
532
+ NaN,
533
+ NaN,
534
+ NaN,
535
+ NaN,
536
+ NaN,
537
+ NaN,
538
+ NaN,
539
+ NaN,
540
+ NaN,
541
+ NaN,
542
+ NaN,
543
+ NaN,
544
+ NaN,
545
+ NaN,
546
+ NaN,
547
+ NaN,
548
+ NaN,
549
+ NaN,
550
+ NaN,
551
+ NaN
552
+ ],
553
+ [
554
+ NaN,
555
+ NaN,
556
+ NaN,
557
+ NaN,
558
+ NaN,
559
+ NaN,
560
+ NaN,
561
+ NaN,
562
+ NaN,
563
+ NaN,
564
+ NaN,
565
+ NaN,
566
+ NaN,
567
+ NaN,
568
+ NaN,
569
+ NaN,
570
+ NaN,
571
+ NaN,
572
+ NaN,
573
+ NaN,
574
+ NaN,
575
+ NaN,
576
+ NaN
577
+ ],
578
+ [
579
+ NaN,
580
+ NaN,
581
+ NaN,
582
+ NaN,
583
+ NaN,
584
+ NaN,
585
+ NaN,
586
+ NaN,
587
+ NaN,
588
+ NaN,
589
+ NaN,
590
+ NaN,
591
+ NaN,
592
+ NaN,
593
+ NaN,
594
+ NaN,
595
+ NaN,
596
+ NaN,
597
+ NaN,
598
+ NaN,
599
+ NaN,
600
+ NaN,
601
+ NaN
602
+ ],
603
+ [
604
+ NaN,
605
+ NaN,
606
+ NaN,
607
+ NaN,
608
+ NaN,
609
+ NaN,
610
+ NaN,
611
+ NaN,
612
+ NaN,
613
+ NaN,
614
+ NaN,
615
+ NaN,
616
+ NaN,
617
+ NaN,
618
+ NaN,
619
+ NaN,
620
+ NaN,
621
+ NaN,
622
+ NaN,
623
+ NaN,
624
+ NaN,
625
+ NaN,
626
+ NaN
627
+ ]
628
+ ],
629
+ "q_base": [
630
+ [
631
+ NaN,
632
+ NaN,
633
+ NaN,
634
+ NaN,
635
+ NaN,
636
+ NaN,
637
+ NaN,
638
+ NaN,
639
+ NaN,
640
+ NaN,
641
+ NaN,
642
+ NaN,
643
+ NaN,
644
+ NaN,
645
+ NaN,
646
+ NaN,
647
+ NaN,
648
+ NaN,
649
+ NaN,
650
+ NaN,
651
+ NaN,
652
+ NaN,
653
+ NaN
654
+ ],
655
+ [
656
+ NaN,
657
+ NaN,
658
+ NaN,
659
+ NaN,
660
+ NaN,
661
+ NaN,
662
+ NaN,
663
+ NaN,
664
+ NaN,
665
+ NaN,
666
+ NaN,
667
+ NaN,
668
+ NaN,
669
+ NaN,
670
+ NaN,
671
+ NaN,
672
+ NaN,
673
+ NaN,
674
+ NaN,
675
+ NaN,
676
+ NaN,
677
+ NaN,
678
+ NaN
679
+ ],
680
+ [
681
+ NaN,
682
+ NaN,
683
+ NaN,
684
+ NaN,
685
+ NaN,
686
+ NaN,
687
+ NaN,
688
+ NaN,
689
+ NaN,
690
+ NaN,
691
+ NaN,
692
+ NaN,
693
+ NaN,
694
+ NaN,
695
+ NaN,
696
+ NaN,
697
+ NaN,
698
+ NaN,
699
+ NaN,
700
+ NaN,
701
+ NaN,
702
+ NaN,
703
+ NaN
704
+ ],
705
+ [
706
+ NaN,
707
+ NaN,
708
+ NaN,
709
+ NaN,
710
+ NaN,
711
+ NaN,
712
+ NaN,
713
+ NaN,
714
+ NaN,
715
+ NaN,
716
+ NaN,
717
+ NaN,
718
+ NaN,
719
+ NaN,
720
+ NaN,
721
+ NaN,
722
+ NaN,
723
+ NaN,
724
+ NaN,
725
+ NaN,
726
+ NaN,
727
+ NaN,
728
+ NaN
729
+ ],
730
+ [
731
+ NaN,
732
+ NaN,
733
+ NaN,
734
+ NaN,
735
+ NaN,
736
+ NaN,
737
+ NaN,
738
+ NaN,
739
+ NaN,
740
+ NaN,
741
+ NaN,
742
+ NaN,
743
+ NaN,
744
+ NaN,
745
+ NaN,
746
+ NaN,
747
+ NaN,
748
+ NaN,
749
+ NaN,
750
+ NaN,
751
+ NaN,
752
+ NaN,
753
+ NaN
754
+ ],
755
+ [
756
+ NaN,
757
+ NaN,
758
+ NaN,
759
+ NaN,
760
+ NaN,
761
+ NaN,
762
+ NaN,
763
+ NaN,
764
+ NaN,
765
+ NaN,
766
+ NaN,
767
+ NaN,
768
+ NaN,
769
+ NaN,
770
+ NaN,
771
+ NaN,
772
+ NaN,
773
+ NaN,
774
+ NaN,
775
+ NaN,
776
+ NaN,
777
+ NaN,
778
+ NaN
779
+ ]
780
+ ],
781
+ "property_delta_sd": [
782
+ [
783
+ NaN,
784
+ NaN,
785
+ NaN,
786
+ NaN,
787
+ NaN,
788
+ NaN,
789
+ NaN,
790
+ NaN,
791
+ NaN,
792
+ NaN,
793
+ NaN,
794
+ NaN,
795
+ NaN,
796
+ NaN,
797
+ NaN,
798
+ NaN,
799
+ NaN,
800
+ NaN,
801
+ NaN,
802
+ NaN,
803
+ NaN,
804
+ NaN,
805
+ NaN
806
+ ],
807
+ [
808
+ NaN,
809
+ NaN,
810
+ NaN,
811
+ NaN,
812
+ NaN,
813
+ NaN,
814
+ NaN,
815
+ NaN,
816
+ NaN,
817
+ NaN,
818
+ NaN,
819
+ NaN,
820
+ NaN,
821
+ NaN,
822
+ NaN,
823
+ NaN,
824
+ NaN,
825
+ NaN,
826
+ NaN,
827
+ NaN,
828
+ NaN,
829
+ NaN,
830
+ NaN
831
+ ],
832
+ [
833
+ NaN,
834
+ NaN,
835
+ NaN,
836
+ NaN,
837
+ NaN,
838
+ NaN,
839
+ NaN,
840
+ NaN,
841
+ NaN,
842
+ NaN,
843
+ NaN,
844
+ NaN,
845
+ NaN,
846
+ NaN,
847
+ NaN,
848
+ NaN,
849
+ NaN,
850
+ NaN,
851
+ NaN,
852
+ NaN,
853
+ NaN,
854
+ NaN,
855
+ NaN
856
+ ],
857
+ [
858
+ NaN,
859
+ NaN,
860
+ NaN,
861
+ NaN,
862
+ NaN,
863
+ NaN,
864
+ NaN,
865
+ NaN,
866
+ NaN,
867
+ NaN,
868
+ NaN,
869
+ NaN,
870
+ NaN,
871
+ NaN,
872
+ NaN,
873
+ NaN,
874
+ NaN,
875
+ NaN,
876
+ NaN,
877
+ NaN,
878
+ NaN,
879
+ NaN,
880
+ NaN
881
+ ],
882
+ [
883
+ NaN,
884
+ NaN,
885
+ NaN,
886
+ NaN,
887
+ NaN,
888
+ NaN,
889
+ NaN,
890
+ NaN,
891
+ NaN,
892
+ NaN,
893
+ NaN,
894
+ NaN,
895
+ NaN,
896
+ NaN,
897
+ NaN,
898
+ NaN,
899
+ NaN,
900
+ NaN,
901
+ NaN,
902
+ NaN,
903
+ NaN,
904
+ NaN,
905
+ NaN
906
+ ],
907
+ [
908
+ NaN,
909
+ NaN,
910
+ NaN,
911
+ NaN,
912
+ NaN,
913
+ NaN,
914
+ NaN,
915
+ NaN,
916
+ NaN,
917
+ NaN,
918
+ NaN,
919
+ NaN,
920
+ NaN,
921
+ NaN,
922
+ NaN,
923
+ NaN,
924
+ NaN,
925
+ NaN,
926
+ NaN,
927
+ NaN,
928
+ NaN,
929
+ NaN,
930
+ NaN
931
+ ]
932
+ ],
933
+ "q_property_delta": [
934
+ [
935
+ NaN,
936
+ NaN,
937
+ NaN,
938
+ NaN,
939
+ NaN,
940
+ NaN,
941
+ NaN,
942
+ NaN,
943
+ NaN,
944
+ NaN,
945
+ NaN,
946
+ NaN,
947
+ NaN,
948
+ NaN,
949
+ NaN,
950
+ NaN,
951
+ NaN,
952
+ NaN,
953
+ NaN,
954
+ NaN,
955
+ NaN,
956
+ NaN,
957
+ NaN
958
+ ],
959
+ [
960
+ NaN,
961
+ NaN,
962
+ NaN,
963
+ NaN,
964
+ NaN,
965
+ NaN,
966
+ NaN,
967
+ NaN,
968
+ NaN,
969
+ NaN,
970
+ NaN,
971
+ NaN,
972
+ NaN,
973
+ NaN,
974
+ NaN,
975
+ NaN,
976
+ NaN,
977
+ NaN,
978
+ NaN,
979
+ NaN,
980
+ NaN,
981
+ NaN,
982
+ NaN
983
+ ],
984
+ [
985
+ NaN,
986
+ NaN,
987
+ NaN,
988
+ NaN,
989
+ NaN,
990
+ NaN,
991
+ NaN,
992
+ NaN,
993
+ NaN,
994
+ NaN,
995
+ NaN,
996
+ NaN,
997
+ NaN,
998
+ NaN,
999
+ NaN,
1000
+ NaN,
1001
+ NaN,
1002
+ NaN,
1003
+ NaN,
1004
+ NaN,
1005
+ NaN,
1006
+ NaN,
1007
+ NaN
1008
+ ],
1009
+ [
1010
+ NaN,
1011
+ NaN,
1012
+ NaN,
1013
+ NaN,
1014
+ NaN,
1015
+ NaN,
1016
+ NaN,
1017
+ NaN,
1018
+ NaN,
1019
+ NaN,
1020
+ NaN,
1021
+ NaN,
1022
+ NaN,
1023
+ NaN,
1024
+ NaN,
1025
+ NaN,
1026
+ NaN,
1027
+ NaN,
1028
+ NaN,
1029
+ NaN,
1030
+ NaN,
1031
+ NaN,
1032
+ NaN
1033
+ ],
1034
+ [
1035
+ NaN,
1036
+ NaN,
1037
+ NaN,
1038
+ NaN,
1039
+ NaN,
1040
+ NaN,
1041
+ NaN,
1042
+ NaN,
1043
+ NaN,
1044
+ NaN,
1045
+ NaN,
1046
+ NaN,
1047
+ NaN,
1048
+ NaN,
1049
+ NaN,
1050
+ NaN,
1051
+ NaN,
1052
+ NaN,
1053
+ NaN,
1054
+ NaN,
1055
+ NaN,
1056
+ NaN,
1057
+ NaN
1058
+ ],
1059
+ [
1060
+ NaN,
1061
+ NaN,
1062
+ NaN,
1063
+ NaN,
1064
+ NaN,
1065
+ NaN,
1066
+ NaN,
1067
+ NaN,
1068
+ NaN,
1069
+ NaN,
1070
+ NaN,
1071
+ NaN,
1072
+ NaN,
1073
+ NaN,
1074
+ NaN,
1075
+ NaN,
1076
+ NaN,
1077
+ NaN,
1078
+ NaN,
1079
+ NaN,
1080
+ NaN,
1081
+ NaN,
1082
+ NaN
1083
+ ]
1084
+ ],
1085
+ "q_property_delta_combo_clustered": [
1086
+ [
1087
+ NaN,
1088
+ NaN,
1089
+ NaN,
1090
+ NaN,
1091
+ NaN,
1092
+ NaN,
1093
+ NaN,
1094
+ NaN,
1095
+ NaN,
1096
+ NaN,
1097
+ NaN,
1098
+ NaN,
1099
+ NaN,
1100
+ NaN,
1101
+ NaN,
1102
+ NaN,
1103
+ NaN,
1104
+ NaN,
1105
+ NaN,
1106
+ NaN,
1107
+ NaN,
1108
+ NaN,
1109
+ NaN
1110
+ ],
1111
+ [
1112
+ NaN,
1113
+ NaN,
1114
+ NaN,
1115
+ NaN,
1116
+ NaN,
1117
+ NaN,
1118
+ NaN,
1119
+ NaN,
1120
+ NaN,
1121
+ NaN,
1122
+ NaN,
1123
+ NaN,
1124
+ NaN,
1125
+ NaN,
1126
+ NaN,
1127
+ NaN,
1128
+ NaN,
1129
+ NaN,
1130
+ NaN,
1131
+ NaN,
1132
+ NaN,
1133
+ NaN,
1134
+ NaN
1135
+ ],
1136
+ [
1137
+ NaN,
1138
+ NaN,
1139
+ NaN,
1140
+ NaN,
1141
+ NaN,
1142
+ NaN,
1143
+ NaN,
1144
+ NaN,
1145
+ NaN,
1146
+ NaN,
1147
+ NaN,
1148
+ NaN,
1149
+ NaN,
1150
+ NaN,
1151
+ NaN,
1152
+ NaN,
1153
+ NaN,
1154
+ NaN,
1155
+ NaN,
1156
+ NaN,
1157
+ NaN,
1158
+ NaN,
1159
+ NaN
1160
+ ],
1161
+ [
1162
+ NaN,
1163
+ NaN,
1164
+ NaN,
1165
+ NaN,
1166
+ NaN,
1167
+ NaN,
1168
+ NaN,
1169
+ NaN,
1170
+ NaN,
1171
+ NaN,
1172
+ NaN,
1173
+ NaN,
1174
+ NaN,
1175
+ NaN,
1176
+ NaN,
1177
+ NaN,
1178
+ NaN,
1179
+ NaN,
1180
+ NaN,
1181
+ NaN,
1182
+ NaN,
1183
+ NaN,
1184
+ NaN
1185
+ ],
1186
+ [
1187
+ NaN,
1188
+ NaN,
1189
+ NaN,
1190
+ NaN,
1191
+ NaN,
1192
+ NaN,
1193
+ NaN,
1194
+ NaN,
1195
+ NaN,
1196
+ NaN,
1197
+ NaN,
1198
+ NaN,
1199
+ NaN,
1200
+ NaN,
1201
+ NaN,
1202
+ NaN,
1203
+ NaN,
1204
+ NaN,
1205
+ NaN,
1206
+ NaN,
1207
+ NaN,
1208
+ NaN,
1209
+ NaN
1210
+ ],
1211
+ [
1212
+ NaN,
1213
+ NaN,
1214
+ NaN,
1215
+ NaN,
1216
+ NaN,
1217
+ NaN,
1218
+ NaN,
1219
+ NaN,
1220
+ NaN,
1221
+ NaN,
1222
+ NaN,
1223
+ NaN,
1224
+ NaN,
1225
+ NaN,
1226
+ NaN,
1227
+ NaN,
1228
+ NaN,
1229
+ NaN,
1230
+ NaN,
1231
+ NaN,
1232
+ NaN,
1233
+ NaN,
1234
+ NaN
1235
+ ]
1236
+ ],
1237
+ "metrics": [
1238
+ "weight_cosine",
1239
+ "qmd_raw",
1240
+ "coord_fraction",
1241
+ "subspace_overlap",
1242
+ "spectral_overcounting",
1243
+ "geo_cka_mean",
1244
+ "geo_procrustes_mean",
1245
+ "geo_subspace_overlap_mean",
1246
+ "geo_svcca_mean",
1247
+ "geo_cka_late_minus_early",
1248
+ "ret_identity_p_at_1",
1249
+ "ret_procrustes_p_at_1",
1250
+ "ret_ridge_p_at_1",
1251
+ "ret_gain_over_identity",
1252
+ "grad_cosine",
1253
+ "grad_l2",
1254
+ "grad_norm_ratio",
1255
+ "grad_cosine_layer_min",
1256
+ "beh_js",
1257
+ "beh_logit_cosine",
1258
+ "beh_topk_overlap",
1259
+ "beh_rank_corr",
1260
+ "beh_entropy_gap"
1261
+ ],
1262
+ "rows": [
1263
+ "Task Arithmetic",
1264
+ "TIES",
1265
+ "DARE Linear",
1266
+ "DARE TIES",
1267
+ "WIDEN",
1268
+ "all (within-op.)"
1269
+ ]
1270
+ }
tables/T1_paired_per_checkpoint.csv CHANGED
@@ -1,101 +1,189 @@
1
- operator,operator_label,combo,metric,family,without_aim,with_aim,difference,difference_sd_units
2
- DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,act_deviation_from_base,representation,0.3268741406500339,0.2771998148411512,-0.0496743258088827,-0.6024211401198292
3
- DARETaskArithmetic,DARE Linear,Code-Math,act_deviation_from_base,representation,0.2090091705322265,0.18324254406616086,-0.025766626466065645,-0.3124825599536037
4
- DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.41396001614630223,0.3468391884118318,-0.06712082773447042,-0.814002100907342
5
- DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,act_deviation_from_base,representation,0.39771203845739367,0.34087302163243294,-0.05683901682496073,-0.6893103179844277
6
- DARETies,DARE TIES,Code-Instruction_Tuned,act_deviation_from_base,representation,0.32205841317772865,0.2706678189337253,-0.05139059424400333,-0.6232350388613026
7
- DARETies,DARE TIES,Code-Math,act_deviation_from_base,representation,0.21137537630274889,0.18471119441092013,-0.02666418189182876,-0.3233675866571219
8
- DARETies,DARE TIES,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.41742832884192466,0.34477567411959165,-0.07265265472233301,-0.881088859845858
9
- DARETies,DARE TIES,Math-Instruction_Tuned,act_deviation_from_base,representation,0.40017827302217485,0.3404878456145525,-0.05969042740762237,-0.7238905560890303
10
- TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,act_deviation_from_base,representation,0.31902736425399775,0.2703720036894083,-0.04865536056458947,-0.5900637262201152
11
- TaskArithmetic,Task Arithmetic,Code-Math,act_deviation_from_base,representation,0.2079064922407269,0.18354407707229253,-0.024362415168434376,-0.2954531074726018
12
- TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.405633632466197,0.34357630088925356,-0.06205733157694343,-0.7525949840811219
13
- TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,act_deviation_from_base,representation,0.39210564494132993,0.33807642459869386,-0.054029220342636075,-0.6552347513245169
14
- Ties,TIES,Code-Instruction_Tuned,act_deviation_from_base,representation,0.26751709580421446,0.22590580526739354,-0.041611290536820916,-0.5046373690806403
15
- Ties,TIES,Code-Math,act_deviation_from_base,representation,0.18338995799422259,0.16238840231671925,-0.021001555677503336,-0.25469457128030626
16
- Ties,TIES,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.2203250393271446,0.18917316654697058,-0.031151872780174006,-0.3777916743007811
17
- Ties,TIES,Math-Instruction_Tuned,act_deviation_from_base,representation,0.2414184311404824,0.20530384760349984,-0.03611458353698255,-0.4379765247948554
18
- WIDEN,WIDEN,Code-Instruction_Tuned,act_deviation_from_base,representation,0.21522565763443705,0.18980246856808658,-0.025423189066350466,-0.30831755224535184
19
- WIDEN,WIDEN,Code-Math,act_deviation_from_base,representation,0.1499567064456641,0.13331147627905007,-0.01664523016661404,-0.2018636060227215
20
- WIDEN,WIDEN,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.24535813517868515,0.21107980906963345,-0.03427832610905171,-0.4157074697996532
21
- WIDEN,WIDEN,Math-Instruction_Tuned,act_deviation_from_base,representation,0.27196276579052203,0.23418197268620128,-0.03778079310432075,-0.45818333889629526
22
- DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,cka_to_base,representation,0.9630374073982239,0.9756896287202835,0.012652221322059676,0.8473500161617931
23
- DARETaskArithmetic,DARE Linear,Code-Math,cka_to_base,representation,0.988602839410305,0.9910977050662041,0.00249486565589907,0.16708721733800921
24
- DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9444771707057953,0.9662201166152954,0.0217429459095001,1.4561779389439773
25
- DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,cka_to_base,representation,0.9561630114912987,0.9691109836101532,0.012947972118854567,0.8671571659155738
26
- DARETies,DARE TIES,Code-Instruction_Tuned,cka_to_base,representation,0.9662329286336899,0.9789113625884056,0.012678433954715684,0.8491055398867839
27
- DARETies,DARE TIES,Code-Math,cka_to_base,representation,0.9881209120154381,0.9910706728696823,0.0029497608542442544,0.19755265450181037
28
- DARETies,DARE TIES,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9456815689802169,0.9694720149040222,0.023790445923805303,1.5933039918361938
29
- DARETies,DARE TIES,Math-Instruction_Tuned,cka_to_base,representation,0.9561433091759681,0.9701771914958954,0.014033882319927304,0.9398832116436016
30
- TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,cka_to_base,representation,0.9668349489569664,0.9788408130407333,0.012005864083766915,0.8040619007888237
31
- TaskArithmetic,Task Arithmetic,Code-Math,cka_to_base,representation,0.988736729323864,0.9910610884428024,0.0023243591189383572,0.15566798010118926
32
- TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9509493231773376,0.9675659328699112,0.016616609692573614,1.1128547417209058
33
- TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,cka_to_base,representation,0.959263214468956,0.9698800936341285,0.01061687916517251,0.7110382045334173
34
- Ties,TIES,Code-Instruction_Tuned,cka_to_base,representation,0.9842474162578583,0.9911796778440476,0.006932261586189314,0.4642704089323539
35
- Ties,TIES,Code-Math,cka_to_base,representation,0.9912186995148659,0.99300397336483,0.0017852738499641863,0.1195641292634388
36
- Ties,TIES,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9914436548948288,0.9942833751440048,0.002839720249176003,0.19018296770061102
37
- Ties,TIES,Math-Instruction_Tuned,cka_to_base,representation,0.988047470152378,0.9926868349313736,0.004639364778995514,0.31070953632530524
38
- WIDEN,WIDEN,Code-Instruction_Tuned,cka_to_base,representation,0.9884805098176003,0.9913797855377198,0.0028992757201194985,0.19417154235338818
39
- WIDEN,WIDEN,Code-Math,cka_to_base,representation,0.9945898994803428,0.9956279069185257,0.0010380074381828308,0.06951788125826895
40
- WIDEN,WIDEN,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9861174300312996,0.990828663110733,0.004711233079433441,0.31552273109857565
41
- WIDEN,WIDEN,Math-Instruction_Tuned,cka_to_base,representation,0.9840415135025978,0.9888819053769111,0.004840391874313288,0.32417280954275035
42
- DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.07540532551926302,0.06918316821085496,-0.006222157308408055,-0.30576580173924633
43
- DARETaskArithmetic,DARE Linear,Code-Math,qmd_raw_from_base,weight space,0.027830858351205065,0.025713600634237572,-0.002117257716967493,-0.10404510384884595
44
- DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.08106394490993216,0.07435638722862838,-0.00670755768130378,-0.3296190775126555
45
- DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.08101604946480126,0.07431673551507964,-0.006699313949721619,-0.32921396862973856
46
- DARETies,DARE TIES,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.06789361068416809,0.06231974267677706,-0.005573868007391028,-0.2739079286480945
47
- DARETies,DARE TIES,Code-Math,qmd_raw_from_base,weight space,0.02532894138283997,0.023444206348130697,-0.0018847350347092721,-0.09261860322547368
48
- DARETies,DARE TIES,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.07842935274588043,0.07191654395449834,-0.006512808791382094,-0.32004883563139264
49
- DARETies,DARE TIES,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.07794717115598046,0.0715050015495038,-0.006442169606476658,-0.3165775240048559
50
- TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.06748542751022336,0.06197007668653693,-0.005515350823686434,-0.27103230967800196
51
- TaskArithmetic,Task Arithmetic,Code-Math,qmd_raw_from_base,weight space,0.02494948531349566,0.023119563194417308,-0.0018299221190783525,-0.08992501734153821
52
- TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.07272877584972695,0.06674016389376215,-0.005988611955964798,-0.29428904562512653
53
- TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.07268515892821595,0.06670400911641289,-0.00598114981180306,-0.29392234507751674
54
- Ties,TIES,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.06307845715669955,0.058067960792707346,-0.005010496363992201,-0.24622303196632414
55
- Ties,TIES,Code-Math,qmd_raw_from_base,weight space,0.02326307511752211,0.021652956885821834,-0.0016101182317002777,-0.07912353667843433
56
- Ties,TIES,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.05446767310726107,0.050203098119900015,-0.004264574987361054,-0.20956737759193825
57
- Ties,TIES,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.05786696753653783,0.05320746697291281,-0.004659500563625026,-0.22897459111424992
58
- WIDEN,WIDEN,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.050647287040768006,0.0466550100292541,-0.003992277011513903,-0.19618626156255584
59
- WIDEN,WIDEN,Code-Math,qmd_raw_from_base,weight space,0.019018005658702125,0.017786045338737237,-0.0012319603199648886,-0.06054031042191369
60
- WIDEN,WIDEN,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.05039161050891783,0.046407742875562194,-0.003983867633355634,-0.195773012567514
61
- WIDEN,WIDEN,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.05479621414229463,0.05043220024105492,-0.004364013901239708,-0.21445394951854316
62
- DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0675294344035755,0.054332467446118,-0.013196966957457507,-0.26534451678728566
63
- DARETaskArithmetic,DARE Linear,Code-Math,beh_js_to_base,behaviour,0.0701002699547847,0.0596478397893982,-0.010452430165386502,-0.2101615500310129
64
- DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1649849997656848,0.1531925715371772,-0.011792428228507595,-0.23710419069238398
65
- DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1558226350073653,0.1452851978753024,-0.010537437132062899,-0.21187074067830441
66
- DARETies,DARE TIES,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0645529807269808,0.0522127097471974,-0.012340270979783408,-0.2481193785443587
67
- DARETies,DARE TIES,Code-Math,beh_js_to_base,behaviour,0.0705667963771806,0.0605886174584691,-0.009978178918711493,-0.20062602809704028
68
- DARETies,DARE TIES,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1579099506890481,0.1434274385153199,-0.014482512173728201,-0.2911923025185968
69
- DARETies,DARE TIES,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1516165943578546,0.136825735452024,-0.014790858905830595,-0.2973920690934773
70
- TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0670005273723255,0.0541176966435751,-0.012882830728750391,-0.2590283438302437
71
- TaskArithmetic,Task Arithmetic,Code-Math,beh_js_to_base,behaviour,0.0707630679739104,0.0605505583717166,-0.010212509602193796,-0.20533759266922477
72
- TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1723386903043327,0.1557380097510118,-0.016600680553320896,-0.3337812069971008
73
- TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1630553862854352,0.1491063061455251,-0.013949080139910086,-0.2804668634303113
74
- Ties,TIES,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0401827128762879,0.0320304306451685,-0.008152282231119398,-0.1639136777642354
75
- Ties,TIES,Code-Math,beh_js_to_base,behaviour,0.0486724480616445,0.0419355507932006,-0.006736897268443905,-0.13545527211695216
76
- Ties,TIES,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.0290378217286798,0.0234016886107479,-0.005636133117931898,-0.11332278269298476
77
- Ties,TIES,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.0397027983565525,0.0319435968527206,-0.0077592015038319,-0.15601020903716128
78
- WIDEN,WIDEN,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0363831441722982,0.0279335037242935,-0.008449640448004704,-0.16989250400715278
79
- WIDEN,WIDEN,Code-Math,beh_js_to_base,behaviour,0.0365018882611834,0.0305083815052392,-0.005993506755944199,-0.1205083076394803
80
- WIDEN,WIDEN,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.0814156675451293,0.0615861280156665,-0.0198295395294628,-0.39870218676166064
81
- WIDEN,WIDEN,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1038943280401073,0.0794474548727698,-0.0244468731673375,-0.4915405008179976
82
- DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.3067125690407576,0.2416893078428814,-0.06502326119787621,-0.2990345698933688
83
- DARETaskArithmetic,DARE Linear,Code-Math,beh_kl_to_base,behaviour,0.3002615200469781,0.2594705143337791,-0.04079100571319899,-0.18759318779542877
84
- DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.7147236138251114,0.64177468547017,-0.07294892835494138,-0.33548381014630757
85
- DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.6873062086352962,0.611457267630225,-0.07584894100507122,-0.3488206379144176
86
- DARETies,DARE TIES,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.2927892901591197,0.2313560658544693,-0.0614332243046504,-0.2825243992484339
87
- DARETies,DARE TIES,Code-Math,beh_kl_to_base,behaviour,0.3048238893372912,0.2638059280120493,-0.041017961325241936,-0.18863693079726954
88
- DARETies,DARE TIES,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.7098179752433407,0.6148826686629932,-0.09493530658034743,-0.436596658610481
89
- DARETies,DARE TIES,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.6873289485144918,0.5846292800262997,-0.10269966848819212,-0.4723040744003957
90
- TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.304607594224647,0.2406390341662837,-0.0639685600583633,-0.29418411951899426
91
- TaskArithmetic,Task Arithmetic,Code-Math,beh_kl_to_base,behaviour,0.303996269463866,0.2630344646865175,-0.04096180477734851,-0.18837867323164387
92
- TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.7553384977087472,0.6549430652966949,-0.10039543241205229,-0.46170715521689953
93
- TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.7302715067695107,0.6341559539274912,-0.09611555284201945,-0.44202447669781614
94
- Ties,TIES,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.1751193129121662,0.1364185276023318,-0.038700785309834396,-0.1779805023073793
95
- Ties,TIES,Code-Math,beh_kl_to_base,behaviour,0.2104004919723689,0.1810295336203008,-0.0293709583520681,-0.13507369111245993
96
- Ties,TIES,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.1210440478162356,0.0967837872615697,-0.02426026055466589,-0.11157017422409021
97
- Ties,TIES,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.1695624375570265,0.1348951166119359,-0.03466732094509059,-0.15943105924236753
98
- WIDEN,WIDEN,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.1487084594954175,0.1135898635347647,-0.035118595960652796,-0.16150642162340226
99
- WIDEN,WIDEN,Code-Math,beh_kl_to_base,behaviour,0.1526036240678989,0.127567749296493,-0.025035874771405897,-0.11513713563810195
100
- WIDEN,WIDEN,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.3255842181369209,0.2459603207798351,-0.0796238973570858,-0.3661812320817214
101
- WIDEN,WIDEN,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.4269925540304505,0.3248536287939473,-0.1021389252365032,-0.4697252800735605
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ operator,operator_label,combo,metric,family,without_aim,with_aim,difference,difference_sd_units,GSM8K__aim0,GSM8K__aim1,HV__aim0,HV__aim1,HumanEval__aim0,HumanEval__aim1,IFEval__aim0,IFEval__aim1,MATH__aim0,MATH__aim1,MBPP__aim0,MBPP__aim1,MMLU__aim0,MMLU__aim1,endpoint_scaled__aim0,endpoint_scaled__aim1,endpoint_scaled__diff,HV__diff,HumanEval__diff,MBPP__diff,MMLU__diff,MATH__diff,GSM8K__diff,IFEval__diff
2
+ DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,act_deviation_from_base,representation,0.3268741406500339,0.2771998148411512,-0.0496743258088827,-0.6024211401198292,,,,,,,,,,,,,,,,,,,,,,,,
3
+ DARETaskArithmetic,DARE Linear,Code-Math,act_deviation_from_base,representation,0.2090091705322265,0.18324254406616086,-0.025766626466065645,-0.3124825599536037,,,,,,,,,,,,,,,,,,,,,,,,
4
+ DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.41396001614630223,0.3468391884118318,-0.06712082773447042,-0.814002100907342,,,,,,,,,,,,,,,,,,,,,,,,
5
+ DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,act_deviation_from_base,representation,0.39771203845739367,0.34087302163243294,-0.05683901682496073,-0.6893103179844277,,,,,,,,,,,,,,,,,,,,,,,,
6
+ DARETies,DARE TIES,Code-Instruction_Tuned,act_deviation_from_base,representation,0.32205841317772865,0.2706678189337253,-0.05139059424400333,-0.6232350388613026,,,,,,,,,,,,,,,,,,,,,,,,
7
+ DARETies,DARE TIES,Code-Math,act_deviation_from_base,representation,0.21137537630274889,0.18471119441092013,-0.02666418189182876,-0.3233675866571219,,,,,,,,,,,,,,,,,,,,,,,,
8
+ DARETies,DARE TIES,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.41742832884192466,0.34477567411959165,-0.07265265472233301,-0.881088859845858,,,,,,,,,,,,,,,,,,,,,,,,
9
+ DARETies,DARE TIES,Math-Instruction_Tuned,act_deviation_from_base,representation,0.40017827302217485,0.3404878456145525,-0.05969042740762237,-0.7238905560890303,,,,,,,,,,,,,,,,,,,,,,,,
10
+ TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,act_deviation_from_base,representation,0.31902736425399775,0.2703720036894083,-0.04865536056458947,-0.5900637262201152,,,,,,,,,,,,,,,,,,,,,,,,
11
+ TaskArithmetic,Task Arithmetic,Code-Math,act_deviation_from_base,representation,0.2079064922407269,0.18354407707229253,-0.024362415168434376,-0.2954531074726018,,,,,,,,,,,,,,,,,,,,,,,,
12
+ TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.405633632466197,0.34357630088925356,-0.06205733157694343,-0.7525949840811219,,,,,,,,,,,,,,,,,,,,,,,,
13
+ TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,act_deviation_from_base,representation,0.39210564494132993,0.33807642459869386,-0.054029220342636075,-0.6552347513245169,,,,,,,,,,,,,,,,,,,,,,,,
14
+ Ties,TIES,Code-Instruction_Tuned,act_deviation_from_base,representation,0.26751709580421446,0.22590580526739354,-0.041611290536820916,-0.5046373690806403,,,,,,,,,,,,,,,,,,,,,,,,
15
+ Ties,TIES,Code-Math,act_deviation_from_base,representation,0.18338995799422259,0.16238840231671925,-0.021001555677503336,-0.25469457128030626,,,,,,,,,,,,,,,,,,,,,,,,
16
+ Ties,TIES,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.2203250393271446,0.18917316654697058,-0.031151872780174006,-0.3777916743007811,,,,,,,,,,,,,,,,,,,,,,,,
17
+ Ties,TIES,Math-Instruction_Tuned,act_deviation_from_base,representation,0.2414184311404824,0.20530384760349984,-0.03611458353698255,-0.4379765247948554,,,,,,,,,,,,,,,,,,,,,,,,
18
+ WIDEN,WIDEN,Code-Instruction_Tuned,act_deviation_from_base,representation,0.21522565763443705,0.18980246856808658,-0.025423189066350466,-0.30831755224535184,,,,,,,,,,,,,,,,,,,,,,,,
19
+ WIDEN,WIDEN,Code-Math,act_deviation_from_base,representation,0.1499567064456641,0.13331147627905007,-0.01664523016661404,-0.2018636060227215,,,,,,,,,,,,,,,,,,,,,,,,
20
+ WIDEN,WIDEN,Code-Math-Instruction_Tuned,act_deviation_from_base,representation,0.24535813517868515,0.21107980906963345,-0.03427832610905171,-0.4157074697996532,,,,,,,,,,,,,,,,,,,,,,,,
21
+ WIDEN,WIDEN,Math-Instruction_Tuned,act_deviation_from_base,representation,0.27196276579052203,0.23418197268620128,-0.03778079310432075,-0.45818333889629526,,,,,,,,,,,,,,,,,,,,,,,,
22
+ DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,cka_to_base,representation,0.9630374073982239,0.9756896287202835,0.012652221322059676,0.8473500161617931,,,,,,,,,,,,,,,,,,,,,,,,
23
+ DARETaskArithmetic,DARE Linear,Code-Math,cka_to_base,representation,0.988602839410305,0.9910977050662041,0.00249486565589907,0.16708721733800921,,,,,,,,,,,,,,,,,,,,,,,,
24
+ DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9444771707057953,0.9662201166152954,0.0217429459095001,1.4561779389439773,,,,,,,,,,,,,,,,,,,,,,,,
25
+ DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,cka_to_base,representation,0.9561630114912987,0.9691109836101532,0.012947972118854567,0.8671571659155738,,,,,,,,,,,,,,,,,,,,,,,,
26
+ DARETies,DARE TIES,Code-Instruction_Tuned,cka_to_base,representation,0.9662329286336899,0.9789113625884056,0.012678433954715684,0.8491055398867839,,,,,,,,,,,,,,,,,,,,,,,,
27
+ DARETies,DARE TIES,Code-Math,cka_to_base,representation,0.9881209120154381,0.9910706728696823,0.0029497608542442544,0.19755265450181037,,,,,,,,,,,,,,,,,,,,,,,,
28
+ DARETies,DARE TIES,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9456815689802169,0.9694720149040222,0.023790445923805303,1.5933039918361938,,,,,,,,,,,,,,,,,,,,,,,,
29
+ DARETies,DARE TIES,Math-Instruction_Tuned,cka_to_base,representation,0.9561433091759681,0.9701771914958954,0.014033882319927304,0.9398832116436016,,,,,,,,,,,,,,,,,,,,,,,,
30
+ TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,cka_to_base,representation,0.9668349489569664,0.9788408130407333,0.012005864083766915,0.8040619007888237,,,,,,,,,,,,,,,,,,,,,,,,
31
+ TaskArithmetic,Task Arithmetic,Code-Math,cka_to_base,representation,0.988736729323864,0.9910610884428024,0.0023243591189383572,0.15566798010118926,,,,,,,,,,,,,,,,,,,,,,,,
32
+ TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9509493231773376,0.9675659328699112,0.016616609692573614,1.1128547417209058,,,,,,,,,,,,,,,,,,,,,,,,
33
+ TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,cka_to_base,representation,0.959263214468956,0.9698800936341285,0.01061687916517251,0.7110382045334173,,,,,,,,,,,,,,,,,,,,,,,,
34
+ Ties,TIES,Code-Instruction_Tuned,cka_to_base,representation,0.9842474162578583,0.9911796778440476,0.006932261586189314,0.4642704089323539,,,,,,,,,,,,,,,,,,,,,,,,
35
+ Ties,TIES,Code-Math,cka_to_base,representation,0.9912186995148659,0.99300397336483,0.0017852738499641863,0.1195641292634388,,,,,,,,,,,,,,,,,,,,,,,,
36
+ Ties,TIES,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9914436548948288,0.9942833751440048,0.002839720249176003,0.19018296770061102,,,,,,,,,,,,,,,,,,,,,,,,
37
+ Ties,TIES,Math-Instruction_Tuned,cka_to_base,representation,0.988047470152378,0.9926868349313736,0.004639364778995514,0.31070953632530524,,,,,,,,,,,,,,,,,,,,,,,,
38
+ WIDEN,WIDEN,Code-Instruction_Tuned,cka_to_base,representation,0.9884805098176003,0.9913797855377198,0.0028992757201194985,0.19417154235338818,,,,,,,,,,,,,,,,,,,,,,,,
39
+ WIDEN,WIDEN,Code-Math,cka_to_base,representation,0.9945898994803428,0.9956279069185257,0.0010380074381828308,0.06951788125826895,,,,,,,,,,,,,,,,,,,,,,,,
40
+ WIDEN,WIDEN,Code-Math-Instruction_Tuned,cka_to_base,representation,0.9861174300312996,0.990828663110733,0.004711233079433441,0.31552273109857565,,,,,,,,,,,,,,,,,,,,,,,,
41
+ WIDEN,WIDEN,Math-Instruction_Tuned,cka_to_base,representation,0.9840415135025978,0.9888819053769111,0.004840391874313288,0.32417280954275035,,,,,,,,,,,,,,,,,,,,,,,,
42
+ DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.07540532551926302,0.06918316821085496,-0.006222157308408055,-0.30576580173924633,,,,,,,,,,,,,,,,,,,,,,,,
43
+ DARETaskArithmetic,DARE Linear,Code-Math,qmd_raw_from_base,weight space,0.027830858351205065,0.025713600634237572,-0.002117257716967493,-0.10404510384884595,,,,,,,,,,,,,,,,,,,,,,,,
44
+ DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.08106394490993216,0.07435638722862838,-0.00670755768130378,-0.3296190775126555,,,,,,,,,,,,,,,,,,,,,,,,
45
+ DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.08101604946480126,0.07431673551507964,-0.006699313949721619,-0.32921396862973856,,,,,,,,,,,,,,,,,,,,,,,,
46
+ DARETies,DARE TIES,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.06789361068416809,0.06231974267677706,-0.005573868007391028,-0.2739079286480945,,,,,,,,,,,,,,,,,,,,,,,,
47
+ DARETies,DARE TIES,Code-Math,qmd_raw_from_base,weight space,0.02532894138283997,0.023444206348130697,-0.0018847350347092721,-0.09261860322547368,,,,,,,,,,,,,,,,,,,,,,,,
48
+ DARETies,DARE TIES,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.07842935274588043,0.07191654395449834,-0.006512808791382094,-0.32004883563139264,,,,,,,,,,,,,,,,,,,,,,,,
49
+ DARETies,DARE TIES,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.07794717115598046,0.0715050015495038,-0.006442169606476658,-0.3165775240048559,,,,,,,,,,,,,,,,,,,,,,,,
50
+ TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.06748542751022336,0.06197007668653693,-0.005515350823686434,-0.27103230967800196,,,,,,,,,,,,,,,,,,,,,,,,
51
+ TaskArithmetic,Task Arithmetic,Code-Math,qmd_raw_from_base,weight space,0.02494948531349566,0.023119563194417308,-0.0018299221190783525,-0.08992501734153821,,,,,,,,,,,,,,,,,,,,,,,,
52
+ TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.07272877584972695,0.06674016389376215,-0.005988611955964798,-0.29428904562512653,,,,,,,,,,,,,,,,,,,,,,,,
53
+ TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.07268515892821595,0.06670400911641289,-0.00598114981180306,-0.29392234507751674,,,,,,,,,,,,,,,,,,,,,,,,
54
+ Ties,TIES,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.06307845715669955,0.058067960792707346,-0.005010496363992201,-0.24622303196632414,,,,,,,,,,,,,,,,,,,,,,,,
55
+ Ties,TIES,Code-Math,qmd_raw_from_base,weight space,0.02326307511752211,0.021652956885821834,-0.0016101182317002777,-0.07912353667843433,,,,,,,,,,,,,,,,,,,,,,,,
56
+ Ties,TIES,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.05446767310726107,0.050203098119900015,-0.004264574987361054,-0.20956737759193825,,,,,,,,,,,,,,,,,,,,,,,,
57
+ Ties,TIES,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.05786696753653783,0.05320746697291281,-0.004659500563625026,-0.22897459111424992,,,,,,,,,,,,,,,,,,,,,,,,
58
+ WIDEN,WIDEN,Code-Instruction_Tuned,qmd_raw_from_base,weight space,0.050647287040768006,0.0466550100292541,-0.003992277011513903,-0.19618626156255584,,,,,,,,,,,,,,,,,,,,,,,,
59
+ WIDEN,WIDEN,Code-Math,qmd_raw_from_base,weight space,0.019018005658702125,0.017786045338737237,-0.0012319603199648886,-0.06054031042191369,,,,,,,,,,,,,,,,,,,,,,,,
60
+ WIDEN,WIDEN,Code-Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.05039161050891783,0.046407742875562194,-0.003983867633355634,-0.195773012567514,,,,,,,,,,,,,,,,,,,,,,,,
61
+ WIDEN,WIDEN,Math-Instruction_Tuned,qmd_raw_from_base,weight space,0.05479621414229463,0.05043220024105492,-0.004364013901239708,-0.21445394951854316,,,,,,,,,,,,,,,,,,,,,,,,
62
+ DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0675294344035755,0.054332467446118,-0.013196966957457507,-0.26534451678728566,,,,,,,,,,,,,,,,,,,,,,,,
63
+ DARETaskArithmetic,DARE Linear,Code-Math,beh_js_to_base,behaviour,0.0701002699547847,0.0596478397893982,-0.010452430165386502,-0.2101615500310129,,,,,,,,,,,,,,,,,,,,,,,,
64
+ DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1649849997656848,0.1531925715371772,-0.011792428228507595,-0.23710419069238398,,,,,,,,,,,,,,,,,,,,,,,,
65
+ DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1558226350073653,0.1452851978753024,-0.010537437132062899,-0.21187074067830441,,,,,,,,,,,,,,,,,,,,,,,,
66
+ DARETies,DARE TIES,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0645529807269808,0.0522127097471974,-0.012340270979783408,-0.2481193785443587,,,,,,,,,,,,,,,,,,,,,,,,
67
+ DARETies,DARE TIES,Code-Math,beh_js_to_base,behaviour,0.0705667963771806,0.0605886174584691,-0.009978178918711493,-0.20062602809704028,,,,,,,,,,,,,,,,,,,,,,,,
68
+ DARETies,DARE TIES,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1579099506890481,0.1434274385153199,-0.014482512173728201,-0.2911923025185968,,,,,,,,,,,,,,,,,,,,,,,,
69
+ DARETies,DARE TIES,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1516165943578546,0.136825735452024,-0.014790858905830595,-0.2973920690934773,,,,,,,,,,,,,,,,,,,,,,,,
70
+ TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0670005273723255,0.0541176966435751,-0.012882830728750391,-0.2590283438302437,,,,,,,,,,,,,,,,,,,,,,,,
71
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_js_to_base,behaviour,0.0707630679739104,0.0605505583717166,-0.010212509602193796,-0.20533759266922477,,,,,,,,,,,,,,,,,,,,,,,,
72
+ TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1723386903043327,0.1557380097510118,-0.016600680553320896,-0.3337812069971008,,,,,,,,,,,,,,,,,,,,,,,,
73
+ TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1630553862854352,0.1491063061455251,-0.013949080139910086,-0.2804668634303113,,,,,,,,,,,,,,,,,,,,,,,,
74
+ Ties,TIES,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0401827128762879,0.0320304306451685,-0.008152282231119398,-0.1639136777642354,,,,,,,,,,,,,,,,,,,,,,,,
75
+ Ties,TIES,Code-Math,beh_js_to_base,behaviour,0.0486724480616445,0.0419355507932006,-0.006736897268443905,-0.13545527211695216,,,,,,,,,,,,,,,,,,,,,,,,
76
+ Ties,TIES,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.0290378217286798,0.0234016886107479,-0.005636133117931898,-0.11332278269298476,,,,,,,,,,,,,,,,,,,,,,,,
77
+ Ties,TIES,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.0397027983565525,0.0319435968527206,-0.0077592015038319,-0.15601020903716128,,,,,,,,,,,,,,,,,,,,,,,,
78
+ WIDEN,WIDEN,Code-Instruction_Tuned,beh_js_to_base,behaviour,0.0363831441722982,0.0279335037242935,-0.008449640448004704,-0.16989250400715278,,,,,,,,,,,,,,,,,,,,,,,,
79
+ WIDEN,WIDEN,Code-Math,beh_js_to_base,behaviour,0.0365018882611834,0.0305083815052392,-0.005993506755944199,-0.1205083076394803,,,,,,,,,,,,,,,,,,,,,,,,
80
+ WIDEN,WIDEN,Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.0814156675451293,0.0615861280156665,-0.0198295395294628,-0.39870218676166064,,,,,,,,,,,,,,,,,,,,,,,,
81
+ WIDEN,WIDEN,Math-Instruction_Tuned,beh_js_to_base,behaviour,0.1038943280401073,0.0794474548727698,-0.0244468731673375,-0.4915405008179976,,,,,,,,,,,,,,,,,,,,,,,,
82
+ DARETaskArithmetic,DARE Linear,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.3067125690407576,0.2416893078428814,-0.06502326119787621,-0.2990345698933688,,,,,,,,,,,,,,,,,,,,,,,,
83
+ DARETaskArithmetic,DARE Linear,Code-Math,beh_kl_to_base,behaviour,0.3002615200469781,0.2594705143337791,-0.04079100571319899,-0.18759318779542877,,,,,,,,,,,,,,,,,,,,,,,,
84
+ DARETaskArithmetic,DARE Linear,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.7147236138251114,0.64177468547017,-0.07294892835494138,-0.33548381014630757,,,,,,,,,,,,,,,,,,,,,,,,
85
+ DARETaskArithmetic,DARE Linear,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.6873062086352962,0.611457267630225,-0.07584894100507122,-0.3488206379144176,,,,,,,,,,,,,,,,,,,,,,,,
86
+ DARETies,DARE TIES,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.2927892901591197,0.2313560658544693,-0.0614332243046504,-0.2825243992484339,,,,,,,,,,,,,,,,,,,,,,,,
87
+ DARETies,DARE TIES,Code-Math,beh_kl_to_base,behaviour,0.3048238893372912,0.2638059280120493,-0.041017961325241936,-0.18863693079726954,,,,,,,,,,,,,,,,,,,,,,,,
88
+ DARETies,DARE TIES,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.7098179752433407,0.6148826686629932,-0.09493530658034743,-0.436596658610481,,,,,,,,,,,,,,,,,,,,,,,,
89
+ DARETies,DARE TIES,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.6873289485144918,0.5846292800262997,-0.10269966848819212,-0.4723040744003957,,,,,,,,,,,,,,,,,,,,,,,,
90
+ TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.304607594224647,0.2406390341662837,-0.0639685600583633,-0.29418411951899426,,,,,,,,,,,,,,,,,,,,,,,,
91
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_kl_to_base,behaviour,0.303996269463866,0.2630344646865175,-0.04096180477734851,-0.18837867323164387,,,,,,,,,,,,,,,,,,,,,,,,
92
+ TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.7553384977087472,0.6549430652966949,-0.10039543241205229,-0.46170715521689953,,,,,,,,,,,,,,,,,,,,,,,,
93
+ TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.7302715067695107,0.6341559539274912,-0.09611555284201945,-0.44202447669781614,,,,,,,,,,,,,,,,,,,,,,,,
94
+ Ties,TIES,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.1751193129121662,0.1364185276023318,-0.038700785309834396,-0.1779805023073793,,,,,,,,,,,,,,,,,,,,,,,,
95
+ Ties,TIES,Code-Math,beh_kl_to_base,behaviour,0.2104004919723689,0.1810295336203008,-0.0293709583520681,-0.13507369111245993,,,,,,,,,,,,,,,,,,,,,,,,
96
+ Ties,TIES,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.1210440478162356,0.0967837872615697,-0.02426026055466589,-0.11157017422409021,,,,,,,,,,,,,,,,,,,,,,,,
97
+ Ties,TIES,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.1695624375570265,0.1348951166119359,-0.03466732094509059,-0.15943105924236753,,,,,,,,,,,,,,,,,,,,,,,,
98
+ WIDEN,WIDEN,Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.1487084594954175,0.1135898635347647,-0.035118595960652796,-0.16150642162340226,,,,,,,,,,,,,,,,,,,,,,,,
99
+ WIDEN,WIDEN,Code-Math,beh_kl_to_base,behaviour,0.1526036240678989,0.127567749296493,-0.025035874771405897,-0.11513713563810195,,,,,,,,,,,,,,,,,,,,,,,,
100
+ WIDEN,WIDEN,Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.3255842181369209,0.2459603207798351,-0.0796238973570858,-0.3661812320817214,,,,,,,,,,,,,,,,,,,,,,,,
101
+ WIDEN,WIDEN,Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.4269925540304505,0.3248536287939473,-0.1021389252365032,-0.4697252800735605,,,,,,,,,,,,,,,,,,,,,,,,
102
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_cka_mean,representation,0.9930885569927596,0.9945506464794936,0.0014620894867340972,1.2173321397865435,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
103
+ Ties,TIES,Code-Math,geo_cka_mean,representation,0.9950276483976797,0.995968085759268,0.000940437361588331,0.78300585299656,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
104
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_cka_late_minus_early,representation,-0.0769884921105492,-0.055673375970133,0.0213151161404162,1.0632110594726967,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
105
+ Ties,TIES,Code-Math,geo_cka_late_minus_early,representation,-0.039826488481819,-0.0314126548347336,0.008413833647085402,0.4196871800844909,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
106
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_cka_peak_depth,representation,0.1794871794871795,0.1794871794871795,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
107
+ Ties,TIES,Code-Math,geo_cka_peak_depth,representation,0.1794871794871795,0.1794871794871795,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
108
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_rowwise_cosine_mean,representation,0.9908710673451424,0.9926170140504836,0.001745946705341228,1.1351066300904185,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
109
+ Ties,TIES,Code-Math,geo_rowwise_cosine_mean,representation,0.9932829976081848,0.9945648342370988,0.0012818366289140126,0.8333709452423851,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
110
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_rowwise_cosine_late_minus_early,representation,-0.0816612243652343,-0.0660529732704162,0.015608251094818101,1.0706216037880014,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
111
+ Ties,TIES,Code-Math,geo_rowwise_cosine_late_minus_early,representation,-0.0572614073753356,-0.047279417514801,0.009981989860534606,0.6846977235668102,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
112
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_rowwise_cosine_peak_depth,representation,0.0769230769230769,0.0769230769230769,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
113
+ Ties,TIES,Code-Math,geo_rowwise_cosine_peak_depth,representation,0.0769230769230769,0.0769230769230769,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
114
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_procrustes_mean,representation,0.0027219471897545,0.0021906655665475,-0.0005312816232070001,-1.1530871863426981,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
115
+ Ties,TIES,Code-Math,geo_procrustes_mean,representation,0.0019867146024871,0.0016178933505263,-0.00036882125196079974,-0.8004851685245838,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
116
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_procrustes_late_minus_early,representation,0.0140495389049124,0.0104719520430905,-0.003577586861821901,-1.0284580258562392,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
117
+ Ties,TIES,Code-Math,geo_procrustes_late_minus_early,representation,0.0076813466885877,0.0061267972029742,-0.0015545494856135005,-0.44689030813795616,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
118
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_procrustes_peak_depth,representation,1.0,1.0,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
119
+ Ties,TIES,Code-Math,geo_procrustes_peak_depth,representation,1.0,1.0,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
120
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_subspace_overlap_mean,representation,0.7774382523120954,0.7996645381073592,0.022226285795263756,0.664232633423485,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
121
+ Ties,TIES,Code-Math,geo_subspace_overlap_mean,representation,0.8307646546247305,0.8532723782550617,0.022507723630331267,0.6726434041682638,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
122
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_subspace_overlap_late_minus_early,representation,-0.2073746343252057,-0.1796519648387671,0.02772266948643859,0.7734565102552258,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
123
+ Ties,TIES,Code-Math,geo_subspace_overlap_late_minus_early,representation,-0.1515655741559813,-0.1241927094204411,0.027372864735540195,0.763697033014662,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
124
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_subspace_overlap_peak_depth,representation,0.0,0.0,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
125
+ Ties,TIES,Code-Math,geo_subspace_overlap_peak_depth,representation,0.0,0.0,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
126
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_svcca_mean,representation,0.9891374250741723,0.990373600650402,0.001236175576229681,0.8413674529886523,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
127
+ Ties,TIES,Code-Math,geo_svcca_mean,representation,0.9918700300931974,0.9923686480991026,0.000498618005905227,0.33937004557418526,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
128
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_svcca_late_minus_early,representation,0.0066625898499054,0.0056903944721771,-0.0009721953777283004,-1.4932141936467185,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
129
+ Ties,TIES,Code-Math,geo_svcca_late_minus_early,representation,0.0053088284958217,0.0052577493998841,-5.107909593760063e-05,-0.07845339815427867,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
130
+ TaskArithmetic,Task Arithmetic,Code-Math,geo_svcca_peak_depth,representation,0.5641025641025641,0.5641025641025641,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
131
+ Ties,TIES,Code-Math,geo_svcca_peak_depth,representation,0.5641025641025641,0.5641025641025641,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
132
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_identity_p_at_1,retrieval,1.0,1.0,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
133
+ Ties,TIES,Code-Math,ret_identity_p_at_1,retrieval,1.0,1.0,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
134
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_identity_p_at_5,retrieval,1.0,1.0,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
135
+ Ties,TIES,Code-Math,ret_identity_p_at_5,retrieval,1.0,1.0,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
136
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_identity_p_at_10,retrieval,1.0,1.0,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
137
+ Ties,TIES,Code-Math,ret_identity_p_at_10,retrieval,1.0,1.0,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
138
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_procrustes_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
139
+ Ties,TIES,Code-Math,ret_procrustes_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
140
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_procrustes_p_at_5,retrieval,0.6153846153846154,0.5384615384615384,-0.07692307692307698,-1.7320508075688774,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
141
+ Ties,TIES,Code-Math,ret_procrustes_p_at_5,retrieval,0.6153846153846154,0.5384615384615384,-0.07692307692307698,-1.7320508075688774,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
142
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_procrustes_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
143
+ Ties,TIES,Code-Math,ret_procrustes_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
144
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_ridge_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
145
+ Ties,TIES,Code-Math,ret_ridge_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
146
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_ridge_p_at_5,retrieval,0.5384615384615384,0.5384615384615384,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
147
+ Ties,TIES,Code-Math,ret_ridge_p_at_5,retrieval,0.5384615384615384,0.5384615384615384,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
148
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_ridge_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
149
+ Ties,TIES,Code-Math,ret_ridge_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
150
+ TaskArithmetic,Task Arithmetic,Code-Math,ret_gain_over_identity,retrieval,-0.8461538461538461,-0.8461538461538461,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
151
+ Ties,TIES,Code-Math,ret_gain_over_identity,retrieval,-0.8461538461538461,-0.8461538461538461,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
152
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_logit_kl,behaviour,0.3039962523567662,0.2630344502284437,-0.04096180212832251,-0.7487239953367144,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
153
+ Ties,TIES,Code-Math,beh_logit_kl,behaviour,0.2104004760078036,0.1810295204287913,-0.029370955579012287,-0.5368596610833689,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
154
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_kl_sym,behaviour,0.3235272971571356,0.2737488723688236,-0.04977842478831196,-0.8032850337920455,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
155
+ Ties,TIES,Code-Math,beh_kl_sym,behaviour,0.2156694802772487,0.1838637848608179,-0.0318056954164308,-0.5132552752727136,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
156
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_js,behaviour,0.0707630679714202,0.0605505583697879,-0.010212509601632294,-0.7998420285240777,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
157
+ Ties,TIES,Code-Math,beh_js,behaviour,0.0486724480599814,0.04193555079188,-0.006736897268101401,-0.5276326571105784,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
158
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_logit_cosine,behaviour,0.9613301176577806,0.968836484476924,0.007506366819143406,1.0193235056671888,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
159
+ Ties,TIES,Code-Math,beh_logit_cosine,behaviour,0.973597852513194,0.9786411840468644,0.005043331533670425,0.684856802632177,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
160
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_logit_l2,behaviour,162.13931274414062,145.14276123046875,-16.996551513671875,-0.9610635668823926,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
161
+ Ties,TIES,Code-Math,beh_logit_l2,behaviour,134.2702178955078,120.259521484375,-14.010696411132812,-0.7922295211801272,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
162
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_entropy_gap,behaviour,-0.9580553334901714,-0.8577658281539922,0.1002895053361792,0.7250644648580261,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
163
+ Ties,TIES,Code-Math,beh_entropy_gap,behaviour,-0.7119038706301577,-0.6537038415514891,0.058200029078668614,0.42076957900223394,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
164
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_topk_overlap,behaviour,0.75,0.759375,0.009375000000000022,0.40730653998127936,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
165
+ Ties,TIES,Code-Math,beh_topk_overlap,behaviour,0.790625,0.796875,0.006249999999999978,0.2715376933208513,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
166
+ TaskArithmetic,Task Arithmetic,Code-Math,beh_rank_corr,behaviour,0.9612488259798522,0.9668371863411456,0.0055883603612933985,0.9043130842475362,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
167
+ Ties,TIES,Code-Math,beh_rank_corr,behaviour,0.9714328588858698,0.9756381849494506,0.0042053260635808165,0.6805093331424227,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
168
+ TaskArithmetic,Task Arithmetic,Code-Math,grad_cosine,gradient,0.592158773428496,0.5782269179049978,-0.013931855523498182,-1.9955597222160073,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
169
+ Ties,TIES,Code-Math,grad_cosine,gradient,0.5770794629628759,0.5798330660603503,0.002753603097474411,0.3944183474355629,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
170
+ TaskArithmetic,Task Arithmetic,Code-Math,grad_l2,gradient,2306.759099282534,2370.258570451704,63.49947116917019,1.2804236528840072,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
171
+ Ties,TIES,Code-Math,grad_l2,gradient,2248.9716232321766,2303.823981372773,54.85235814059615,1.106060499190166,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
172
+ TaskArithmetic,Task Arithmetic,Code-Math,grad_norm_ratio,gradient,0.9110638542598302,0.93503639362456,0.023972539364729872,0.3872846452052494,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
173
+ Ties,TIES,Code-Math,grad_norm_ratio,gradient,0.7936686956273205,0.8719713407752943,0.0783026451479738,1.2650062508347577,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
174
+ TaskArithmetic,Task Arithmetic,Code-Math,grad_norm_a,gradient,2425.0298323354004,2488.8388868209768,63.809054485576326,0.387284645205249,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
175
+ Ties,TIES,Code-Math,grad_norm_a,gradient,2112.552544904357,2320.975093496056,208.42254859169907,1.2650062508347617,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
176
+ TaskArithmetic,Task Arithmetic,Code-Math,grad_norm_b,gradient,2661.756166702006,2661.756166702006,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
177
+ Ties,TIES,Code-Math,grad_norm_b,gradient,2661.756166702006,2661.756166702006,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
178
+ TaskArithmetic,Task Arithmetic,Code-Math,grad_cosine_layer_mean,gradient,0.5155714649707079,0.5337781719863415,0.01820670701563354,0.522265593839657,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
179
+ Ties,TIES,Code-Math,grad_cosine_layer_mean,gradient,0.5680897474288941,0.5935959197580815,0.025506172329187393,0.7316532433152987,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
180
+ TaskArithmetic,Task Arithmetic,Code-Math,grad_cosine_layer_late_minus_early,gradient,0.2209797352552414,0.3219577968120575,0.1009780615568161,1.8692998590654468,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
181
+ Ties,TIES,Code-Math,grad_cosine_layer_late_minus_early,gradient,0.2794316709041595,0.3435364365577698,0.06410476565361029,1.1867036022898234,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
182
+ TaskArithmetic,Task Arithmetic,Code-Math,grad_cosine_layer_min,gradient,0.2175084501504898,0.1312640011310577,-0.0862444490194321,-2.1313491949485486,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
183
+ Ties,TIES,Code-Math,grad_cosine_layer_min,gradient,0.1861082315444946,0.1399699449539184,-0.0461382865905762,-1.1402101943856504,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
184
+ TaskArithmetic,Task Arithmetic,Code-Math,weight_cosine,weight space,0.999807596206665,0.999807596206665,0.0,,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
185
+ Ties,TIES,Code-Math,weight_cosine,weight space,0.999807596206665,0.999807596206665,0.0,,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
186
+ TaskArithmetic,Task Arithmetic,Code-Math,subspace_overlap,weight space,0.99853515625,0.999355137348175,0.0008199810981750488,2.449081454332485,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
187
+ Ties,TIES,Code-Math,subspace_overlap,weight space,0.9989471435546876,0.998933732509613,-1.3411045074573913e-05,-0.040055486459938276,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
188
+ TaskArithmetic,Task Arithmetic,Code-Math,spectral_overcounting,weight space,-0.9999980330467224,-0.9999980926513672,-5.960464477539063e-08,-2.0,64.7,64.5,0.24,0.24,18.29,17.68,21.92,21.54,15.0,14.6,28.6,29.2,52.1,52.52,0.3547705078172823,0.40357481522062155,0.04880430740333924,0.0,-0.6099999999999994,0.5999999999999979,0.4200000000000017,-0.40000000000000036,-0.20000000000000284,-0.38000000000000256
189
+ Ties,TIES,Code-Math,spectral_overcounting,weight space,-0.9999980926513672,-0.9999980926513672,0.0,0.0,62.6,63.8,0.2,0.23,15.85,15.85,21.63,22.64,14.3,15.3,26.8,28.6,51.86,52.29,0.21736985324982597,0.3515020017556185,0.13413214850579255,0.03,0.0,1.8000000000000007,0.4299999999999997,1.0,1.1999999999999957,1.0100000000000016
tables/T2_per_operator.csv CHANGED
@@ -1,31 +1,163 @@
1
  operator,metric,family,mean_without,mean_with,mean_diff_raw,n_pairs,mean_diff,median_diff,n_positive,p_wilcoxon,p_sign,q_bh
2
- ALL,beh_js_to_base,behaviour,0.08960160711301804,0.07769059418763206,-0.011911012925385985,20,-0.23948851121034828,-0.2244874656853442,0,1.9073486328125e-06,1.9073486328125e-06,1.1444091796875e-05
3
- ALL,beh_kl_to_base,behaviour,0.3913996514478821,0.3301468381705516,-0.061252813277330495,20,-0.28169470948872694,-0.28835425938371406,0,1.9073486328125e-06,1.9073486328125e-06,1.1444091796875e-05
4
- ALL,act_deviation_from_base,other,0.2909211340174079,0.24877564283087844,-0.0421454911865294,20,-0.5111158417968739,-0.48141035398846777,0,1.9073486328125e-06,1.9073486328125e-06,1.1444091796875e-05
5
- ALL,cka_to_base,other,0.9741214978694913,0.9828484863042831,0.008726988434791572,20,0.5844676284923385,0.3942216092375521,20,1.9073486328125e-06,1.9073486328125e-06,1.1444091796875e-05
6
- ALL,qmd_raw_from_base,other,0.05631466960422178,0.0517850840132395,-0.0045295855909822675,20,-0.22259038161919803,-0.23759881154028703,0,1.9073486328125e-06,1.9073486328125e-06,1.1444091796875e-05
7
- DARETaskArithmetic,beh_js_to_base,behaviour,0.11460933478285257,0.10311451916199893,-0.011494815620853626,4,-0.23112024954724675,-0.2244874656853442,0,,0.125,0.125
8
- DARETaskArithmetic,beh_kl_to_base,behaviour,0.5022509778870359,0.43859794381926387,-0.06365303406777195,4,-0.2927330514373807,-0.31725919001983816,0,,0.125,0.125
9
- DARETaskArithmetic,act_deviation_from_base,other,0.3368888414464891,0.2870386422378942,-0.049850199208594875,4,-0.6045540297413006,-0.6458657290521284,0,,0.125,0.125
10
- DARETaskArithmetic,cka_to_base,other,0.9630701072514057,0.9755296085029841,0.012459501251578353,4,0.8344430845898384,0.8572535910386834,4,,0.125,0.125
11
- DARETaskArithmetic,qmd_raw_from_base,other,0.06632904456130037,0.06089247289720014,-0.005436571664100237,4,-0.2671609879326216,-0.31748988518449245,0,,0.125,0.125
12
- DARETies,beh_js_to_base,behaviour,0.11116158053776601,0.09826362529325261,-0.012897955244513424,4,-0.25933244456336824,-0.26965584053147773,0,,0.125,0.125
13
- DARETies,beh_kl_to_base,behaviour,0.4986900258135608,0.4236684856389529,-0.07502154017460796,4,-0.345015515764145,-0.35956052892945745,0,,0.125,0.125
14
- DARETies,act_deviation_from_base,other,0.33776009783614425,0.2851606332696974,-0.05259946456644687,4,-0.6378955103633283,-0.6735627974751665,0,,0.125,0.125
15
- DARETies,cka_to_base,other,0.9640446797013282,0.9774078104645014,0.013363130763173137,4,0.8949613494670974,0.8944943757651928,4,,0.125,0.125
16
- DARETies,qmd_raw_from_base,other,0.06239976899221724,0.057296373632227476,-0.005103395359989763,4,-0.25078822287745417,-0.2952427263264752,0,,0.125,0.125
17
- TaskArithmetic,beh_js_to_base,behaviour,0.11828941798400094,0.10487814272795715,-0.013411275256043792,4,-0.26965350173172015,-0.2697476036302775,0,,0.125,0.125
18
- TaskArithmetic,beh_kl_to_base,behaviour,0.5235534670416927,0.44819312951924684,-0.07536033752244589,4,-0.3465736061663384,-0.36810429810840517,0,,0.125,0.125
19
- TaskArithmetic,act_deviation_from_base,other,0.33116828347556293,0.28389220156241207,-0.04727608191315084,4,-0.5733366422745889,-0.622649238772316,0,,0.125,0.125
20
- TaskArithmetic,cka_to_base,other,0.9664460539817811,0.9768369819968938,0.01039092801511285,4,0.695905706786084,0.7575500526611205,4,,0.125,0.125
21
- TaskArithmetic,qmd_raw_from_base,other,0.05946221190041548,0.05463345322278232,-0.004828758677633161,4,-0.23729217943054587,-0.2824773273777593,0,,0.125,0.125
22
- Ties,beh_js_to_base,behaviour,0.03939894525579118,0.0323278167254594,-0.007071128530331775,4,-0.14217548540283342,-0.1457327405770567,0,,0.125,0.125
23
- Ties,beh_kl_to_base,behaviour,0.16903157256444928,0.13728174127403456,-0.03174983129041474,4,-0.14601385672157424,-0.14725237517741374,0,,0.125,0.125
24
- Ties,act_deviation_from_base,other,0.22816263106651602,0.1956928054336458,-0.0324698256328702,4,-0.3937750348641458,-0.40788409954781824,0,,0.125,0.125
25
- Ties,cka_to_base,other,0.9887393102049828,0.9927884653210641,0.0040491551160812544,4,0.27118176055542725,0.2504462520129581,4,,0.125,0.125
26
- Ties,qmd_raw_from_base,other,0.04966904322950514,0.0457828706928355,-0.0038861725366696395,4,-0.19097213433773666,-0.21927098435309408,0,,0.125,0.125
27
- WIDEN,beh_js_to_base,behaviour,0.06454875700467955,0.04986886702949225,-0.0146798899751873,4,-0.2951608748065728,-0.2842973453844067,0,,0.125,0.125
28
- WIDEN,beh_kl_to_base,behaviour,0.26347221393267195,0.20299289060126002,-0.06047932333141193,4,-0.2781375173541965,-0.2638438268525618,0,,0.125,0.125
29
- WIDEN,act_deviation_from_base,other,0.2206258162623271,0.19209393165074284,-0.028531884611584242,4,-0.34601799174100545,-0.36201251102250254,0,,0.125,0.125
30
- WIDEN,cka_to_base,other,0.9883073382079601,0.9916795652359724,0.0033722270280122646,4,0.2258462410632458,0.2548471367259819,4,,0.125,0.125
31
- WIDEN,qmd_raw_from_base,other,0.04371327933767065,0.04032024962115212,-0.0033930297165185334,4,-0.16673838351763168,-0.1959796370650349,0,,0.125,0.125
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  operator,metric,family,mean_without,mean_with,mean_diff_raw,n_pairs,mean_diff,median_diff,n_positive,p_wilcoxon,p_sign,q_bh
2
+ ALL,beh_entropy_gap,behaviour,-0.8349796020601645,-0.7557348348527406,0.0792447672074239,2,0.57291702193013,0.57291702193013,2.0,,0.5,1.0
3
+ ALL,beh_js,behaviour,0.0597177580157008,0.05124305458083395,-0.008474703434866847,2,-0.6637373428173281,-0.6637373428173281,0.0,,0.5,1.0
4
+ ALL,beh_js_to_base,behaviour,0.08960160711301804,0.07769059418763206,-0.011911012925385985,20,-0.23948851121034828,-0.2244874656853442,0.0,1.9073486328125e-06,1.9073486328125e-06,4.31060791015625e-05
5
+ ALL,beh_kl_sym,behaviour,0.2695983887171921,0.22880632861482075,-0.04079206010237138,2,-0.6582701545323796,-0.6582701545323796,0.0,,0.5,1.0
6
+ ALL,beh_kl_to_base,behaviour,0.3913996514478821,0.3301468381705516,-0.061252813277330495,20,-0.28169470948872694,-0.28835425938371406,0.0,1.9073486328125e-06,1.9073486328125e-06,4.31060791015625e-05
7
+ ALL,beh_logit_cosine,behaviour,0.9674639850854874,0.9737388342618942,0.006274849176406916,2,0.8520901541496828,0.8520901541496828,2.0,,0.5,1.0
8
+ ALL,beh_logit_kl,behaviour,0.2571983641822849,0.2220319853286175,-0.0351663788536674,2,-0.6427918282100417,-0.6427918282100417,0.0,,0.5,1.0
9
+ ALL,beh_logit_l2,behaviour,148.20476531982422,132.70114135742188,-15.503623962402344,2,-0.8766465440312599,-0.8766465440312599,0.0,,0.5,1.0
10
+ ALL,beh_rank_corr,behaviour,0.966340842432861,0.9712376856452981,0.0048968432124371075,2,0.7924112086949795,0.7924112086949795,2.0,,0.5,1.0
11
+ ALL,beh_topk_overlap,behaviour,0.7703125,0.778125,0.0078125,2,0.33942211665106536,0.33942211665106536,2.0,,0.5,1.0
12
+ ALL,grad_cosine,gradient,0.5846191181956859,0.579029991982674,-0.005589126213011886,2,-0.8005706873902222,-0.8005706873902222,1.0,,1.0,1.0
13
+ ALL,grad_cosine_layer_late_minus_early,gradient,0.25020570307970047,0.33274711668491364,0.08254141360521319,2,1.528001730677635,1.528001730677635,2.0,,0.5,1.0
14
+ ALL,grad_cosine_layer_mean,gradient,0.5418306061998011,0.5636870458722114,0.021856439672410466,2,0.6269594185774778,0.6269594185774778,2.0,,0.5,1.0
15
+ ALL,grad_cosine_layer_min,gradient,0.20180834084749222,0.13561697304248804,-0.06619136780500415,2,-1.6357796946670995,-1.6357796946670995,0.0,,0.5,1.0
16
+ ALL,grad_l2,gradient,2277.8653612573553,2337.0412759122382,59.17591465488317,2,1.1932420760370865,1.1932420760370865,2.0,,0.5,1.0
17
+ ALL,grad_norm_a,gradient,2268.791188619879,2404.9069901585162,136.1158015386377,2,0.8261454480200053,0.8261454480200053,2.0,,0.5,1.0
18
+ ALL,grad_norm_b,gradient,2661.756166702006,2661.756166702006,0.0,0,,,,,,
19
+ ALL,grad_norm_ratio,gradient,0.8523662749435753,0.9035038671999271,0.05113759225635184,2,0.8261454480200036,0.8261454480200036,2.0,,0.5,1.0
20
+ ALL,act_deviation_from_base,other,0.2909211340174079,0.24877564283087844,-0.0421454911865294,20,-0.5111158417968739,-0.48141035398846777,0.0,1.9073486328125e-06,1.9073486328125e-06,4.31060791015625e-05
21
+ ALL,cka_to_base,other,0.9741214978694913,0.9828484863042831,0.008726988434791572,20,0.5844676284923385,0.3942216092375521,20.0,1.9073486328125e-06,1.9073486328125e-06,4.31060791015625e-05
22
+ ALL,qmd_raw_from_base,other,0.05631466960422178,0.0517850840132395,-0.0045295855909822675,20,-0.22259038161919803,-0.23759881154028703,0.0,1.9073486328125e-06,1.9073486328125e-06,4.31060791015625e-05
23
+ ALL,geo_cka_late_minus_early,representation,-0.0584074902961841,-0.0435430154024333,0.014864474893750802,2,0.7414491197785937,0.7414491197785937,2.0,,0.5,1.0
24
+ ALL,geo_cka_mean,representation,0.9940581026952195,0.9952593661193808,0.0012012634241612141,2,1.0001689963915519,1.0001689963915519,2.0,,0.5,1.0
25
+ ALL,geo_cka_peak_depth,representation,0.1794871794871795,0.1794871794871795,0.0,0,,,,,,
26
+ ALL,geo_procrustes_late_minus_early,representation,0.01086544279675005,0.00829937462303235,-0.0025660681737177007,2,-0.7376741669970976,-0.7376741669970976,0.0,,0.5,1.0
27
+ ALL,geo_procrustes_mean,representation,0.0023543308961208,0.0019042794585369,-0.0004500514375838999,2,-0.976786177433641,-0.976786177433641,0.0,,0.5,1.0
28
+ ALL,geo_procrustes_peak_depth,representation,1.0,1.0,0.0,0,,,,,,
29
+ ALL,geo_rowwise_cosine_late_minus_early,representation,-0.06946131587028495,-0.0566661953926086,0.012795120477676353,2,0.8776596636774059,0.8776596636774059,2.0,,0.5,1.0
30
+ ALL,geo_rowwise_cosine_mean,representation,0.9920770324766637,0.9935909241437912,0.0015138916671276204,2,0.9842387876664018,0.9842387876664018,2.0,,0.5,1.0
31
+ ALL,geo_rowwise_cosine_peak_depth,representation,0.0769230769230769,0.0769230769230769,0.0,0,,,,,,
32
+ ALL,geo_subspace_overlap_late_minus_early,representation,-0.1794701042405935,-0.1519223371296041,0.027547767110989392,2,0.7685767716349439,0.7685767716349439,2.0,,0.5,1.0
33
+ ALL,geo_subspace_overlap_mean,representation,0.804101453468413,0.8264684581812105,0.02236700471279751,2,0.6684380187958745,0.6684380187958745,2.0,,0.5,1.0
34
+ ALL,geo_subspace_overlap_peak_depth,representation,0.0,0.0,0.0,0,,,,,,
35
+ ALL,geo_svcca_late_minus_early,representation,0.00598570917286355,0.0054740719360306,-0.0005116372368329505,2,-0.7858337959004986,-0.7858337959004986,0.0,,0.5,1.0
36
+ ALL,geo_svcca_mean,representation,0.9905037275836849,0.9913711243747523,0.000867396791067454,2,0.5903687492814188,0.5903687492814188,2.0,,0.5,1.0
37
+ ALL,geo_svcca_peak_depth,representation,0.5641025641025641,0.5641025641025641,0.0,0,,,,,,
38
+ ALL,ret_gain_over_identity,retrieval,-0.8461538461538461,-0.8461538461538461,0.0,0,,,,,,
39
+ ALL,ret_identity_p_at_1,retrieval,1.0,1.0,0.0,0,,,,,,
40
+ ALL,ret_identity_p_at_10,retrieval,1.0,1.0,0.0,0,,,,,,
41
+ ALL,ret_identity_p_at_5,retrieval,1.0,1.0,0.0,0,,,,,,
42
+ ALL,ret_procrustes_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
43
+ ALL,ret_procrustes_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
44
+ ALL,ret_procrustes_p_at_5,retrieval,0.6153846153846154,0.5384615384615384,-0.07692307692307698,2,-1.7320508075688774,-1.7320508075688774,0.0,,0.5,1.0
45
+ ALL,ret_ridge_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
46
+ ALL,ret_ridge_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
47
+ ALL,ret_ridge_p_at_5,retrieval,0.5384615384615384,0.5384615384615384,0.0,0,,,,,,
48
+ ALL,spectral_overcounting,weight space,-0.9999980628490448,-0.9999980926513672,-2.9802322387695312e-08,2,-1.0,-1.0,0.0,,1.0,1.0
49
+ ALL,subspace_overlap,weight space,0.9987411499023438,0.999144434928894,0.00040328502655023746,2,1.2045129839362734,1.2045129839362734,1.0,,1.0,1.0
50
+ ALL,weight_cosine,weight space,0.999807596206665,0.999807596206665,0.0,0,,,,,,
51
+ DARETaskArithmetic,beh_js_to_base,behaviour,0.11460933478285257,0.10311451916199893,-0.011494815620853626,4,-0.23112024954724675,-0.2244874656853442,0.0,,0.125,0.4708333333333333
52
+ DARETaskArithmetic,beh_kl_to_base,behaviour,0.5022509778870359,0.43859794381926387,-0.06365303406777195,4,-0.2927330514373807,-0.31725919001983816,0.0,,0.125,0.4708333333333333
53
+ DARETaskArithmetic,act_deviation_from_base,other,0.3368888414464891,0.2870386422378942,-0.049850199208594875,4,-0.6045540297413006,-0.6458657290521284,0.0,,0.125,0.4708333333333333
54
+ DARETaskArithmetic,cka_to_base,other,0.9630701072514057,0.9755296085029841,0.012459501251578353,4,0.8344430845898384,0.8572535910386834,4.0,,0.125,0.4708333333333333
55
+ DARETaskArithmetic,qmd_raw_from_base,other,0.06632904456130037,0.06089247289720014,-0.005436571664100237,4,-0.2671609879326216,-0.31748988518449245,0.0,,0.125,0.4708333333333333
56
+ DARETies,beh_js_to_base,behaviour,0.11116158053776601,0.09826362529325261,-0.012897955244513424,4,-0.25933244456336824,-0.26965584053147773,0.0,,0.125,0.4708333333333333
57
+ DARETies,beh_kl_to_base,behaviour,0.4986900258135608,0.4236684856389529,-0.07502154017460796,4,-0.345015515764145,-0.35956052892945745,0.0,,0.125,0.4708333333333333
58
+ DARETies,act_deviation_from_base,other,0.33776009783614425,0.2851606332696974,-0.05259946456644687,4,-0.6378955103633283,-0.6735627974751665,0.0,,0.125,0.4708333333333333
59
+ DARETies,cka_to_base,other,0.9640446797013282,0.9774078104645014,0.013363130763173137,4,0.8949613494670974,0.8944943757651928,4.0,,0.125,0.4708333333333333
60
+ DARETies,qmd_raw_from_base,other,0.06239976899221724,0.057296373632227476,-0.005103395359989763,4,-0.25078822287745417,-0.2952427263264752,0.0,,0.125,0.4708333333333333
61
+ TaskArithmetic,beh_entropy_gap,behaviour,-0.9580553334901714,-0.8577658281539922,0.1002895053361792,1,0.7250644648580261,0.7250644648580261,1.0,,1.0,1.0
62
+ TaskArithmetic,beh_js,behaviour,0.0707630679714202,0.0605505583697879,-0.010212509601632294,1,-0.7998420285240777,-0.7998420285240777,0.0,,1.0,1.0
63
+ TaskArithmetic,beh_js_to_base,behaviour,0.11828941798400094,0.10487814272795715,-0.013411275256043792,4,-0.26965350173172015,-0.2697476036302775,0.0,,0.125,0.4708333333333333
64
+ TaskArithmetic,beh_kl_sym,behaviour,0.3235272971571356,0.2737488723688236,-0.04977842478831196,1,-0.8032850337920455,-0.8032850337920455,0.0,,1.0,1.0
65
+ TaskArithmetic,beh_kl_to_base,behaviour,0.5235534670416927,0.44819312951924684,-0.07536033752244589,4,-0.3465736061663384,-0.36810429810840517,0.0,,0.125,0.4708333333333333
66
+ TaskArithmetic,beh_logit_cosine,behaviour,0.9613301176577806,0.968836484476924,0.007506366819143406,1,1.0193235056671888,1.0193235056671888,1.0,,1.0,1.0
67
+ TaskArithmetic,beh_logit_kl,behaviour,0.3039962523567662,0.2630344502284437,-0.04096180212832251,1,-0.7487239953367144,-0.7487239953367144,0.0,,1.0,1.0
68
+ TaskArithmetic,beh_logit_l2,behaviour,162.13931274414062,145.14276123046875,-16.996551513671875,1,-0.9610635668823926,-0.9610635668823926,0.0,,1.0,1.0
69
+ TaskArithmetic,beh_rank_corr,behaviour,0.9612488259798522,0.9668371863411456,0.0055883603612933985,1,0.9043130842475362,0.9043130842475362,1.0,,1.0,1.0
70
+ TaskArithmetic,beh_topk_overlap,behaviour,0.75,0.759375,0.009375000000000022,1,0.40730653998127936,0.40730653998127936,1.0,,1.0,1.0
71
+ TaskArithmetic,grad_cosine,gradient,0.592158773428496,0.5782269179049978,-0.013931855523498182,1,-1.9955597222160073,-1.9955597222160073,0.0,,1.0,1.0
72
+ TaskArithmetic,grad_cosine_layer_late_minus_early,gradient,0.2209797352552414,0.3219577968120575,0.1009780615568161,1,1.8692998590654468,1.8692998590654468,1.0,,1.0,1.0
73
+ TaskArithmetic,grad_cosine_layer_mean,gradient,0.5155714649707079,0.5337781719863415,0.01820670701563354,1,0.522265593839657,0.522265593839657,1.0,,1.0,1.0
74
+ TaskArithmetic,grad_cosine_layer_min,gradient,0.2175084501504898,0.1312640011310577,-0.0862444490194321,1,-2.1313491949485486,-2.1313491949485486,0.0,,1.0,1.0
75
+ TaskArithmetic,grad_l2,gradient,2306.759099282534,2370.258570451704,63.49947116917019,1,1.2804236528840072,1.2804236528840072,1.0,,1.0,1.0
76
+ TaskArithmetic,grad_norm_a,gradient,2425.0298323354004,2488.8388868209768,63.809054485576326,1,0.387284645205249,0.387284645205249,1.0,,1.0,1.0
77
+ TaskArithmetic,grad_norm_b,gradient,2661.756166702006,2661.756166702006,0.0,0,,,,,,
78
+ TaskArithmetic,grad_norm_ratio,gradient,0.9110638542598302,0.93503639362456,0.023972539364729872,1,0.3872846452052494,0.3872846452052494,1.0,,1.0,1.0
79
+ TaskArithmetic,act_deviation_from_base,other,0.33116828347556293,0.28389220156241207,-0.04727608191315084,4,-0.5733366422745889,-0.622649238772316,0.0,,0.125,0.4708333333333333
80
+ TaskArithmetic,cka_to_base,other,0.9664460539817811,0.9768369819968938,0.01039092801511285,4,0.695905706786084,0.7575500526611205,4.0,,0.125,0.4708333333333333
81
+ TaskArithmetic,qmd_raw_from_base,other,0.05946221190041548,0.05463345322278232,-0.004828758677633161,4,-0.23729217943054587,-0.2824773273777593,0.0,,0.125,0.4708333333333333
82
+ TaskArithmetic,geo_cka_late_minus_early,representation,-0.0769884921105492,-0.055673375970133,0.0213151161404162,1,1.0632110594726967,1.0632110594726967,1.0,,1.0,1.0
83
+ TaskArithmetic,geo_cka_mean,representation,0.9930885569927596,0.9945506464794936,0.0014620894867340972,1,1.2173321397865435,1.2173321397865435,1.0,,1.0,1.0
84
+ TaskArithmetic,geo_cka_peak_depth,representation,0.1794871794871795,0.1794871794871795,0.0,0,,,,,,
85
+ TaskArithmetic,geo_procrustes_late_minus_early,representation,0.0140495389049124,0.0104719520430905,-0.003577586861821901,1,-1.0284580258562392,-1.0284580258562392,0.0,,1.0,1.0
86
+ TaskArithmetic,geo_procrustes_mean,representation,0.0027219471897545,0.0021906655665475,-0.0005312816232070001,1,-1.1530871863426981,-1.1530871863426981,0.0,,1.0,1.0
87
+ TaskArithmetic,geo_procrustes_peak_depth,representation,1.0,1.0,0.0,0,,,,,,
88
+ TaskArithmetic,geo_rowwise_cosine_late_minus_early,representation,-0.0816612243652343,-0.0660529732704162,0.015608251094818101,1,1.0706216037880014,1.0706216037880014,1.0,,1.0,1.0
89
+ TaskArithmetic,geo_rowwise_cosine_mean,representation,0.9908710673451424,0.9926170140504836,0.001745946705341228,1,1.1351066300904185,1.1351066300904185,1.0,,1.0,1.0
90
+ TaskArithmetic,geo_rowwise_cosine_peak_depth,representation,0.0769230769230769,0.0769230769230769,0.0,0,,,,,,
91
+ TaskArithmetic,geo_subspace_overlap_late_minus_early,representation,-0.2073746343252057,-0.1796519648387671,0.02772266948643859,1,0.7734565102552258,0.7734565102552258,1.0,,1.0,1.0
92
+ TaskArithmetic,geo_subspace_overlap_mean,representation,0.7774382523120954,0.7996645381073592,0.022226285795263756,1,0.664232633423485,0.664232633423485,1.0,,1.0,1.0
93
+ TaskArithmetic,geo_subspace_overlap_peak_depth,representation,0.0,0.0,0.0,0,,,,,,
94
+ TaskArithmetic,geo_svcca_late_minus_early,representation,0.0066625898499054,0.0056903944721771,-0.0009721953777283004,1,-1.4932141936467185,-1.4932141936467185,0.0,,1.0,1.0
95
+ TaskArithmetic,geo_svcca_mean,representation,0.9891374250741723,0.990373600650402,0.001236175576229681,1,0.8413674529886523,0.8413674529886523,1.0,,1.0,1.0
96
+ TaskArithmetic,geo_svcca_peak_depth,representation,0.5641025641025641,0.5641025641025641,0.0,0,,,,,,
97
+ TaskArithmetic,ret_gain_over_identity,retrieval,-0.8461538461538461,-0.8461538461538461,0.0,0,,,,,,
98
+ TaskArithmetic,ret_identity_p_at_1,retrieval,1.0,1.0,0.0,0,,,,,,
99
+ TaskArithmetic,ret_identity_p_at_10,retrieval,1.0,1.0,0.0,0,,,,,,
100
+ TaskArithmetic,ret_identity_p_at_5,retrieval,1.0,1.0,0.0,0,,,,,,
101
+ TaskArithmetic,ret_procrustes_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
102
+ TaskArithmetic,ret_procrustes_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
103
+ TaskArithmetic,ret_procrustes_p_at_5,retrieval,0.6153846153846154,0.5384615384615384,-0.07692307692307698,1,-1.7320508075688774,-1.7320508075688774,0.0,,1.0,1.0
104
+ TaskArithmetic,ret_ridge_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
105
+ TaskArithmetic,ret_ridge_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
106
+ TaskArithmetic,ret_ridge_p_at_5,retrieval,0.5384615384615384,0.5384615384615384,0.0,0,,,,,,
107
+ TaskArithmetic,spectral_overcounting,weight space,-0.9999980330467224,-0.9999980926513672,-5.960464477539063e-08,1,-2.0,-2.0,0.0,,1.0,1.0
108
+ TaskArithmetic,subspace_overlap,weight space,0.99853515625,0.999355137348175,0.0008199810981750488,1,2.449081454332485,2.449081454332485,1.0,,1.0,1.0
109
+ TaskArithmetic,weight_cosine,weight space,0.999807596206665,0.999807596206665,0.0,0,,,,,,
110
+ Ties,beh_entropy_gap,behaviour,-0.7119038706301577,-0.6537038415514891,0.058200029078668614,1,0.42076957900223394,0.42076957900223394,1.0,,1.0,1.0
111
+ Ties,beh_js,behaviour,0.0486724480599814,0.04193555079188,-0.006736897268101401,1,-0.5276326571105784,-0.5276326571105784,0.0,,1.0,1.0
112
+ Ties,beh_js_to_base,behaviour,0.03939894525579118,0.0323278167254594,-0.007071128530331775,4,-0.14217548540283342,-0.1457327405770567,0.0,,0.125,0.4708333333333333
113
+ Ties,beh_kl_sym,behaviour,0.2156694802772487,0.1838637848608179,-0.0318056954164308,1,-0.5132552752727136,-0.5132552752727136,0.0,,1.0,1.0
114
+ Ties,beh_kl_to_base,behaviour,0.16903157256444928,0.13728174127403456,-0.03174983129041474,4,-0.14601385672157424,-0.14725237517741374,0.0,,0.125,0.4708333333333333
115
+ Ties,beh_logit_cosine,behaviour,0.973597852513194,0.9786411840468644,0.005043331533670425,1,0.684856802632177,0.684856802632177,1.0,,1.0,1.0
116
+ Ties,beh_logit_kl,behaviour,0.2104004760078036,0.1810295204287913,-0.029370955579012287,1,-0.5368596610833689,-0.5368596610833689,0.0,,1.0,1.0
117
+ Ties,beh_logit_l2,behaviour,134.2702178955078,120.259521484375,-14.010696411132812,1,-0.7922295211801272,-0.7922295211801272,0.0,,1.0,1.0
118
+ Ties,beh_rank_corr,behaviour,0.9714328588858698,0.9756381849494506,0.0042053260635808165,1,0.6805093331424227,0.6805093331424227,1.0,,1.0,1.0
119
+ Ties,beh_topk_overlap,behaviour,0.790625,0.796875,0.006249999999999978,1,0.2715376933208513,0.2715376933208513,1.0,,1.0,1.0
120
+ Ties,grad_cosine,gradient,0.5770794629628759,0.5798330660603503,0.002753603097474411,1,0.3944183474355629,0.3944183474355629,1.0,,1.0,1.0
121
+ Ties,grad_cosine_layer_late_minus_early,gradient,0.2794316709041595,0.3435364365577698,0.06410476565361029,1,1.1867036022898234,1.1867036022898234,1.0,,1.0,1.0
122
+ Ties,grad_cosine_layer_mean,gradient,0.5680897474288941,0.5935959197580815,0.025506172329187393,1,0.7316532433152987,0.7316532433152987,1.0,,1.0,1.0
123
+ Ties,grad_cosine_layer_min,gradient,0.1861082315444946,0.1399699449539184,-0.0461382865905762,1,-1.1402101943856504,-1.1402101943856504,0.0,,1.0,1.0
124
+ Ties,grad_l2,gradient,2248.9716232321766,2303.823981372773,54.85235814059615,1,1.106060499190166,1.106060499190166,1.0,,1.0,1.0
125
+ Ties,grad_norm_a,gradient,2112.552544904357,2320.975093496056,208.42254859169907,1,1.2650062508347617,1.2650062508347617,1.0,,1.0,1.0
126
+ Ties,grad_norm_b,gradient,2661.756166702006,2661.756166702006,0.0,0,,,,,,
127
+ Ties,grad_norm_ratio,gradient,0.7936686956273205,0.8719713407752943,0.0783026451479738,1,1.2650062508347577,1.2650062508347577,1.0,,1.0,1.0
128
+ Ties,act_deviation_from_base,other,0.22816263106651602,0.1956928054336458,-0.0324698256328702,4,-0.3937750348641458,-0.40788409954781824,0.0,,0.125,0.4708333333333333
129
+ Ties,cka_to_base,other,0.9887393102049828,0.9927884653210641,0.0040491551160812544,4,0.27118176055542725,0.2504462520129581,4.0,,0.125,0.4708333333333333
130
+ Ties,qmd_raw_from_base,other,0.04966904322950514,0.0457828706928355,-0.0038861725366696395,4,-0.19097213433773666,-0.21927098435309408,0.0,,0.125,0.4708333333333333
131
+ Ties,geo_cka_late_minus_early,representation,-0.039826488481819,-0.0314126548347336,0.008413833647085402,1,0.4196871800844909,0.4196871800844909,1.0,,1.0,1.0
132
+ Ties,geo_cka_mean,representation,0.9950276483976797,0.995968085759268,0.000940437361588331,1,0.78300585299656,0.78300585299656,1.0,,1.0,1.0
133
+ Ties,geo_cka_peak_depth,representation,0.1794871794871795,0.1794871794871795,0.0,0,,,,,,
134
+ Ties,geo_procrustes_late_minus_early,representation,0.0076813466885877,0.0061267972029742,-0.0015545494856135005,1,-0.44689030813795616,-0.44689030813795616,0.0,,1.0,1.0
135
+ Ties,geo_procrustes_mean,representation,0.0019867146024871,0.0016178933505263,-0.00036882125196079974,1,-0.8004851685245838,-0.8004851685245838,0.0,,1.0,1.0
136
+ Ties,geo_procrustes_peak_depth,representation,1.0,1.0,0.0,0,,,,,,
137
+ Ties,geo_rowwise_cosine_late_minus_early,representation,-0.0572614073753356,-0.047279417514801,0.009981989860534606,1,0.6846977235668102,0.6846977235668102,1.0,,1.0,1.0
138
+ Ties,geo_rowwise_cosine_mean,representation,0.9932829976081848,0.9945648342370988,0.0012818366289140126,1,0.8333709452423851,0.8333709452423851,1.0,,1.0,1.0
139
+ Ties,geo_rowwise_cosine_peak_depth,representation,0.0769230769230769,0.0769230769230769,0.0,0,,,,,,
140
+ Ties,geo_subspace_overlap_late_minus_early,representation,-0.1515655741559813,-0.1241927094204411,0.027372864735540195,1,0.763697033014662,0.763697033014662,1.0,,1.0,1.0
141
+ Ties,geo_subspace_overlap_mean,representation,0.8307646546247305,0.8532723782550617,0.022507723630331267,1,0.6726434041682638,0.6726434041682638,1.0,,1.0,1.0
142
+ Ties,geo_subspace_overlap_peak_depth,representation,0.0,0.0,0.0,0,,,,,,
143
+ Ties,geo_svcca_late_minus_early,representation,0.0053088284958217,0.0052577493998841,-5.107909593760063e-05,1,-0.07845339815427867,-0.07845339815427867,0.0,,1.0,1.0
144
+ Ties,geo_svcca_mean,representation,0.9918700300931974,0.9923686480991026,0.000498618005905227,1,0.33937004557418526,0.33937004557418526,1.0,,1.0,1.0
145
+ Ties,geo_svcca_peak_depth,representation,0.5641025641025641,0.5641025641025641,0.0,0,,,,,,
146
+ Ties,ret_gain_over_identity,retrieval,-0.8461538461538461,-0.8461538461538461,0.0,0,,,,,,
147
+ Ties,ret_identity_p_at_1,retrieval,1.0,1.0,0.0,0,,,,,,
148
+ Ties,ret_identity_p_at_10,retrieval,1.0,1.0,0.0,0,,,,,,
149
+ Ties,ret_identity_p_at_5,retrieval,1.0,1.0,0.0,0,,,,,,
150
+ Ties,ret_procrustes_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
151
+ Ties,ret_procrustes_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
152
+ Ties,ret_procrustes_p_at_5,retrieval,0.6153846153846154,0.5384615384615384,-0.07692307692307698,1,-1.7320508075688774,-1.7320508075688774,0.0,,1.0,1.0
153
+ Ties,ret_ridge_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
154
+ Ties,ret_ridge_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
155
+ Ties,ret_ridge_p_at_5,retrieval,0.5384615384615384,0.5384615384615384,0.0,0,,,,,,
156
+ Ties,spectral_overcounting,weight space,-0.9999980926513672,-0.9999980926513672,0.0,1,0.0,0.0,0.0,,,
157
+ Ties,subspace_overlap,weight space,0.9989471435546876,0.998933732509613,-1.3411045074573913e-05,1,-0.040055486459938276,-0.040055486459938276,0.0,,1.0,1.0
158
+ Ties,weight_cosine,weight space,0.999807596206665,0.999807596206665,0.0,0,,,,,,
159
+ WIDEN,beh_js_to_base,behaviour,0.06454875700467955,0.04986886702949225,-0.0146798899751873,4,-0.2951608748065728,-0.2842973453844067,0.0,,0.125,0.4708333333333333
160
+ WIDEN,beh_kl_to_base,behaviour,0.26347221393267195,0.20299289060126002,-0.06047932333141193,4,-0.2781375173541965,-0.2638438268525618,0.0,,0.125,0.4708333333333333
161
+ WIDEN,act_deviation_from_base,other,0.2206258162623271,0.19209393165074284,-0.028531884611584242,4,-0.34601799174100545,-0.36201251102250254,0.0,,0.125,0.4708333333333333
162
+ WIDEN,cka_to_base,other,0.9883073382079601,0.9916795652359724,0.0033722270280122646,4,0.2258462410632458,0.2548471367259819,4.0,,0.125,0.4708333333333333
163
+ WIDEN,qmd_raw_from_base,other,0.04371327933767065,0.04032024962115212,-0.0033930297165185334,4,-0.16673838351763168,-0.1959796370650349,0.0,,0.125,0.4708333333333333
tables/T3_per_combo.csv CHANGED
@@ -1,26 +1,114 @@
1
  combo,metric,family,mean_without,mean_with,mean_diff_raw,n_pairs,mean_diff,median_diff,n_positive,p_wilcoxon,p_sign,q_bh
2
- ALL,beh_js_to_base,behaviour,0.08960160711301804,0.07769059418763206,-0.011911012925385985,20,-0.23948851121034828,-0.2244874656853442,0,1.9073486328125e-06,1.9073486328125e-06,9.5367431640625e-06
3
- ALL,beh_kl_to_base,behaviour,0.3913996514478821,0.3301468381705516,-0.061252813277330495,20,-0.28169470948872694,-0.28835425938371406,0,1.9073486328125e-06,1.9073486328125e-06,9.5367431640625e-06
4
- ALL,act_deviation_from_base,other,0.2909211340174079,0.24877564283087844,-0.0421454911865294,20,-0.5111158417968739,-0.48141035398846777,0,1.9073486328125e-06,1.9073486328125e-06,9.5367431640625e-06
5
- ALL,cka_to_base,other,0.9741214978694913,0.9828484863042831,0.008726988434791572,20,0.5844676284923385,0.3942216092375521,20,1.9073486328125e-06,1.9073486328125e-06,9.5367431640625e-06
6
- ALL,qmd_raw_from_base,other,0.05631466960422178,0.0517850840132395,-0.0045295855909822675,20,-0.22259038161919803,-0.23759881154028703,0,1.9073486328125e-06,1.9073486328125e-06,9.5367431640625e-06
7
- Code-Instruction_Tuned,beh_js_to_base,behaviour,0.05512975991029358,0.0441253616412705,-0.011004398269023082,5,-0.22125968418665526,-0.2481193785443587,0,0.0625,0.0625,0.0625
8
- Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.24558744516642164,0.19273855980014618,-0.052848885366275425,5,-0.24304600251831573,-0.2825243992484339,0,0.0625,0.0625,0.0625
9
- Code-Instruction_Tuned,act_deviation_from_base,other,0.29014053430408243,0.24678958225995298,-0.04335095204412938,5,-0.5257349653054478,-0.5900637262201152,0,0.0625,0.0625,0.0625
10
- Code-Instruction_Tuned,cka_to_base,other,0.9737666422128678,0.983200253546238,0.009433611333370218,5,0.6317918816246285,0.8040619007888237,5,0.0625,0.0625,0.0625
11
- Code-Instruction_Tuned,qmd_raw_from_base,other,0.06490202158222438,0.05963919167922608,-0.005262829902998324,5,-0.25862306671884455,-0.27103230967800196,0,0.0625,0.0625,0.0625
12
- Code-Math,beh_js_to_base,behaviour,0.05932089412574072,0.05064618958360474,-0.008674704542135979,5,-0.17441775011074206,-0.20062602809704028,0,0.0625,0.0625,0.0625
13
- Code-Math,beh_kl_to_base,behaviour,0.25441715897768064,0.21898163798982795,-0.03543552098785269,5,-0.16296392371498078,-0.18759318779542877,0,0.0625,0.0625,0.0625
14
- Code-Math,act_deviation_from_base,other,0.1923275407031178,0.16943953882902857,-0.02288800187408923,5,-0.27757228627727104,-0.2954531074726018,0,0.0625,0.0625,0.0625
15
- Code-Math,cka_to_base,other,0.9902538159489632,0.9923722693324087,0.0021184533834457397,5,0.14187797249254336,0.15566798010118926,5,0.0625,0.0625,0.0625
16
- Code-Math,qmd_raw_from_base,other,0.024078073164752987,0.02234327448026893,-0.0017347986844840567,5,-0.08525051430324117,-0.08992501734153821,0,0.0625,0.0625,0.0625
17
- Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.12113742600657493,0.10746916728598466,-0.01366825872059028,5,-0.2748205339325454,-0.2911923025185968,0,0.0625,0.0625,0.0625
18
- Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.5253016705460712,0.45086890549425257,-0.07443276505181856,5,-0.3423078060559,-0.3661812320817214,0,0.0625,0.0625,0.0625
19
- Code-Math-Instruction_Tuned,act_deviation_from_base,other,0.34054103039205075,0.2870888278074562,-0.05345220258459451,5,-0.6482370177869512,-0.7525949840811219,0,0.0625,0.0625,0.0625
20
- Code-Math-Instruction_Tuned,cka_to_base,other,0.9637338295578957,0.9776740205287933,0.013940190970897692,5,0.9336084742600528,1.1128547417209058,5,0.0625,0.0625,0.0625
21
- Code-Math-Instruction_Tuned,qmd_raw_from_base,other,0.06741627142434369,0.06192478721447022,-0.005491484209873472,5,-0.2698594697857254,-0.29428904562512653,0,0.0625,0.0625,0.0625
22
- Math-Instruction_Tuned,beh_js_to_base,behaviour,0.122818348409463,0.10852165823966838,-0.014296690169794596,5,-0.2874560766114504,-0.2804668634303113,0,0.0625,0.0625,0.0625
23
- Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.5402923311013552,0.45799824939797984,-0.08229408170337532,5,-0.3784611056657115,-0.44202447669781614,0,0.0625,0.0625,0.0625
24
- Math-Instruction_Tuned,act_deviation_from_base,other,0.3406754306703806,0.2917846224270761,-0.048890808243304495,5,-0.5929190978178251,-0.6552347513245169,0,0.0625,0.0625,0.0625
25
- Math-Instruction_Tuned,cka_to_base,other,0.9687317037582398,0.9781474018096924,0.009415698051452637,5,0.6305921855921296,0.7110382045334173,5,0.0625,0.0625,0.0625
26
- Math-Instruction_Tuned,qmd_raw_from_base,other,0.06886231224556602,0.06323308267899282,-0.005629229566573214,5,-0.27662847566898086,-0.29392234507751674,0,0.0625,0.0625,0.0625
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  combo,metric,family,mean_without,mean_with,mean_diff_raw,n_pairs,mean_diff,median_diff,n_positive,p_wilcoxon,p_sign,q_bh
2
+ ALL,beh_entropy_gap,behaviour,-0.8349796020601645,-0.7557348348527406,0.0792447672074239,2,0.57291702193013,0.57291702193013,2.0,,0.5,0.54
3
+ ALL,beh_js,behaviour,0.0597177580157008,0.05124305458083395,-0.008474703434866847,2,-0.6637373428173281,-0.6637373428173281,0.0,,0.5,0.54
4
+ ALL,beh_js_to_base,behaviour,0.08960160711301804,0.07769059418763206,-0.011911012925385985,20,-0.23948851121034828,-0.2244874656853442,0.0,1.9073486328125e-06,1.9073486328125e-06,3.08990478515625e-05
5
+ ALL,beh_kl_sym,behaviour,0.2695983887171921,0.22880632861482075,-0.04079206010237138,2,-0.6582701545323796,-0.6582701545323796,0.0,,0.5,0.54
6
+ ALL,beh_kl_to_base,behaviour,0.3913996514478821,0.3301468381705516,-0.061252813277330495,20,-0.28169470948872694,-0.28835425938371406,0.0,1.9073486328125e-06,1.9073486328125e-06,3.08990478515625e-05
7
+ ALL,beh_logit_cosine,behaviour,0.9674639850854874,0.9737388342618942,0.006274849176406916,2,0.8520901541496828,0.8520901541496828,2.0,,0.5,0.54
8
+ ALL,beh_logit_kl,behaviour,0.2571983641822849,0.2220319853286175,-0.0351663788536674,2,-0.6427918282100417,-0.6427918282100417,0.0,,0.5,0.54
9
+ ALL,beh_logit_l2,behaviour,148.20476531982422,132.70114135742188,-15.503623962402344,2,-0.8766465440312599,-0.8766465440312599,0.0,,0.5,0.54
10
+ ALL,beh_rank_corr,behaviour,0.966340842432861,0.9712376856452981,0.0048968432124371075,2,0.7924112086949795,0.7924112086949795,2.0,,0.5,0.54
11
+ ALL,beh_topk_overlap,behaviour,0.7703125,0.778125,0.0078125,2,0.33942211665106536,0.33942211665106536,2.0,,0.5,0.54
12
+ ALL,grad_cosine,gradient,0.5846191181956859,0.579029991982674,-0.005589126213011886,2,-0.8005706873902222,-0.8005706873902222,1.0,,1.0,1.0
13
+ ALL,grad_cosine_layer_late_minus_early,gradient,0.25020570307970047,0.33274711668491364,0.08254141360521319,2,1.528001730677635,1.528001730677635,2.0,,0.5,0.54
14
+ ALL,grad_cosine_layer_mean,gradient,0.5418306061998011,0.5636870458722114,0.021856439672410466,2,0.6269594185774778,0.6269594185774778,2.0,,0.5,0.54
15
+ ALL,grad_cosine_layer_min,gradient,0.20180834084749222,0.13561697304248804,-0.06619136780500415,2,-1.6357796946670995,-1.6357796946670995,0.0,,0.5,0.54
16
+ ALL,grad_l2,gradient,2277.8653612573553,2337.0412759122382,59.17591465488317,2,1.1932420760370865,1.1932420760370865,2.0,,0.5,0.54
17
+ ALL,grad_norm_a,gradient,2268.791188619879,2404.9069901585162,136.1158015386377,2,0.8261454480200053,0.8261454480200053,2.0,,0.5,0.54
18
+ ALL,grad_norm_b,gradient,2661.756166702006,2661.756166702006,0.0,0,,,,,,
19
+ ALL,grad_norm_ratio,gradient,0.8523662749435753,0.9035038671999271,0.05113759225635184,2,0.8261454480200036,0.8261454480200036,2.0,,0.5,0.54
20
+ ALL,act_deviation_from_base,other,0.2909211340174079,0.24877564283087844,-0.0421454911865294,20,-0.5111158417968739,-0.48141035398846777,0.0,1.9073486328125e-06,1.9073486328125e-06,3.08990478515625e-05
21
+ ALL,cka_to_base,other,0.9741214978694913,0.9828484863042831,0.008726988434791572,20,0.5844676284923385,0.3942216092375521,20.0,1.9073486328125e-06,1.9073486328125e-06,3.08990478515625e-05
22
+ ALL,qmd_raw_from_base,other,0.05631466960422178,0.0517850840132395,-0.0045295855909822675,20,-0.22259038161919803,-0.23759881154028703,0.0,1.9073486328125e-06,1.9073486328125e-06,3.08990478515625e-05
23
+ ALL,geo_cka_late_minus_early,representation,-0.0584074902961841,-0.0435430154024333,0.014864474893750802,2,0.7414491197785937,0.7414491197785937,2.0,,0.5,0.54
24
+ ALL,geo_cka_mean,representation,0.9940581026952195,0.9952593661193808,0.0012012634241612141,2,1.0001689963915519,1.0001689963915519,2.0,,0.5,0.54
25
+ ALL,geo_cka_peak_depth,representation,0.1794871794871795,0.1794871794871795,0.0,0,,,,,,
26
+ ALL,geo_procrustes_late_minus_early,representation,0.01086544279675005,0.00829937462303235,-0.0025660681737177007,2,-0.7376741669970976,-0.7376741669970976,0.0,,0.5,0.54
27
+ ALL,geo_procrustes_mean,representation,0.0023543308961208,0.0019042794585369,-0.0004500514375838999,2,-0.976786177433641,-0.976786177433641,0.0,,0.5,0.54
28
+ ALL,geo_procrustes_peak_depth,representation,1.0,1.0,0.0,0,,,,,,
29
+ ALL,geo_rowwise_cosine_late_minus_early,representation,-0.06946131587028495,-0.0566661953926086,0.012795120477676353,2,0.8776596636774059,0.8776596636774059,2.0,,0.5,0.54
30
+ ALL,geo_rowwise_cosine_mean,representation,0.9920770324766637,0.9935909241437912,0.0015138916671276204,2,0.9842387876664018,0.9842387876664018,2.0,,0.5,0.54
31
+ ALL,geo_rowwise_cosine_peak_depth,representation,0.0769230769230769,0.0769230769230769,0.0,0,,,,,,
32
+ ALL,geo_subspace_overlap_late_minus_early,representation,-0.1794701042405935,-0.1519223371296041,0.027547767110989392,2,0.7685767716349439,0.7685767716349439,2.0,,0.5,0.54
33
+ ALL,geo_subspace_overlap_mean,representation,0.804101453468413,0.8264684581812105,0.02236700471279751,2,0.6684380187958745,0.6684380187958745,2.0,,0.5,0.54
34
+ ALL,geo_subspace_overlap_peak_depth,representation,0.0,0.0,0.0,0,,,,,,
35
+ ALL,geo_svcca_late_minus_early,representation,0.00598570917286355,0.0054740719360306,-0.0005116372368329505,2,-0.7858337959004986,-0.7858337959004986,0.0,,0.5,0.54
36
+ ALL,geo_svcca_mean,representation,0.9905037275836849,0.9913711243747523,0.000867396791067454,2,0.5903687492814188,0.5903687492814188,2.0,,0.5,0.54
37
+ ALL,geo_svcca_peak_depth,representation,0.5641025641025641,0.5641025641025641,0.0,0,,,,,,
38
+ ALL,ret_gain_over_identity,retrieval,-0.8461538461538461,-0.8461538461538461,0.0,0,,,,,,
39
+ ALL,ret_identity_p_at_1,retrieval,1.0,1.0,0.0,0,,,,,,
40
+ ALL,ret_identity_p_at_10,retrieval,1.0,1.0,0.0,0,,,,,,
41
+ ALL,ret_identity_p_at_5,retrieval,1.0,1.0,0.0,0,,,,,,
42
+ ALL,ret_procrustes_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
43
+ ALL,ret_procrustes_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
44
+ ALL,ret_procrustes_p_at_5,retrieval,0.6153846153846154,0.5384615384615384,-0.07692307692307698,2,-1.7320508075688774,-1.7320508075688774,0.0,,0.5,0.54
45
+ ALL,ret_ridge_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
46
+ ALL,ret_ridge_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
47
+ ALL,ret_ridge_p_at_5,retrieval,0.5384615384615384,0.5384615384615384,0.0,0,,,,,,
48
+ ALL,spectral_overcounting,weight space,-0.9999980628490448,-0.9999980926513672,-2.9802322387695312e-08,2,-1.0,-1.0,0.0,,1.0,1.0
49
+ ALL,subspace_overlap,weight space,0.9987411499023438,0.999144434928894,0.00040328502655023746,2,1.2045129839362734,1.2045129839362734,1.0,,1.0,1.0
50
+ ALL,weight_cosine,weight space,0.999807596206665,0.999807596206665,0.0,0,,,,,,
51
+ Code-Instruction_Tuned,beh_js_to_base,behaviour,0.05512975991029358,0.0441253616412705,-0.011004398269023082,5,-0.22125968418665526,-0.2481193785443587,0.0,0.0625,0.0625,0.2025
52
+ Code-Instruction_Tuned,beh_kl_to_base,behaviour,0.24558744516642164,0.19273855980014618,-0.052848885366275425,5,-0.24304600251831573,-0.2825243992484339,0.0,0.0625,0.0625,0.2025
53
+ Code-Instruction_Tuned,act_deviation_from_base,other,0.29014053430408243,0.24678958225995298,-0.04335095204412938,5,-0.5257349653054478,-0.5900637262201152,0.0,0.0625,0.0625,0.2025
54
+ Code-Instruction_Tuned,cka_to_base,other,0.9737666422128678,0.983200253546238,0.009433611333370218,5,0.6317918816246285,0.8040619007888237,5.0,0.0625,0.0625,0.2025
55
+ Code-Instruction_Tuned,qmd_raw_from_base,other,0.06490202158222438,0.05963919167922608,-0.005262829902998324,5,-0.25862306671884455,-0.27103230967800196,0.0,0.0625,0.0625,0.2025
56
+ Code-Math,beh_entropy_gap,behaviour,-0.8349796020601645,-0.7557348348527406,0.0792447672074239,2,0.57291702193013,0.57291702193013,2.0,,0.5,0.54
57
+ Code-Math,beh_js,behaviour,0.0597177580157008,0.05124305458083395,-0.008474703434866847,2,-0.6637373428173281,-0.6637373428173281,0.0,,0.5,0.54
58
+ Code-Math,beh_js_to_base,behaviour,0.05932089412574072,0.05064618958360474,-0.008674704542135979,5,-0.17441775011074206,-0.20062602809704028,0.0,0.0625,0.0625,0.2025
59
+ Code-Math,beh_kl_sym,behaviour,0.2695983887171921,0.22880632861482075,-0.04079206010237138,2,-0.6582701545323796,-0.6582701545323796,0.0,,0.5,0.54
60
+ Code-Math,beh_kl_to_base,behaviour,0.25441715897768064,0.21898163798982795,-0.03543552098785269,5,-0.16296392371498078,-0.18759318779542877,0.0,0.0625,0.0625,0.2025
61
+ Code-Math,beh_logit_cosine,behaviour,0.9674639850854874,0.9737388342618942,0.006274849176406916,2,0.8520901541496828,0.8520901541496828,2.0,,0.5,0.54
62
+ Code-Math,beh_logit_kl,behaviour,0.2571983641822849,0.2220319853286175,-0.0351663788536674,2,-0.6427918282100417,-0.6427918282100417,0.0,,0.5,0.54
63
+ Code-Math,beh_logit_l2,behaviour,148.20476531982422,132.70114135742188,-15.503623962402344,2,-0.8766465440312599,-0.8766465440312599,0.0,,0.5,0.54
64
+ Code-Math,beh_rank_corr,behaviour,0.966340842432861,0.9712376856452981,0.0048968432124371075,2,0.7924112086949795,0.7924112086949795,2.0,,0.5,0.54
65
+ Code-Math,beh_topk_overlap,behaviour,0.7703125,0.778125,0.0078125,2,0.33942211665106536,0.33942211665106536,2.0,,0.5,0.54
66
+ Code-Math,grad_cosine,gradient,0.5846191181956859,0.579029991982674,-0.005589126213011886,2,-0.8005706873902222,-0.8005706873902222,1.0,,1.0,1.0
67
+ Code-Math,grad_cosine_layer_late_minus_early,gradient,0.25020570307970047,0.33274711668491364,0.08254141360521319,2,1.528001730677635,1.528001730677635,2.0,,0.5,0.54
68
+ Code-Math,grad_cosine_layer_mean,gradient,0.5418306061998011,0.5636870458722114,0.021856439672410466,2,0.6269594185774778,0.6269594185774778,2.0,,0.5,0.54
69
+ Code-Math,grad_cosine_layer_min,gradient,0.20180834084749222,0.13561697304248804,-0.06619136780500415,2,-1.6357796946670995,-1.6357796946670995,0.0,,0.5,0.54
70
+ Code-Math,grad_l2,gradient,2277.8653612573553,2337.0412759122382,59.17591465488317,2,1.1932420760370865,1.1932420760370865,2.0,,0.5,0.54
71
+ Code-Math,grad_norm_a,gradient,2268.791188619879,2404.9069901585162,136.1158015386377,2,0.8261454480200053,0.8261454480200053,2.0,,0.5,0.54
72
+ Code-Math,grad_norm_b,gradient,2661.756166702006,2661.756166702006,0.0,0,,,,,,
73
+ Code-Math,grad_norm_ratio,gradient,0.8523662749435753,0.9035038671999271,0.05113759225635184,2,0.8261454480200036,0.8261454480200036,2.0,,0.5,0.54
74
+ Code-Math,act_deviation_from_base,other,0.1923275407031178,0.16943953882902857,-0.02288800187408923,5,-0.27757228627727104,-0.2954531074726018,0.0,0.0625,0.0625,0.2025
75
+ Code-Math,cka_to_base,other,0.9902538159489632,0.9923722693324087,0.0021184533834457397,5,0.14187797249254336,0.15566798010118926,5.0,0.0625,0.0625,0.2025
76
+ Code-Math,qmd_raw_from_base,other,0.024078073164752987,0.02234327448026893,-0.0017347986844840567,5,-0.08525051430324117,-0.08992501734153821,0.0,0.0625,0.0625,0.2025
77
+ Code-Math,geo_cka_late_minus_early,representation,-0.0584074902961841,-0.0435430154024333,0.014864474893750802,2,0.7414491197785937,0.7414491197785937,2.0,,0.5,0.54
78
+ Code-Math,geo_cka_mean,representation,0.9940581026952195,0.9952593661193808,0.0012012634241612141,2,1.0001689963915519,1.0001689963915519,2.0,,0.5,0.54
79
+ Code-Math,geo_cka_peak_depth,representation,0.1794871794871795,0.1794871794871795,0.0,0,,,,,,
80
+ Code-Math,geo_procrustes_late_minus_early,representation,0.01086544279675005,0.00829937462303235,-0.0025660681737177007,2,-0.7376741669970976,-0.7376741669970976,0.0,,0.5,0.54
81
+ Code-Math,geo_procrustes_mean,representation,0.0023543308961208,0.0019042794585369,-0.0004500514375838999,2,-0.976786177433641,-0.976786177433641,0.0,,0.5,0.54
82
+ Code-Math,geo_procrustes_peak_depth,representation,1.0,1.0,0.0,0,,,,,,
83
+ Code-Math,geo_rowwise_cosine_late_minus_early,representation,-0.06946131587028495,-0.0566661953926086,0.012795120477676353,2,0.8776596636774059,0.8776596636774059,2.0,,0.5,0.54
84
+ Code-Math,geo_rowwise_cosine_mean,representation,0.9920770324766637,0.9935909241437912,0.0015138916671276204,2,0.9842387876664018,0.9842387876664018,2.0,,0.5,0.54
85
+ Code-Math,geo_rowwise_cosine_peak_depth,representation,0.0769230769230769,0.0769230769230769,0.0,0,,,,,,
86
+ Code-Math,geo_subspace_overlap_late_minus_early,representation,-0.1794701042405935,-0.1519223371296041,0.027547767110989392,2,0.7685767716349439,0.7685767716349439,2.0,,0.5,0.54
87
+ Code-Math,geo_subspace_overlap_mean,representation,0.804101453468413,0.8264684581812105,0.02236700471279751,2,0.6684380187958745,0.6684380187958745,2.0,,0.5,0.54
88
+ Code-Math,geo_subspace_overlap_peak_depth,representation,0.0,0.0,0.0,0,,,,,,
89
+ Code-Math,geo_svcca_late_minus_early,representation,0.00598570917286355,0.0054740719360306,-0.0005116372368329505,2,-0.7858337959004986,-0.7858337959004986,0.0,,0.5,0.54
90
+ Code-Math,geo_svcca_mean,representation,0.9905037275836849,0.9913711243747523,0.000867396791067454,2,0.5903687492814188,0.5903687492814188,2.0,,0.5,0.54
91
+ Code-Math,geo_svcca_peak_depth,representation,0.5641025641025641,0.5641025641025641,0.0,0,,,,,,
92
+ Code-Math,ret_gain_over_identity,retrieval,-0.8461538461538461,-0.8461538461538461,0.0,0,,,,,,
93
+ Code-Math,ret_identity_p_at_1,retrieval,1.0,1.0,0.0,0,,,,,,
94
+ Code-Math,ret_identity_p_at_10,retrieval,1.0,1.0,0.0,0,,,,,,
95
+ Code-Math,ret_identity_p_at_5,retrieval,1.0,1.0,0.0,0,,,,,,
96
+ Code-Math,ret_procrustes_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
97
+ Code-Math,ret_procrustes_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
98
+ Code-Math,ret_procrustes_p_at_5,retrieval,0.6153846153846154,0.5384615384615384,-0.07692307692307698,2,-1.7320508075688774,-1.7320508075688774,0.0,,0.5,0.54
99
+ Code-Math,ret_ridge_p_at_1,retrieval,0.1538461538461538,0.1538461538461538,0.0,0,,,,,,
100
+ Code-Math,ret_ridge_p_at_10,retrieval,0.8461538461538461,0.8461538461538461,0.0,0,,,,,,
101
+ Code-Math,ret_ridge_p_at_5,retrieval,0.5384615384615384,0.5384615384615384,0.0,0,,,,,,
102
+ Code-Math,spectral_overcounting,weight space,-0.9999980628490448,-0.9999980926513672,-2.9802322387695312e-08,2,-1.0,-1.0,0.0,,1.0,1.0
103
+ Code-Math,subspace_overlap,weight space,0.9987411499023438,0.999144434928894,0.00040328502655023746,2,1.2045129839362734,1.2045129839362734,1.0,,1.0,1.0
104
+ Code-Math,weight_cosine,weight space,0.999807596206665,0.999807596206665,0.0,0,,,,,,
105
+ Code-Math-Instruction_Tuned,beh_js_to_base,behaviour,0.12113742600657493,0.10746916728598466,-0.01366825872059028,5,-0.2748205339325454,-0.2911923025185968,0.0,0.0625,0.0625,0.2025
106
+ Code-Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.5253016705460712,0.45086890549425257,-0.07443276505181856,5,-0.3423078060559,-0.3661812320817214,0.0,0.0625,0.0625,0.2025
107
+ Code-Math-Instruction_Tuned,act_deviation_from_base,other,0.34054103039205075,0.2870888278074562,-0.05345220258459451,5,-0.6482370177869512,-0.7525949840811219,0.0,0.0625,0.0625,0.2025
108
+ Code-Math-Instruction_Tuned,cka_to_base,other,0.9637338295578957,0.9776740205287933,0.013940190970897692,5,0.9336084742600528,1.1128547417209058,5.0,0.0625,0.0625,0.2025
109
+ Code-Math-Instruction_Tuned,qmd_raw_from_base,other,0.06741627142434369,0.06192478721447022,-0.005491484209873472,5,-0.2698594697857254,-0.29428904562512653,0.0,0.0625,0.0625,0.2025
110
+ Math-Instruction_Tuned,beh_js_to_base,behaviour,0.122818348409463,0.10852165823966838,-0.014296690169794596,5,-0.2874560766114504,-0.2804668634303113,0.0,0.0625,0.0625,0.2025
111
+ Math-Instruction_Tuned,beh_kl_to_base,behaviour,0.5402923311013552,0.45799824939797984,-0.08229408170337532,5,-0.3784611056657115,-0.44202447669781614,0.0,0.0625,0.0625,0.2025
112
+ Math-Instruction_Tuned,act_deviation_from_base,other,0.3406754306703806,0.2917846224270761,-0.048890808243304495,5,-0.5929190978178251,-0.6552347513245169,0.0,0.0625,0.0625,0.2025
113
+ Math-Instruction_Tuned,cka_to_base,other,0.9687317037582398,0.9781474018096924,0.009415698051452637,5,0.6305921855921296,0.7110382045334173,5.0,0.0625,0.0625,0.2025
114
+ Math-Instruction_Tuned,qmd_raw_from_base,other,0.06886231224556602,0.06323308267899282,-0.005629229566573214,5,-0.27662847566898086,-0.29392234507751674,0.0,0.0625,0.0625,0.2025
tables/T5_coverage.csv CHANGED
@@ -1,9 +1,9 @@
1
  operator,operator_label,combo,repo_baseline,repo_with_aim,ledger_status,seconds,weight_mechanism,activation_mechanism,property_panel_both_arms,published_benchmarks,note
2
- TaskArithmetic,Task Arithmetic,Code-Math,ahn1376/TaskArithmetic___Code-Math,ahn1376/TaskArithmetic___Code-Math___AIM,not-run,,True,True,False,True,
3
  TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,ahn1376/TaskArithmetic___Code-Instruction_Tuned,ahn1376/TaskArithmetic___Code-Instruction_Tuned___AIM,not-run,,True,True,False,True,
4
  TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,ahn1376/TaskArithmetic___Math-Instruction_Tuned,ahn1376/TaskArithmetic___Math-Instruction_Tuned___AIM,not-run,,True,True,False,True,
5
  TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,ahn1376/TaskArithmetic___Code-Math-Instruction_Tuned,ahn1376/TaskArithmetic___Code-Math-Instruction_Tuned___AIM,not-run,,True,True,False,True,
6
- Ties,TIES,Code-Math,ahn1376/Ties___Code-Math,ahn1376/Ties___Code-Math___AIM,not-run,,True,True,False,True,
7
  Ties,TIES,Code-Instruction_Tuned,ahn1376/Ties___Code-Instruction_Tuned,ahn1376/Ties___Code-Instruction_Tuned___AIM,not-run,,True,True,False,True,
8
  Ties,TIES,Math-Instruction_Tuned,ahn1376/Ties___Math-Instruction_Tuned,ahn1376/Ties___Math-Instruction_Tuned___AIM,not-run,,True,True,False,True,
9
  Ties,TIES,Code-Math-Instruction_Tuned,ahn1376/Ties___Code-Math-Instruction_Tuned,ahn1376/Ties___Code-Math-Instruction_Tuned___AIM,not-run,,True,True,False,True,
 
1
  operator,operator_label,combo,repo_baseline,repo_with_aim,ledger_status,seconds,weight_mechanism,activation_mechanism,property_panel_both_arms,published_benchmarks,note
2
+ TaskArithmetic,Task Arithmetic,Code-Math,ahn1376/TaskArithmetic___Code-Math,ahn1376/TaskArithmetic___Code-Math___AIM,not-run,,True,True,True,True,
3
  TaskArithmetic,Task Arithmetic,Code-Instruction_Tuned,ahn1376/TaskArithmetic___Code-Instruction_Tuned,ahn1376/TaskArithmetic___Code-Instruction_Tuned___AIM,not-run,,True,True,False,True,
4
  TaskArithmetic,Task Arithmetic,Math-Instruction_Tuned,ahn1376/TaskArithmetic___Math-Instruction_Tuned,ahn1376/TaskArithmetic___Math-Instruction_Tuned___AIM,not-run,,True,True,False,True,
5
  TaskArithmetic,Task Arithmetic,Code-Math-Instruction_Tuned,ahn1376/TaskArithmetic___Code-Math-Instruction_Tuned,ahn1376/TaskArithmetic___Code-Math-Instruction_Tuned___AIM,not-run,,True,True,False,True,
6
+ Ties,TIES,Code-Math,ahn1376/Ties___Code-Math,ahn1376/Ties___Code-Math___AIM,not-run,,True,True,True,True,
7
  Ties,TIES,Code-Instruction_Tuned,ahn1376/Ties___Code-Instruction_Tuned,ahn1376/Ties___Code-Instruction_Tuned___AIM,not-run,,True,True,False,True,
8
  Ties,TIES,Math-Instruction_Tuned,ahn1376/Ties___Math-Instruction_Tuned,ahn1376/Ties___Math-Instruction_Tuned___AIM,not-run,,True,True,False,True,
9
  Ties,TIES,Code-Math-Instruction_Tuned,ahn1376/Ties___Code-Math-Instruction_Tuned,ahn1376/Ties___Code-Math-Instruction_Tuned___AIM,not-run,,True,True,False,True,
tables/T5b_column_provenance.csv CHANGED
@@ -1,26 +1,26 @@
1
  column,family,in_F8,populated,reason_if_not
2
- weight_cosine,weight space,True,False,sweep incomplete
3
  qmd_raw,weight space,True,False,sweep incomplete
4
  coord_fraction,weight space,True,False,NaN by construction: coordinate_gap / qmd_raw with a zero numerator.
5
- subspace_overlap,weight space,True,False,sweep incomplete
6
- spectral_overcounting,weight space,True,False,sweep incomplete
7
- geo_cka_mean,representation,True,False,sweep incomplete
8
- geo_procrustes_mean,representation,True,False,sweep incomplete
9
- geo_subspace_overlap_mean,representation,True,False,sweep incomplete
10
- geo_svcca_mean,representation,True,False,sweep incomplete
11
- geo_cka_late_minus_early,representation,True,False,sweep incomplete
12
- ret_identity_p_at_1,retrieval,True,False,sweep incomplete
13
- ret_procrustes_p_at_1,retrieval,True,False,sweep incomplete
14
- ret_ridge_p_at_1,retrieval,True,False,sweep incomplete
15
- ret_gain_over_identity,retrieval,True,False,sweep incomplete
16
- grad_cosine,gradient,True,False,sweep incomplete
17
- grad_l2,gradient,True,False,sweep incomplete
18
- grad_norm_ratio,gradient,True,False,sweep incomplete
19
- grad_cosine_layer_min,gradient,True,False,sweep incomplete
20
- beh_js,behaviour,True,False,sweep incomplete
21
- beh_logit_cosine,behaviour,True,False,sweep incomplete
22
- beh_topk_overlap,behaviour,True,False,sweep incomplete
23
- beh_rank_corr,behaviour,True,False,sweep incomplete
24
- beh_entropy_gap,behaviour,True,False,sweep incomplete
25
  geoX_*,representation,False,False,"extended 512-sentence probe not run: it is the robustness check on the n/d problem, not the figure, and the 41-layer Procrustes/SVCCA geometry at 512x5120 dominated the sweep's cost on a contended box."
26
  retX_*,retrieval,False,False,"extended 512-sentence probe not run: it is the robustness check on the n/d problem, not the figure, and the 41-layer Procrustes/SVCCA geometry at 512x5120 dominated the sweep's cost on a contended box."
 
1
  column,family,in_F8,populated,reason_if_not
2
+ weight_cosine,weight space,True,True,
3
  qmd_raw,weight space,True,False,sweep incomplete
4
  coord_fraction,weight space,True,False,NaN by construction: coordinate_gap / qmd_raw with a zero numerator.
5
+ subspace_overlap,weight space,True,True,
6
+ spectral_overcounting,weight space,True,True,
7
+ geo_cka_mean,representation,True,True,
8
+ geo_procrustes_mean,representation,True,True,
9
+ geo_subspace_overlap_mean,representation,True,True,
10
+ geo_svcca_mean,representation,True,True,
11
+ geo_cka_late_minus_early,representation,True,True,
12
+ ret_identity_p_at_1,retrieval,True,True,
13
+ ret_procrustes_p_at_1,retrieval,True,True,
14
+ ret_ridge_p_at_1,retrieval,True,True,
15
+ ret_gain_over_identity,retrieval,True,True,
16
+ grad_cosine,gradient,True,True,
17
+ grad_l2,gradient,True,True,
18
+ grad_norm_ratio,gradient,True,True,
19
+ grad_cosine_layer_min,gradient,True,True,
20
+ beh_js,behaviour,True,True,
21
+ beh_logit_cosine,behaviour,True,True,
22
+ beh_topk_overlap,behaviour,True,True,
23
+ beh_rank_corr,behaviour,True,True,
24
+ beh_entropy_gap,behaviour,True,True,
25
  geoX_*,representation,False,False,"extended 512-sentence probe not run: it is the robustness check on the n/d problem, not the figure, and the 41-layer Procrustes/SVCCA geometry at 512x5120 dominated the sweep's cost on a contended box."
26
  retX_*,retrieval,False,False,"extended 512-sentence probe not run: it is the robustness check on the n/d problem, not the figure, and the 41-layer Procrustes/SVCCA geometry at 512x5120 dominated the sweep's cost on a contended box."