Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -355,15 +355,17 @@ Every table is also a CSV under `results/aim/tables/` and on the Hub at [`Mergea
|
|
| 355 |
|
| 356 |
The direct test of AIM's claim. Needs no benchmark scores. `rho_W` is the weight-space shrink `‖ΔW_AIM‖/‖ΔW_merge‖`; `rho_A_*` is the same ratio measured on activations (4096 pile-val token positions, mean over layers); `aim_advantage = rho_A_uniform − rho_A_aim` is **positive when the activation-informed shrink preserves the base model's activations better than an activation-agnostic shrink of exactly the same weight-space size**. `omega_hat` and `col_r2_median` are the recovered closed form; `salience_recovery_r` is the correlation between the salience implied by the published checkpoint pair and the salience measured independently from the base model.
|
| 357 |
|
| 358 |
-
| operator_label | combo
|
| 359 |
-
|:-----------------|:----------|------------:|----------------:|-------------------:|----------------------
|
| 360 |
-
| DARE Linear | Code-Math
|
| 361 |
-
| DARE TIES | Code-Math
|
| 362 |
-
| Task Arithmetic | Code-
|
| 363 |
-
|
|
|
|
|
|
|
|
| 364 |
|
| 365 |
|
| 366 |
-
**Paired test of the targeting advantage** over the magnitude-matched uniform control, across the
|
| 367 |
|
| 368 |
|
| 369 |
### T0 — The merge outcome, paired (published numbers, nothing re-run)
|
|
@@ -411,12 +413,29 @@ Per benchmark, paired across the 20 matched pairs:
|
|
| 411 |
|
| 412 |
### T2 — Does AIM's effect differ by merge operator?
|
| 413 |
|
| 414 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 415 |
|
| 416 |
|
| 417 |
### T3 — Does it differ by which domains were merged?
|
| 418 |
|
| 419 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 420 |
|
| 421 |
|
| 422 |
### T1 — Paired per-checkpoint table
|
|
@@ -431,7 +450,7 @@ One row per (operator, task combination, property): the with-AIM value, its base
|
|
| 431 |
| operator_label | combo | ledger_status | weight_mechanism | activation_mechanism | property_panel_both_arms | published_benchmarks |
|
| 432 |
|:-----------------|:----------------------------|:----------------|:-------------------|:-----------------------|:---------------------------|:-----------------------|
|
| 433 |
| Task Arithmetic | Code-Math | not-run | True | True | False | True |
|
| 434 |
-
| Task Arithmetic | Code-Instruction_Tuned | not-run |
|
| 435 |
| Task Arithmetic | Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 436 |
| Task Arithmetic | Code-Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 437 |
| TIES | Code-Math | not-run | True | True | False | True |
|
|
@@ -446,7 +465,7 @@ One row per (operator, task combination, property): the with-AIM value, its base
|
|
| 446 |
| DARE TIES | Code-Instruction_Tuned | not-run | False | False | False | True |
|
| 447 |
| DARE TIES | Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 448 |
| DARE TIES | Code-Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 449 |
-
| WIDEN | Code-Math | not-run |
|
| 450 |
| WIDEN | Code-Instruction_Tuned | not-run | False | False | False | True |
|
| 451 |
| WIDEN | Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 452 |
| WIDEN | Code-Math-Instruction_Tuned | not-run | False | False | False | True |
|
|
|
|
| 355 |
|
| 356 |
The direct test of AIM's claim. Needs no benchmark scores. `rho_W` is the weight-space shrink `‖ΔW_AIM‖/‖ΔW_merge‖`; `rho_A_*` is the same ratio measured on activations (4096 pile-val token positions, mean over layers); `aim_advantage = rho_A_uniform − rho_A_aim` is **positive when the activation-informed shrink preserves the base model's activations better than an activation-agnostic shrink of exactly the same weight-space size**. `omega_hat` and `col_r2_median` are the recovered closed form; `salience_recovery_r` is the correlation between the salience implied by the published checkpoint pair and the salience measured independently from the base model.
|
| 357 |
|
| 358 |
+
| operator_label | combo | omega_hat | col_r2_median | scalar_r2_median | salience_recovery_r | rho_W | rel_dev_merge | rel_dev_uniform | rel_dev_aim | rho_A_uniform | rho_A_aim | aim_advantage | leverage_uniform | leverage_aim |
|
| 359 |
+
|:-----------------|:-----------------------|------------:|----------------:|-------------------:|:----------------------|--------:|----------------:|:------------------|:--------------|:----------------|:------------|:----------------|:-------------------|:---------------|
|
| 360 |
+
| DARE Linear | Code-Math | 0.3971 | 0.9955 | 0.9949 | 0.9934 | 0.9239 | 0.209 | 0.2017 | 0.1832 | 0.9652 | 0.8767 | 0.08843 | 0.4579 | 1.62 |
|
| 361 |
+
| DARE TIES | Code-Math | 0.3961 | 0.9989 | 0.9984 | 0.9923 | 0.9256 | 0.2114 | 0.2053 | 0.1847 | 0.971 | 0.8739 | 0.09717 | 0.3893 | 1.695 |
|
| 362 |
+
| Task Arithmetic | Code-Instruction_Tuned | 0.3999 | 0.9995 | 0.9991 | — | 0.9185 | 0.319 | 0.3071 | 0.2704 | 0.9626 | 0.8475 | 0.1151 | 0.4592 | 1.871 |
|
| 363 |
+
| Task Arithmetic | Code-Math | 0.3958 | 0.9988 | 0.9985 | 0.9921 | 0.9266 | 0.2079 | 0.2019 | 0.1835 | 0.9711 | 0.8828 | 0.08823 | 0.3946 | 1.597 |
|
| 364 |
+
| TIES | Code-Math | 0.3939 | 0.9988 | 0.9984 | 0.99 | 0.9308 | 0.1834 | 0.1784 | 0.1624 | 0.9729 | 0.8855 | 0.08738 | 0.392 | 1.654 |
|
| 365 |
+
| WIDEN | Code-Math | 0.3717 | 0.9899 | 0.9894 | — | 0.9352 | 0.15 | — | — | — | — | — | — | — |
|
| 366 |
|
| 367 |
|
| 368 |
+
**Paired test of the targeting advantage** over the magnitude-matched uniform control, across the 5 matched pairs: mean +0.09526, median +0.08843, 5/5 positive, Wilcoxon signed rank p = 0.0625, sign test p = 0.0625.
|
| 369 |
|
| 370 |
|
| 371 |
### T0 — The merge outcome, paired (published numbers, nothing re-run)
|
|
|
|
| 413 |
|
| 414 |
### T2 — Does AIM's effect differ by merge operator?
|
| 415 |
|
| 416 |
+
| operator | n_metrics | n_moved_q<0.05 | mean_|diff| (SD units) | largest movers |
|
| 417 |
+
|:-------------------|------------:|-----------------:|-------------------------:|:--------------------------------------------------------------------------------|
|
| 418 |
+
| ALL | 5 | 0 | 0.5915 | beh_js_to_base (-0.85), beh_kl_to_base (-0.82), act_deviation_from_base (-0.60) |
|
| 419 |
+
| DARETaskArithmetic | 5 | 0 | 0.5261 | beh_js_to_base (-0.89), beh_kl_to_base (-0.78), act_deviation_from_base (-0.53) |
|
| 420 |
+
| DARETies | 5 | 0 | 0.5311 | beh_js_to_base (-0.85), beh_kl_to_base (-0.78), act_deviation_from_base (-0.54) |
|
| 421 |
+
| TaskArithmetic | 5 | 0 | 0.7688 | beh_kl_to_base (-1.00), beh_js_to_base (-0.98), cka_to_base (+0.90) |
|
| 422 |
+
| Ties | 5 | 0 | 0.3762 | beh_js_to_base (-0.57), beh_kl_to_base (-0.56), act_deviation_from_base (-0.43) |
|
| 423 |
+
| WIDEN | 5 | 0 | 0.07494 | qmd_raw_from_base (-0.07), beh_js_to_base (+nan), beh_kl_to_base (+nan) |
|
| 424 |
+
|
| 425 |
+
|
| 426 |
+
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).
|
| 427 |
|
| 428 |
|
| 429 |
### T3 — Does it differ by which domains were merged?
|
| 430 |
|
| 431 |
+
| combo | n_metrics | n_moved_q<0.05 | mean_|diff| (SD units) | largest movers |
|
| 432 |
+
|:-----------------------|------------:|-----------------:|-------------------------:|:--------------------------------------------------------------------------------|
|
| 433 |
+
| ALL | 5 | 0 | 0.5915 | beh_js_to_base (-0.85), beh_kl_to_base (-0.82), act_deviation_from_base (-0.60) |
|
| 434 |
+
| Code-Instruction_Tuned | 5 | 0 | 1.029 | cka_to_base (+1.50), beh_kl_to_base (-1.22), beh_js_to_base (-1.09) |
|
| 435 |
+
| Code-Math | 5 | 0 | 0.4841 | beh_js_to_base (-0.79), beh_kl_to_base (-0.72), act_deviation_from_base (-0.50) |
|
| 436 |
+
|
| 437 |
+
|
| 438 |
+
Full table: `results/aim/tables/T3_per_combo.csv` (15 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).
|
| 439 |
|
| 440 |
|
| 441 |
### T1 — Paired per-checkpoint table
|
|
|
|
| 450 |
| operator_label | combo | ledger_status | weight_mechanism | activation_mechanism | property_panel_both_arms | published_benchmarks |
|
| 451 |
|:-----------------|:----------------------------|:----------------|:-------------------|:-----------------------|:---------------------------|:-----------------------|
|
| 452 |
| Task Arithmetic | Code-Math | not-run | True | True | False | True |
|
| 453 |
+
| Task Arithmetic | Code-Instruction_Tuned | not-run | True | True | False | True |
|
| 454 |
| Task Arithmetic | Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 455 |
| Task Arithmetic | Code-Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 456 |
| TIES | Code-Math | not-run | True | True | False | True |
|
|
|
|
| 465 |
| DARE TIES | Code-Instruction_Tuned | not-run | False | False | False | True |
|
| 466 |
| DARE TIES | Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 467 |
| DARE TIES | Code-Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 468 |
+
| WIDEN | Code-Math | not-run | True | True | False | True |
|
| 469 |
| WIDEN | Code-Instruction_Tuned | not-run | False | False | False | True |
|
| 470 |
| WIDEN | Math-Instruction_Tuned | not-run | False | False | False | True |
|
| 471 |
| WIDEN | Code-Math-Instruction_Tuned | not-run | False | False | False | True |
|