Design
This diagnostic tests the draft scaling suspicion raised after the grouped
GSM8K reward follow-ups: raw decomposed direction and magnitude learning rates
may not correspond to comparable effective weight-space updates. It is not an
accuracy sweep. The run uses one grouped GSM8K reward batch, applies one ES
update, and measures the resulting parameter-space and effective DoRA
weight-space update sizes for plain EGGROLL and decomposed EGGROLL.
Results
20260803-104230-pretrained_llm_gsm8k_scale_diagnostic_jean_zay-cb7f7a9
commit: cb7f7a9
| metric | value |
|---|---|
| decomposed_lr_direction | 0.02 |
| decomposed_lr_magnitude_0_mean_effective_weight_relative_update | 0.0008017193681629564 |
| decomposed_lr_magnitude_0_mean_magnitude_effective_weight_delta_norm | 0 |
| decomposed_lr_magnitude_0_mean_magnitude_param_relative_update | 0 |
| decomposed_lr_magnitude_0_mean_magnitude_to_direction_effective_weight_delta_ratio | null |
| decomposed_lr_magnitude_0_value | 0 |
| decomposed_lr_magnitude_1_mean_effective_weight_relative_update | 0.0008040715523299729 |
| decomposed_lr_magnitude_1_mean_magnitude_effective_weight_delta_norm | 0.12167434429702485 |
| decomposed_lr_magnitude_1_mean_magnitude_param_relative_update | 6.145854126400007e-05 |
| decomposed_lr_magnitude_1_mean_magnitude_to_direction_effective_weight_delta_ratio | null |
| decomposed_lr_magnitude_1_value | 0.02 |
| decomposed_lr_magnitude_2_mean_effective_weight_relative_update | 0.0008585934467492261 |
| decomposed_lr_magnitude_2_mean_magnitude_effective_weight_delta_norm | 0.608371534614009 |
| decomposed_lr_magnitude_2_mean_magnitude_param_relative_update | 0.00030729263619639946 |
| decomposed_lr_magnitude_2_mean_magnitude_to_direction_effective_weight_delta_ratio | null |
| decomposed_lr_magnitude_2_value | 0.1 |
| decomposed_mean_direction_effective_weight_delta_norm | 1.5872272590950383 |
| decomposed_mean_direction_param_relative_update | 0.0008023099603004136 |
| decomposed_mean_effective_weight_relative_update | 0.0008214614557473851 |
| decomposed_mean_fitness_std | 0.7432134648164114 |
| decomposed_mean_magnitude_effective_weight_delta_norm | 0.24334862630367793 |
| decomposed_mean_magnitude_param_relative_update | 0.00012291705915346652 |
| decomposed_mean_magnitude_to_direction_effective_weight_delta_ratio | null |
| decomposed_mean_magnitude_to_direction_param_update_ratio | null |
| decomposed_mean_train_accuracy | 0.020833333333333332 |
| decomposed_mean_train_parseable_rate | 0.9895833333333334 |
| decomposed_mean_used_parseability_shaping | 0.5 |
| decomposed_sigma_direction | 0.01 |
| decomposed_sigma_magnitude | 0.02 |
| decomposed_tangent_project_direction | 0 |
| eggroll_lr | 0.02 |
| eggroll_mean_fitness_std | 0.6009588241577148 |
| eggroll_mean_train_accuracy | 0.0078125 |
| eggroll_mean_train_parseable_rate | 0.9869791666666666 |
| eggroll_mean_used_parseability_shaping | 0.6666666666666666 |
| eggroll_mean_weight_relative_update | 0.0006769713713991269 |
| eggroll_sigma | 0.01 |
| generations_per_prompt | 8 |
| num_decomposed_lr_magnitudes | 3 |
| num_ranks | 2 |
| num_seeds | 3 |
| num_target_modules | 7 |
| population | 64 |
| prompts_per_epoch | 8 |
| wall_clock_seconds | 2001.3164142081514 |
Interpretation
Run 20260803-104230-pretrained_llm_gsm8k_scale_diagnostic_jean_zay-cb7f7a9
supports the scale-mismatch explanation. With the old equal decomposed learning
rates, lr_magnitude=0.02 produced a mean magnitude effective-weight delta of
0.12167434429702485, while the mean direction effective-weight delta was
1.5872272590950383; the ratio of mean deltas is therefore about
0.07665842657364502. Raising lr_magnitude to 0.1 raised the magnitude
delta to 0.608371534614009, or about 0.38329201513390937 of the direction
delta. That makes the previous observation that 0.1 was the least bad
magnitude LR much less weird: 0.02 was probably under-scaling magnitude
relative to direction.
The boring alternative explanation is still that the GSM8K reward batch is
mostly sparse/noisy. The diagnostic does not rule that out: train accuracy was
low, parseability shaping was used on a substantial fraction of batches, and
this was only a one-step measurement. What it does rule against is the idea
that equal raw direction/magnitude learning rates were already a fair scale
comparison.
Next, do not run another broad accuracy sweep with equal decomposed rates.
Either tune magnitude to match a target fraction of the direction effective
delta, or run a small two-dimensional scale grid over direction and magnitude
where the reported metric is both reward movement and effective update size.