5.1 KiB
Qwen3.5-2B parameter similarity pilot
- Revision:
15852e8c16360a2fea060d615a32b45270f8a8fc - Stored parameters: 2,274,069,824
- Weight bytes: 4,548,144,832
- Tensor dimensionalities: {'1': 319, '2': 294, '3': 18, '5': 1}
- Analysis and grouping/export wall time: 6661.1 seconds
- Pair-comparison cutoff: 6600 seconds; final grouping/export is additional and remains under the supervisor hard limit.
- Peak analysis RSS: 201.3 MiB
Verified sharing estimates
| Representation | Weight-error limit | Verified sharing groups | Conservative total groups* | Largest group | Net model savings | |
|---|---|---|---|---|---|---|
| whole | 1% | 0 | 632 | 1 | 0.000000% | |
| whole | 2% | 0 | 632 | 1 | 0.000000% | |
| whole | 5% | 1 | 631 | 2 | 0.000000% | |
| whole | 10% | 11 | 613 | 6 | 0.000719% | |
| whole | 20% | 16 | 572 | 15 | 0.002626% | |
| 2d-128 | 1% | 0 | 138576 | 1 | 0.000000% | |
| 2d-128 | 2% | 0 | 138576 | 1 | 0.000000% | |
| 2d-128 | 5% | 0 | 138576 | 1 | 0.000000% | |
| 2d-128 | 10% | 0 | 138576 | 1 | 0.000000% | |
| 2d-128 | 20% | 1 | 138575 | 2 | 0.000000% | |
| 2d-64 | 1% | 0 | 554304 | 1 | 0.000000% | |
| 2d-64 | 2% | 0 | 554304 | 1 | 0.000000% | |
| 2d-64 | 5% | 0 | 554304 | 1 | 0.000000% | |
| 2d-64 | 10% | 1 | 554303 | 2 | 0.000000% | |
| 2d-64 | 20% | 3 | 554299 | 3 | 0.000000% | |
| 3d-axis0-slice | 1% | 0 | 110592 | 1 | 0.000000% | |
| 3d-axis0-slice | 2% | 0 | 110592 | 1 | 0.000000% | |
| 3d-axis0-slice | 5% | 9 | 110581 | 3 | 0.000000% | |
| 3d-axis0-slice | 10% | 33 | 110542 | 7 | 0.000000% | |
| 3d-axis0-slice | 20% | 45 | 110483 | 12 | 0.000000% | |
| 3d-volume | 1% | 0 | 27648 | 1 | 0.000000% | |
| 3d-volume | 2% | 0 | 27648 | 1 | 0.000000% | |
| 3d-volume | 5% | 0 | 27648 | 1 | 0.000000% | |
| 3d-volume | 10% | 0 | 27648 | 1 | 0.000000% | |
| 3d-volume | 20% | 2 | 27646 | 2 | 0.000000% |
*Conservative total groups counts unexamined full blocks as singletons; boundary regions and ineligible tensors stay unchanged and are outside that block count. Savings count only verified replacement edges. Unsampled blocks remain unchanged. Different block scales must not be summed. Theoretical encoding: one stored representative per group, 16 bytes per object mapping and 64 bytes per shared-group descriptor; identity fallback if larger. Exact grouping uses direct replacement only. Scaled and spectral results are diagnostic.
Stratified diagnostics
See analysis/role_summary.csv for same-role, same-shape statistics, separated by tensor dimensionality and block scale.
See analysis/dimensional_bytes.csv for tensor counts versus weight-byte contributions by dimensionality.
Heatmap missing pairs are hatched (heatmap_pairs.csv); identity entries are not empirical matches.
Layer-distance medians are split by dimension/scale (layer_distance_stratified.csv); role and selection bias remain.
Controls are split by dimension and density-normalized, with sample counts (null_histograms.csv); these are not significance tests.
These are screened-candidate statistics, not estimates of all possible pairs. Small normalization vectors must not be conflated with large matrices.
Coverage
- tensor:
{"eligible_pairs_in_screened_objects": 6159, "candidate_pairs_before_cap": 4724, "selected_pairs": 2000, "objects_screened": 632, "pairs_verified": 2000, "timed_out": false, "screen_sample_elements": 2048, "candidate_neighbors_per_metric": 10} - block:
{"eligible_pairs_in_screened_objects": 1678968, "candidate_pairs_before_cap": 175036, "selected_pairs": 20000, "objects_screened": 13328, "pairs_verified": 14778, "timed_out": true, "screen_sample_elements": 2048, "candidate_neighbors_per_metric": 10} - Complete-block partition totals (per alternative):
{"2d-64": {"blocks": 554304, "bytes": 4540858368, "edge_bytes": 0}, "2d-128": {"blocks": 138576, "bytes": 4540858368, "edge_bytes": 0}, "3d-axis0-slice": {"blocks": 110592, "bytes": 884736, "edge_bytes": 0}, "3d-volume": {"blocks": 27648, "bytes": 884736, "edge_bytes": 0}}
Interpretation limits
- No inference or functional equivalence measured
- No exhaustive pair search
- Block search uses per-role/shape/scale reservoirs of 256
- Different block scales overlap; savings must not be added
- Block savings count only verified disjoint sampled blocks, no extrapolation
- Spectral probes use aligned bounded subvolumes, not whole tensors
- Shared encoding is theoretical, not a runnable compressed checkpoint
- No cross-role grouping, neuron permutation alignment, or training
- Candidate histograms are selection-biased and do not represent the distribution of all tensor pairs.
- 1D normalization weights may cluster strongly but carry few bytes; do not infer material compression from group counts.
- No task accuracy, perplexity, inference speed or runtime-memory improvement has been measured.
- Stored parameter counts may omit tied aliases; no duplicate-weight equality search is performed.
Figures
