ar-overfit-i1-d262031-4icons-9b9b1699
completed —
Measured behaviour
Table view
| optimizer step | free-token accuracy |
|---|---|
| 100 | 0.9978 |
| 200 | 1.0000 |
| 300 | 1.0000 |
| 400 | 1.0000 |
| 500 | 1.0000 |
| 600 | 1.0000 |
| 700 | 1.0000 |
| 800 | 1.0000 |
| 900 | 1.0000 |
| 1000 | 1.0000 |
| 1100 | 1.0000 |
| 1200 | 1.0000 |
| 1300 | 1.0000 |
| 1400 | 1.0000 |
| 1500 | 1.0000 |
| 1600 | 1.0000 |
| 1700 | 1.0000 |
| 1800 | 1.0000 |
| 1900 | 1.0000 |
| 2000 | 1.0000 |
Measured result
Verbatim from the run's summary.json.
checkpoint_round_trip
true
checkpoint_sha256
b282b73556bb027d…checks
all_valid
true
exact_match_rate
true
free_token_accuracy
true
loss_reduction_factor
true
config_sha256
a39ad1eebe6a156f…criteria
min_exact_match_rate
1.0
min_free_token_accuracy
0.99
min_loss_reduction_factor
100.0
require_all_valid
true
deterministic_algorithms
true
device
cuda
exact_match_rate
1.0
final
free_token_accuracy
1.0
loss
0.00023123926075641066
step
2000
free_positions_per_icon
464
icons
color/svg/1F3F4-E0064-E0065-E0062-E0065-E007F.svg, color/svg/1F468-1F3FC-200D-1F9BC.svg, color/svg/1F561.svg, color/svg/1F469-1F3FD-200D-2764-FE0F-200D-1F48B-200D-1F469-1F3FE.svg
initial_loss
4.602105617523193
loss_reduction_factor
19901.921509648353
metrics_sha256
94bc16eb8f9c2fab…model_parameters
375106
predeclared_outcome
passed
samples
decode_calls
1376
exact_match
true
hexcode
1F3F4-E0064-E0065-E0062-E0065-E007F
rendered
true
subgroup
subdivision-flag
svg_sha256
057a2da3aa29e10c…valid
true
decode_calls
1376
exact_match
true
hexcode
1F468-1F3FC-200D-1F9BC
rendered
true
subgroup
person-activity
svg_sha256
9193529ce7f4bbb2…valid
true
decode_calls
1376
exact_match
true
hexcode
1F561
rendered
true
subgroup
time
svg_sha256
2409a243d236ec20…valid
true
decode_calls
1376
exact_match
true
hexcode
1F469-1F3FD-200D-2764-FE0F-200D-1F48B-200D-1F469-1F3FE
rendered
true
subgroup
family
svg_sha256
334c71748707b2ab…valid
true
schema_version
1
scope
autoregressive learnability diagnostic; not a generation result
sequence_length
1376
study_version
ar-overfit-i1
torch_version
2.14.0a0+4fdf77b940.nv26.08
vocabulary
418
State transitions
- planned2026-09-21T09:10:00Z
- completed2026-09-21T09:18:00Zthe causal model memorises four icons exactly through cache and masks; learnability only, not generation
Run record
This run has no run.yaml. What follows is the identity and configuration carried by its rows in state/runs.jsonl, the append-only registry.
run_id
ar-overfit-i1-d262031-4icons-9b9b1699
config
configs/learning/ar-overfit-i1.yaml