CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Tue 17 Nov, 23:36 UTC
%
0.8580429 of 500 eval, 71 wrong
▲ +69.00 pts across 4 evals▲ +1.60 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Familiar decider: ModernBERT-base (backfilled)

finishedRadeon 8060S

One of the small models that decide whether Familiar, a desktop companion, reacts to something happening on the PC, and how. Backfilled: recorded after the fact from the run's own logs and records. Times inside a run are estimated, and there are no GPU sensor readings.

started
Tue 17 Nov, 23:17 UTC
ran for
14:48
config
learning rate 0.0001, batch 16, seed 0

run1/5step1125/1125epoch3/3config1/3seed0lr0.0001batch16modelModernBERT-base 149Mrev8949b90items6000+500

Eval measures

Elapsed

296.0 s

epoch interval

296.0 sto eval 2296.0 sto eval 3296.0 sto eval 4

296.0 s to eval 2, 296.0 s to eval 3, 296.0 s to eval 4

1125 of 1125 steps, 3 of 3 epochs, 100%

0.79 s per step, 76.0 steps a minute

Eval accuracy by eval

+0.0160

▲ on the eval before

eval 1 0.1680, eval 4 0.8580

Against the other runs

5 of 5

by last eval accuracy

best so far 0.9180, run 5

Set-up

6000 train, 500 eval

synthetic, learning rate 0.0001, batch size 16, epochs 3, max length 512, precision bf16-mixed, loss ce, optimizer adamw, scheduler linear, seed 0

Input

  • Runs128 runs
  • Agentsnone showing
  • Treesnone showing
  • Projectsnone showing

Check

Every entry is checked on its way out; some are held for review.

Not shown on any run: data samples, predictions, raw logs, file paths, machine names

Delay

2 min, set

Watching the agents work, about 2 min behind.

One way

Nothing here reaches the agents.

This page

This page updates by itself.

visits counted with Google Analytics · choose

Runsheight is eval accuracyall recordings ▸ | finished× failed■ stopped○ lost● live ◂ older

1 of 5 finished, 0.8580, config 1 of 3 (learning rate 0.0001, batch 16), seed 0, Tue 17 Nov, 23:17 UTC

◂
Runs as a table
StartedStatusEval accuracyStepsConfig
Wed 18 Nov, 02:41 UTC| finished0.91801125 of 1125learning rate 0.00005, batch 16, seed 2 of 3
Wed 18 Nov, 02:22 UTC| finished0.89401125 of 1125learning rate 0.00005, batch 16, seed 1 of 3
Tue 17 Nov, 23:55 UTC| finished0.88801125 of 1125learning rate 0.00005, batch 16, seed 0 of 3
Tue 17 Nov, 23:36 UTC| finished0.86401125 of 1125learning rate 0.00002, batch 16, seed 0
Tue 17 Nov, 23:17 UTC| finished0.85801125 of 1125learning rate 0.0001, batch 16, seed 0

Plate

Model
answerdotai/ModernBERT-base, 149M parameters revision 8949b90
Data
Familiar decisions, shared draw, synthetic6000 train500 eval
Config
learning rate 0.0001, batch size 16, epochs 3, max length 512, precision bf16-mixed, loss ce, optimizer adamw, scheduler linear, seed 0
Hardware
Radeon 8060S, ROCm

Log

Since startMark
0.0 sphase starting
0.0 seval 1
296.0 seval 2
296.0 sepoch 1
592.0 seval 3
592.0 sepoch 2
888.0 seval 4
888.0 sepoch 3
13 h 34 min laterwent public

Readings

StepSecondsEval accuracy
00.00.1680
375296.00.7480
750592.00.8420
1125888.00.8580

Worked out

ReadoutValueWorked out from
Run complete100%1125 steps over 1125
Step time0.79 s296.0 s between epoch marks, over 375 steps an epoch. The reading interval holds the eval, so it is left out
Steps per minute76.060 s over the step time of 0.79 s
Epoch interval296.0 sMean of 2 gaps between epoch marks
Eval accuracy, as shown85.8%0.8580 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.69000.8580 minus 0.1680
Eval change+0.0160Eval 4 at 0.8580 minus eval 3 at 0.8420
Wrong answers71500 eval items minus 429 correct. 429 over 500 is the only count that gives 0.8580
Rank5 of 5By last eval accuracy, across this project's 5 runs with an eval
Best so far0.9180Highest last eval accuracy, run 5
Items per second20.31 step over 0.79 s, times batch size 16