CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Sat 21 Nov, 13:54 UTC
%
0.9191409 of 445 eval, 36 wrong
▲ +7.87 pts across 3 evals▲ +2.70 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Familiar decider: ModernBERT-large

finishedRadeon 8060S

One of the small models that decide whether Familiar, a desktop companion, reacts to something happening on the PC, and how. Every candidate trains on the same 6,000 decision points from synthetic timelines. The encoders also get one run on the full pool.

started
Sat 21 Nov, 13:15 UTC
ran for
25:53
config
learning rate 0.00002, batch 16, seed 0

run9/12step1032/1032epoch3/3config7/9seed0lr0.00002batch16modelModernBERT-large 395Mrev45bb465items5503+445

Loss train 0.0743 ▼ −1.6645 lowest 0.0173 at step 720 run before, run 8 ▼ −0.3085 step 1032
Timebase
Range
Smoothing
Reference
104 readings, 10 min/div5 min/div5 min/div
012300:0010:0020:00phase starting, 0.0 s to 37.4 sphase training, 37.4 s to 25:53epoch 1, 09:02epoch 2, 17:28epoch 3, 25:38trainingepoch 1epoch 2−1.6645since first1.7388Lowest last train loss across this project's 12 runs, run 11lowest last 0.0007, run 110.0743012300:0005:0010:0015:0020:0025:00phase starting, 0.0 s to 37.4 sphase training, 37.4 s to 25:53epoch 1, 09:02epoch 2, 17:28epoch 3, 25:38trainingepoch 1epoch 21.7388Lowest last train loss across this project's 12 runs, run 11lowest last 0.0007, run 110.0743▼ −1.6645012300:0005:0010:0015:0020:0025:00phase starting, 0.0 s to 37.4 sphase training, 37.4 s to 25:53epoch 1, 09:02epoch 2, 17:28epoch 3, 25:38trainingepoch 1epoch 21.7388Lowest last train loss across this project's 12 runs, run 11lowest last 0.0007, run 110.0743▼ −1.6645

Eval measures

Elapsed

498.1 s

epoch interval

1032 of 1032 steps, 3 of 3 epochs, 100%

1.47 s per step, 40.9 steps a minute

Eval accuracy by eval

+0.0270

▲ on the eval before

eval 1 0.8404, eval 3 0.9191

Against the other runs

3 of 11

by last eval accuracy

best so far 0.9600, run 7

▲ +0.0315 against run 8

Set-up

5503 train, 445 eval

synthetic, learning rate 0.00002, batch size 16, epochs 3, max length 512, precision bf16-mixed, loss ce, optimizer adamw, scheduler linear, seed 0

Learning rate

0.00e+0

▼ −3.28e-6 since first

On one axis

phase starting, 0.0 s to 37.4 sphase training, 37.4 s to 25:53E1E2E3traininglowest last 0.0007Train loss0.0743▼ −1.66453.000.001.73880.0743last ▼ −1.6645best 0.9600, run 7Last-value range, 11 runs, 0.8854 to 0.9600Eval accuracy0.9191▲ +0.07871.000.800.84040.9191last ▲ +0.0787Learning rate0.00e+0▼ −3.28e-62.0e-50.0e+03.28e-6last 0.00e+0Power, whole chip, CPU and GPU118 W▼ −4 W125 W110 W122 Wlast 118 WHottest, one board zone93 °C▼ −1 °C94 °C90 °C94 °Clast 93 °C00:0005:0010:0015:0020:0025:00
GPUpower: whole chip, CPU and GPUtemperature: one board zone

GPU 0

Wwhole chip, CPU and GPU

118 W whole chip, CPU and GPU

use91%
memory, whole GPU2.1 GB

peak 23.7 GB, whole GPU

93 °C one board zone

history 0 to 124 W

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
1.000.851.000.85finished, eval accuracy 0.9000, config 1 of 9 (learning rate 0.0001, batch 16), seed 0, Wed 18 Nov, 12:57 UTCfinished, eval accuracy 0.9200, config 2 of 9 (learning rate 0.00002, batch 16), seed 0, Wed 18 Nov, 14:41 UTCfinished, eval accuracy 0.9020, config 3 of 9 (learning rate 0.00005, batch 16), seed 0, Wed 18 Nov, 16:41 UTCfinished, eval accuracy 0.9080, config 4 of 9 (batch 16), seed 1 of 2, Thu 19 Nov, 00:21 UTCfinished, eval accuracy 0.9160, config 4 of 9 (batch 16), seed 2 of 2, Thu 19 Nov, 02:14 UTCfailed at step 470 of 7274, eval accuracy -, config 5 of 9 (batch 16), seed 0, Fri 20 Nov, 13:18 UTCfinished, eval accuracy 0.9600, config 5 of 9 (batch 16), seed 0, Sat 21 Nov, 01:57 UTCfinished, eval accuracy 0.8876, config 6 of 9 (learning rate 0.0001, batch 16), seed 0, Sat 21 Nov, 12:36 UTCfinished, eval accuracy 0.9191, config 7 of 9 (learning rate 0.00002, batch 16), seed 0, Sat 21 Nov, 13:15 UTCfinished, eval accuracy 0.8944, config 8 of 9 (learning rate 0.00005, batch 16), seed 0, Sat 21 Nov, 13:54 UTCfinished, eval accuracy 0.8854, config 9 of 9 (batch 16), seed 1 of 2, Sat 21 Nov, 19:01 UTCfinished, eval accuracy 0.8921, config 9 of 9 (batch 16), seed 2 of 2, Sat 21 Nov, 19:40 UTC1.000.85finished, eval accuracy 0.9000, config 1 of 9 (learning rate 0.0001, batch 16), seed 0, Wed 18 Nov, 12:57 UTCfinished, eval accuracy 0.9200, config 2 of 9 (learning rate 0.00002, batch 16), seed 0, Wed 18 Nov, 14:41 UTCfinished, eval accuracy 0.9020, config 3 of 9 (learning rate 0.00005, batch 16), seed 0, Wed 18 Nov, 16:41 UTCfinished, eval accuracy 0.9080, config 4 of 9 (batch 16), seed 1 of 2, Thu 19 Nov, 00:21 UTCfinished, eval accuracy 0.9160, config 4 of 9 (batch 16), seed 2 of 2, Thu 19 Nov, 02:14 UTCfailed at step 470 of 7274, eval accuracy -, config 5 of 9 (batch 16), seed 0, Fri 20 Nov, 13:18 UTCfinished, eval accuracy 0.9600, config 5 of 9 (batch 16), seed 0, Sat 21 Nov, 01:57 UTCfinished, eval accuracy 0.8876, config 6 of 9 (learning rate 0.0001, batch 16), seed 0, Sat 21 Nov, 12:36 UTCfinished, eval accuracy 0.9191, config 7 of 9 (learning rate 0.00002, batch 16), seed 0, Sat 21 Nov, 13:15 UTCfinished, eval accuracy 0.8944, config 8 of 9 (learning rate 0.00005, batch 16), seed 0, Sat 21 Nov, 13:54 UTCfinished, eval accuracy 0.8854, config 9 of 9 (batch 16), seed 1 of 2, Sat 21 Nov, 19:01 UTCfinished, eval accuracy 0.8921, config 9 of 9 (batch 16), seed 2 of 2, Sat 21 Nov, 19:40 UTC

9 of 12 finished, 0.9191, config 7 of 9 (learning rate 0.00002, batch 16), seed 0, Sat 21 Nov, 13:15 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Sat 21 Nov, 19:40 UTC| finished0.89211032 of 1032batch 16, seed 2 of 2
Sat 21 Nov, 19:01 UTC| finished0.88541032 of 1032batch 16, seed 1 of 2
Sat 21 Nov, 13:54 UTC| finished0.89441032 of 1032learning rate 0.00005, batch 16, seed 0
Sat 21 Nov, 13:15 UTC| finished0.91911032 of 1032learning rate 0.00002, batch 16, seed 0
Sat 21 Nov, 12:36 UTC| finished0.88761032 of 1032learning rate 0.0001, batch 16, seed 0
Sat 21 Nov, 01:57 UTC| finished0.96007274 of 7274batch 16, seed 0
Fri 20 Nov, 13:18 UTC× failed-470 of 7274batch 16, seed 0
Thu 19 Nov, 02:14 UTC| finished0.91601125 of 1125batch 16, seed 2 of 2
Thu 19 Nov, 00:21 UTC| finished0.90801125 of 1125batch 16, seed 1 of 2
Wed 18 Nov, 16:41 UTC| finished0.90201125 of 1125learning rate 0.00005, batch 16, seed 0
Wed 18 Nov, 14:41 UTC| finished0.92001125 of 1125learning rate 0.00002, batch 16, seed 0
Wed 18 Nov, 12:57 UTC| finished0.90001125 of 1125learning rate 0.0001, batch 16, seed 0

Plate

Model
answerdotai/ModernBERT-large, 395M parameters revision 45bb465
Data
Familiar decisions, shared draw, plain-case intents, synthetic5503 train445 eval
Config
learning rate 0.00002, 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
9.8 swent public
37.4 sphase training
530.9 seval 1
542.2 sepoch 1
1042.8 seval 2
1048.7 sepoch 2
1538.4 sepoch 3
1553.7 seval 3

Readings

106 rows as a table

rows 1 to 100 of 106

StepSecondsTrain lossEval accuracyLearning rate
1037.41.7388-3.28e-6
2051.62.1138-6.56e-6
3065.92.0228-9.84e-6
4080.21.4796-1.31e-5
5094.61.0414-1.64e-5
60108.91.6153-1.97e-5
70123.31.3014-1.98e-5
80137.61.5696-1.96e-5
90151.81.5672-1.94e-5
100166.31.3809-1.92e-5
110180.51.9847-1.90e-5
120194.61.1705-1.88e-5
130208.91.1343-1.86e-5
140223.20.8859-1.84e-5
150237.41.0108-1.82e-5
160251.60.5061-1.80e-5
170266.00.9980-1.78e-5
180280.60.6667-1.75e-5
190295.00.9660-1.73e-5
200309.50.5689-1.71e-5
210323.80.7685-1.69e-5
220338.10.7224-1.67e-5
230352.41.1932-1.65e-5
240366.80.4270-1.63e-5
250381.10.4772-1.61e-5
260395.30.9958-1.59e-5
270409.60.4974-1.57e-5
280423.80.9709-1.55e-5
290438.00.4966-1.53e-5
300452.30.1501-1.51e-5
310466.81.1965-1.49e-5
320481.10.8709-1.47e-5
330495.50.4718-1.45e-5
340509.90.3877-1.43e-5
344530.9-0.8404-
350542.20.3997-1.40e-5
360556.60.1834-1.38e-5
370571.01.0838-1.36e-5
380585.30.3056-1.34e-5
390599.90.7000-1.32e-5
400614.11.1323-1.30e-5
410628.40.5023-1.28e-5
420642.60.1056-1.26e-5
430657.00.4486-1.24e-5
440671.30.2642-1.22e-5
450685.70.1895-1.20e-5
460700.00.3591-1.18e-5
470714.30.4763-1.16e-5
480728.80.2210-1.14e-5
490743.20.2725-1.12e-5
500757.50.0741-1.10e-5
510772.00.7527-1.08e-5
520786.30.3519-1.05e-5
530800.80.4276-1.03e-5
540815.01.0065-1.01e-5
550829.50.4349-9.93e-6
560843.80.4607-9.72e-6
570858.20.3276-9.52e-6
580872.30.3167-9.31e-6
590886.90.3601-9.10e-6
600901.20.7948-8.90e-6
610915.70.2393-8.69e-6
620930.00.2208-8.49e-6
630944.30.0952-8.28e-6
640958.80.3617-8.07e-6
650973.10.4048-7.87e-6
660987.30.1472-7.66e-6
6701001.70.3821-7.46e-6
6801016.10.2218-7.25e-6
6881042.8-0.8921-
6901048.70.2053-7.04e-6
7001062.90.1816-6.84e-6
7101077.20.2141-6.63e-6
7201091.40.0173-6.43e-6
7301105.90.2857-6.22e-6
7401120.30.0777-6.01e-6
7501134.70.1286-5.81e-6
7601149.10.6225-5.60e-6
7701163.50.2639-5.40e-6
7801177.80.1091-5.19e-6
7901192.10.2533-4.98e-6
8001206.40.0488-4.78e-6
8101220.90.1196-4.57e-6
8201235.20.1205-4.37e-6
8301249.40.1624-4.16e-6
8401263.70.1172-3.95e-6
8501278.10.2161-3.75e-6
8601292.20.0235-3.54e-6
8701306.50.1853-3.34e-6
8801320.90.2834-3.13e-6
8901335.20.0331-2.92e-6
9001349.70.5559-2.72e-6
9101364.00.0299-2.51e-6
9201378.30.1144-2.31e-6
9301392.60.1392-2.10e-6
9401406.90.1393-1.89e-6
9501421.20.2092-1.69e-6
9601435.40.0286-1.48e-6
9701449.70.2214-1.28e-6
9801464.00.0263-1.07e-6

GPU readings

106 rows as a table
StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
1000121.516.993.8
200100117.917.290.8
500100113.517.290.8
700100118.917.291.8
900100114.320.491.8
11000121.220.492.8
1400100119.220.492.8
1600100115.620.492.8
1800100121.320.492.8
2000100113.323.792.8
23000123.523.792.8
2500100115.123.792.8
27000116.223.792.8
2900100122.823.791.8
3200100113.723.792.8
3400100120.323.792.8
3500100118.923.793.8
3700100113.223.792.8
3900100118.923.792.8
4200100114.423.792.8
44000120.823.792.8
4600100116.723.793.8
48000120.523.793.8
5100100116.123.793.8
53000120.923.793.8
5500100118.223.793.8
5800100113.523.793.8
6000100117.423.793.8
62000120.423.793.8
6400100117.423.793.8
6700100113.223.793.8
680053119.323.793.8
7000100119.123.792.8
7200100115.323.792.8
7400100122.623.792.8
7700100114.123.792.8
79000123.223.792.8
8100100114.223.792.8
83000120.623.792.8
8600100119.223.792.8
8800100114.023.793.8
9000100118.223.792.8
9200100114.023.792.8
95000121.623.792.8
9700100118.023.791.8
9900100113.823.792.8
101000123.823.792.8
1032091117.92.192.8

Worked out

ReadoutValueWorked out from
Run complete100%1032 steps over 1032
Step time1.47 s1478.3 s over 1008 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute40.960 s over the step time of 1.47 s
Epoch interval498.1 sMean of 2 gaps between epoch marks
Train loss since first reading−1.66450.0743 at step 1032 minus 1.7388 at step 10
Lowest train loss0.0173Lowest of 104 readings, at step 720
Eval accuracy, as shown91.9%0.9191 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.07870.9191 minus 0.8404
Eval change+0.0270Eval 3 at 0.9191 minus eval 2 at 0.8921
Wrong answers36445 eval items minus 409 correct. 409 over 445 is the only count that gives 0.9191
Rank3 of 11By last eval accuracy, across this project's 11 runs with an eval
Best so far0.9600Highest last eval accuracy, run 7
Against the run before+0.03150.9191 minus 0.8876 on run 8
Lowest last train loss0.0007Lowest last train loss across this project's 12 runs, run 11
Items per second10.91 step over 1.47 s, times batch size 16
Train loss against run 8−0.30850.0743 minus 0.3828 on run 8, both at step 1032
Learning rate since first reading−3.28e-60.00e+0 minus 3.28e-6
Peak memory per GPU, whole GPU23.7 GBHighest sensor reading of any GPU, GPU 0, counting every process on it