CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Sat 21 Nov, 12:02 UTC
%
0.8404374 of 445 eval, 71 wrong
▲ +2.47 pts across 3 evals▲ +1.57 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Familiar decider: ModernBERT-base

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, 11:45 UTC
ran for
12:04
config
learning rate 0.0001, batch 16, seed 0

run2/6step1032/1032epoch3/3config2/5seed0lr0.0001batch16modelModernBERT-base 149Mrev8949b90items5503+445

Loss train 0.3133 ▼ −1.3714 lowest 0.0518 at step 820 run before, run 1 ▼ −0.0721 step 1030
Timebase
Range
Smoothing
Reference
104 readings, 5 min/div5 min/div2 min/div
0123400:0005:0010:00phase starting, 0.0 s to 20.3 sphase training, 20.3 s to 12:04epoch 1, 04:14epoch 2, 08:09epoch 3, 11:57trainingepoch 1epoch 21.6847Lowest last train loss across this project's 6 runs, run 6lowest last 0.1056, run 60.31330123400:0005:0010:00phase starting, 0.0 s to 20.3 sphase training, 20.3 s to 12:04epoch 1, 04:14epoch 2, 08:09epoch 3, 11:57trainingepoch 1epoch 21.6847Lowest last train loss across this project's 6 runs, run 6lowest last 0.1056, run 60.3133▼ −1.37140123400:0002:0004:0006:0008:0010:0012:00phase starting, 0.0 s to 20.3 sphase training, 20.3 s to 12:04epoch 1, 04:14epoch 2, 08:09epoch 3, 11:57trainingepoch 1epoch 21.6847Lowest last train loss across this project's 6 runs, run 6lowest last 0.1056, run 60.3133▼ −1.3714

Eval measures

Elapsed

231.5 s

epoch interval

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

0.67 s per step, 89.8 steps a minute

Eval accuracy by eval

+0.0157

▲ on the eval before

eval 1 0.8157, eval 3 0.8404

Against the other runs

5 of 5

by last eval accuracy

best so far 0.9101, run 6

Set-up

5503 train, 445 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

Learning rate

0.00e+0

▼ −1.64e-5 since first

On one axis

phase starting, 0.0 s to 20.3 sphase training, 20.3 s to 12:04E1E2E3traininglowest last 0.1056Train loss0.3133▼ −1.37144.000.001.68470.3133last ▼ −1.3714best 0.9101, run 6Last-value range, 5 runs, 0.8404 to 0.9101Eval accuracy0.8404▲ +0.02470.950.800.81570.8404last ▲ +0.0247Learning rate0.00e+0▼ −1.64e-51.0e-40.0e+01.64e-5last 0.00e+0Power, whole chip, CPU and GPU118 W▼ −1 W123 W113 W119 Wlast 118 WHottest, one board zone94 °C▲ +11 °C95 °C80 °C83 °Clast 94 °C00:0002:0004:0006:0008:0010:0012:00
GPUpower: whole chip, CPU and GPUtemperature: one board zone

GPU 0

Wwhole chip, CPU and GPU

118 W whole chip, CPU and GPU

use88%
memory, whole GPU1.1 GB

peak 12.5 GB, whole GPU

94 °C one board zone

history 0 to 122 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

2 of 6 finished, 0.8404, config 2 of 5 (learning rate 0.0001, batch 16), seed 0, Sat 21 Nov, 11:45 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Sat 21 Nov, 18:44 UTC| finished0.91011032 of 1032batch 16, seed 2 of 2
Sat 21 Nov, 18:26 UTC| finished0.85391032 of 1032batch 16, seed 1 of 2
Sat 21 Nov, 12:19 UTC| finished0.88311032 of 1032learning rate 0.00005, batch 16, seed 0
Sat 21 Nov, 12:02 UTC| finished0.84491032 of 1032learning rate 0.00002, batch 16, seed 0
Sat 21 Nov, 11:45 UTC| finished0.84041032 of 1032learning rate 0.0001, batch 16, seed 0
Fri 20 Nov, 07:59 UTC× failed-2180 of 7274batch 16, seed 0

Plate

Model
answerdotai/ModernBERT-base, 149M parameters revision 8949b90
Data
Familiar decisions, shared draw, plain-case intents, synthetic5503 train445 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
0.7 swent public
20.3 sphase training
249.5 seval 1
254.4 sepoch 1
487.3 seval 2
489.7 sepoch 2
717.3 sepoch 3
724.7 seval 3

Readings

106 rows as a table

rows 1 to 100 of 106

StepSecondsTrain lossEval accuracyLearning rate
1020.31.6847-1.64e-5
2026.92.0105-3.28e-5
3033.42.1738-4.92e-5
4040.11.5218-6.56e-5
5046.71.1594-8.20e-5
6053.41.6137-9.84e-5
7060.11.3108-9.91e-5
8066.71.5180-9.80e-5
9073.31.7061-9.70e-5
10080.01.4480-9.60e-5
11086.62.2476-9.50e-5
12093.11.2753-9.39e-5
13099.81.3235-9.29e-5
140106.41.4802-9.19e-5
150112.91.7503-9.08e-5
160119.50.7160-8.98e-5
170126.20.8655-8.88e-5
180133.00.6217-8.77e-5
190139.71.0613-8.67e-5
200146.40.9621-8.57e-5
210153.11.2408-8.47e-5
220159.70.8583-8.36e-5
230166.41.2969-8.26e-5
240173.10.5140-8.16e-5
250179.70.5243-8.05e-5
260186.31.0864-7.95e-5
270193.00.3835-7.85e-5
280199.50.9412-7.74e-5
290206.10.7537-7.64e-5
300212.80.2361-7.54e-5
310219.51.1583-7.44e-5
320226.10.9028-7.33e-5
330232.90.5781-7.23e-5
340239.50.4203-7.13e-5
344249.5-0.8157-
350254.40.5265-7.02e-5
360261.10.3068-6.92e-5
370267.80.9458-6.82e-5
380274.40.4648-6.71e-5
390281.20.4343-6.61e-5
400287.90.8405-6.51e-5
410294.50.5647-6.41e-5
420301.10.2304-6.30e-5
430307.80.5900-6.20e-5
440314.40.2197-6.10e-5
450321.10.6543-5.99e-5
460327.80.4502-5.89e-5
470334.40.3393-5.79e-5
480341.10.4933-5.68e-5
490347.80.2487-5.58e-5
500354.50.2007-5.48e-5
510361.20.7966-5.38e-5
520367.90.4558-5.27e-5
530374.60.5084-5.17e-5
540381.21.2515-5.07e-5
550387.90.7274-4.96e-5
560394.60.8607-4.86e-5
570401.30.4466-4.76e-5
580407.90.3232-4.65e-5
590414.60.3171-4.55e-5
600421.30.9747-4.45e-5
610428.00.2241-4.35e-5
620434.70.1535-4.24e-5
630441.30.8097-4.14e-5
640448.10.4939-4.04e-5
650454.70.3999-3.93e-5
660461.30.2735-3.83e-5
670468.00.3097-3.73e-5
680474.60.4832-3.63e-5
688487.3-0.8247-
690489.70.3437-3.52e-5
700496.20.4369-3.42e-5
710502.90.3297-3.32e-5
720509.50.1965-3.21e-5
730516.20.4642-3.11e-5
740522.90.1272-3.01e-5
750529.60.1505-2.90e-5
760536.40.8252-2.80e-5
770543.00.4739-2.70e-5
780549.70.2869-2.60e-5
790556.30.4999-2.49e-5
800563.00.1045-2.39e-5
810569.70.0888-2.29e-5
820576.40.0518-2.18e-5
830583.00.1869-2.08e-5
840589.60.1143-1.98e-5
850596.30.2450-1.87e-5
860602.90.1518-1.77e-5
870609.50.3147-1.67e-5
880616.20.2046-1.57e-5
890622.90.3214-1.46e-5
900629.60.6831-1.36e-5
910636.30.0855-1.26e-5
920642.90.2500-1.15e-5
930649.50.2348-1.05e-5
940656.20.1003-9.47e-6
950662.80.3495-8.44e-6
960669.40.0825-7.42e-6
970676.10.7717-6.39e-6
980682.70.0698-5.36e-6

GPU readings

106 rows as a table
StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
20065119.08.582.8
700100115.88.584.8
1200100117.510.686.8
1700100119.010.687.8
220067117.012.588.8
270065119.812.589.8
310062121.512.590.8
350065118.612.593.8
4000100116.512.590.8
4500100114.712.590.8
5000100120.412.590.8
5500100115.512.590.8
590064120.612.590.8
6400100115.812.590.8
680098117.012.590.8
7300100115.612.590.8
780067119.812.590.8
8200100116.612.590.8
8700100115.712.590.8
920066117.512.590.8
970063119.812.590.8
10200100118.212.590.8
1032088117.71.193.8

Worked out

ReadoutValueWorked out from
Run complete100%1032 steps over 1032
Step time0.67 s673.2 s over 1008 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute89.860 s over the step time of 0.67 s
Epoch interval231.5 sMean of 2 gaps between epoch marks
Train loss since first reading−1.37140.3133 at step 1032 minus 1.6847 at step 10
Lowest train loss0.0518Lowest of 104 readings, at step 820
Eval accuracy, as shown84.0%0.8404 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.02470.8404 minus 0.8157
Eval change+0.0157Eval 3 at 0.8404 minus eval 2 at 0.8247
Wrong answers71445 eval items minus 374 correct. 374 over 445 is the only count that gives 0.8404
Rank5 of 5By last eval accuracy, across this project's 5 runs with an eval
Best so far0.9101Highest last eval accuracy, run 6
Lowest last train loss0.1056Lowest last train loss across this project's 6 runs, run 6
Items per second24.01 step over 0.67 s, times batch size 16
Train loss against run 1−0.07210.0729 minus 0.1450 on run 1, both at step 1030
Learning rate since first reading−1.64e-50.00e+0 minus 1.64e-5
Peak memory per GPU, whole GPU12.5 GBHighest sensor reading of any GPU, GPU 0, counting every process on it