CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Sat 21 Nov, 18:44 UTC
%
0.8539380 of 445 eval, 65 wrong
▲ +4.49 pts across 3 evals▲ +2.47 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, 18:26 UTC
ran for
12:05
config
batch 16, seed 1 of 2

run5/6step1032/1032epoch3/3config5/5seed1/2batch16modelModernBERT-base 149Mrev8949b90items5503+445

Loss train 0.1429 ▼ −1.5883 lowest 0.0295 at step 790 run before, run 4 ▼ −0.1074 step 1032
Timebase
Range
Smoothing
Reference
104 readings, 5 min/div5 min/div2 min/div
012300:0005:0010:00phase starting, 0.0 s to 20.2 sphase training, 20.2 s to 12:05epoch 1, 04:14epoch 2, 08:10epoch 3, 11:58trainingepoch 1epoch 2−1.5883since first1.7313Lowest last train loss across this project's 6 runs, run 6lowest last 0.1056, run 60.1429012300:0005:0010:00phase starting, 0.0 s to 20.2 sphase training, 20.2 s to 12:05epoch 1, 04:14epoch 2, 08:10epoch 3, 11:58trainingepoch 1epoch 21.7313Lowest last train loss across this project's 6 runs, run 6lowest last 0.1056, run 60.1429▼ −1.5883012300:0002:0004:0006:0008:0010:0012:00phase starting, 0.0 s to 20.2 sphase training, 20.2 s to 12:05epoch 1, 04:14epoch 2, 08:10epoch 3, 11:58trainingepoch 1epoch 21.7313Lowest last train loss across this project's 6 runs, run 6lowest last 0.1056, run 60.1429▼ −1.5883

Eval measures

Elapsed

231.8 s

epoch interval

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

0.67 s per step, 89.7 steps a minute

Eval accuracy by eval

+0.0247

▲ on the eval before

eval 1 0.8090, eval 3 0.8539

Against the other runs

3 of 5

by last eval accuracy

best so far 0.9101, run 6

▼ −0.0292 against run 4

Set-up

5503 train, 445 eval

synthetic, batch size 16, epochs 3, max length 512, precision bf16-mixed, loss ce, optimizer adamw, scheduler linear, seed 1

Seed spread, observed

0.0562
0.850.95

2 finished seeds

Learning rate

0.00e+0

▼ −8.20e-6 since first

On one axis

phase starting, 0.0 s to 20.2 sphase training, 20.2 s to 12:05E1E2E3traininglowest last 0.1056Train loss0.1429▼ −1.58833.000.001.73130.1429last ▼ −1.5883best 0.9101, run 6Last-value range, 5 runs, 0.8404 to 0.9101Eval accuracy0.8539▲ +0.04490.950.800.80900.8539last ▲ +0.0449Learning rate0.00e+0▼ −8.20e-66.0e-50.0e+08.20e-6last 0.00e+0Power, whole chip, CPU and GPU118 W▲ +2 W123 W113 W116 Wlast 118 WHottest, one board zone93 °C▲ +5 °C94 °C86 °C88 °Clast 93 °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

use89%
memory, whole GPU1.1 GB

peak 13.5 GB, whole GPU

93 °C one board zone

history 0 to 121 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

5 of 6 finished, 0.8539, config 5 of 5 (batch 16), seed 1 of 2, Sat 21 Nov, 18:26 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
batch size 16, epochs 3, max length 512, precision bf16-mixed, loss ce, optimizer adamw, scheduler linear, seed 1
Hardware
Radeon 8060S, ROCm

Log

Since startMark
0.0 sphase starting
6.0 swent public
20.2 sphase training
250.2 seval 1
254.7 sepoch 1
487.8 seval 2
490.1 sepoch 2
718.4 sepoch 3
725.7 seval 3

Seeds

SeedStatusEval accuracy
seed 1| finished0.8539
seed 2| finished0.9101

observed spread 0.0562, 2 finished seeds

Readings

106 rows as a table

rows 1 to 100 of 106

StepSecondsTrain lossEval accuracyLearning rate
1020.21.7313-8.20e-6
2026.91.7683-1.64e-5
3033.41.3817-2.46e-5
4040.21.3894-3.28e-5
5046.91.2167-4.10e-5
6053.51.3130-4.92e-5
7060.21.8195-4.95e-5
8066.81.8575-4.90e-5
9073.40.9052-4.85e-5
10080.11.0718-4.80e-5
11086.61.7046-4.75e-5
12093.31.6964-4.70e-5
130100.01.1191-4.64e-5
140106.81.6236-4.59e-5
150113.51.5715-4.54e-5
160120.10.8253-4.49e-5
170126.80.6046-4.44e-5
180133.41.4616-4.39e-5
190140.10.8075-4.34e-5
200146.80.7959-4.28e-5
210153.60.9545-4.23e-5
220160.20.8353-4.18e-5
230166.80.6003-4.13e-5
240173.40.7536-4.08e-5
250180.10.7486-4.03e-5
260186.80.3360-3.98e-5
270193.50.9191-3.92e-5
280200.20.6511-3.87e-5
290206.90.2605-3.82e-5
300213.60.2823-3.77e-5
310220.21.0826-3.72e-5
320226.90.2010-3.67e-5
330233.60.8808-3.61e-5
340240.20.4357-3.56e-5
344250.2-0.8090-
350254.70.4076-3.51e-5
360261.40.3561-3.46e-5
370268.10.3772-3.41e-5
380274.80.1543-3.36e-5
390281.50.0828-3.31e-5
400288.10.2395-3.25e-5
410294.70.3381-3.20e-5
420301.50.4326-3.15e-5
430308.20.4233-3.10e-5
440314.90.3267-3.05e-5
450321.60.6294-3.00e-5
460328.20.5600-2.95e-5
470334.90.7072-2.89e-5
480341.61.1913-2.84e-5
490348.20.4972-2.79e-5
500354.90.5966-2.74e-5
510361.50.3148-2.69e-5
520368.20.1638-2.64e-5
530374.81.2241-2.58e-5
540381.50.6485-2.53e-5
550388.20.4908-2.48e-5
560394.90.6213-2.43e-5
570401.60.2546-2.38e-5
580408.30.2109-2.33e-5
590414.90.8580-2.28e-5
600421.70.1985-2.22e-5
610428.40.7220-2.17e-5
620435.10.1678-2.12e-5
630441.80.0672-2.07e-5
640448.50.2225-2.02e-5
650455.10.5806-1.97e-5
660461.80.5230-1.92e-5
670468.40.2753-1.86e-5
680475.10.3184-1.81e-5
688487.8-0.8292-
690490.10.7566-1.76e-5
700496.50.1877-1.71e-5
710503.20.0754-1.66e-5
720509.80.2632-1.61e-5
730516.40.3088-1.56e-5
740523.10.5010-1.50e-5
750529.80.3768-1.45e-5
760536.50.1720-1.40e-5
770543.20.5734-1.35e-5
780549.90.0824-1.30e-5
790556.70.0295-1.25e-5
800563.40.3184-1.19e-5
810570.10.7231-1.14e-5
820576.70.1352-1.09e-5
830583.30.1680-1.04e-5
840590.00.1303-9.89e-6
850596.60.2236-9.37e-6
860603.40.1744-8.86e-6
870610.00.3037-8.34e-6
880616.70.3356-7.83e-6
890623.40.0618-7.31e-6
900630.10.2885-6.80e-6
910636.70.2054-6.28e-6
920643.40.1945-5.77e-6
930650.20.0743-5.25e-6
940656.90.2947-4.74e-6
950663.70.2802-4.22e-6
960670.40.3430-3.71e-6
970677.10.2210-3.19e-6
980683.70.3076-2.68e-6

GPU readings

106 rows as a table
StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
200100115.89.587.8
70065119.213.587.8
120064118.413.587.8
1700100116.113.587.8
22001119.713.588.8
2600100120.213.589.8
3100100114.713.589.8
350065117.813.592.8
4000100119.013.589.8
450065118.713.589.8
500064120.213.589.8
5400100116.413.589.8
590063119.013.589.8
6400100115.113.590.8
680095117.213.590.8
7300100119.313.590.8
780063118.413.590.8
8200100118.713.590.8
8700100114.813.590.8
920064120.913.590.8
970066117.513.590.8
1020064120.713.590.8
1032089117.71.192.8

Worked out

ReadoutValueWorked out from
Run complete100%1032 steps over 1032
Step time0.67 s674.3 s over 1008 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute89.760 s over the step time of 0.67 s
Epoch interval231.8 sMean of 2 gaps between epoch marks
Train loss since first reading−1.58830.1429 at step 1032 minus 1.7313 at step 10
Lowest train loss0.0295Lowest of 104 readings, at step 790
Eval accuracy, as shown85.4%0.8539 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.04490.8539 minus 0.8090
Eval change+0.0247Eval 3 at 0.8539 minus eval 2 at 0.8292
Wrong answers65445 eval items minus 380 correct. 380 over 445 is the only count that gives 0.8539
Rank3 of 5By last eval accuracy, across this project's 5 runs with an eval
Best so far0.9101Highest last eval accuracy, run 6
Against the run before−0.02920.8539 minus 0.8831 on run 4
Lowest last train loss0.1056Lowest last train loss across this project's 6 runs, run 6
Seed spread, observed0.0562High 0.9101 minus low 0.8539 across 2 finished seeds
Items per second23.91 step over 0.67 s, times batch size 16
Train loss against run 4−0.10740.1429 minus 0.2503 on run 4, both at step 1032
Learning rate since first reading−8.20e-60.00e+0 minus 8.20e-6
Peak memory per GPU, whole GPU13.5 GBHighest sensor reading of any GPU, GPU 0, counting every process on it