CITADEL
StandbyNothing is live right now.
Eval accuracy
%
no eval

Familiar decider: ModernBERT-base

failedRadeon 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
Fri 20 Nov, 07:59 UTC
ran for
135:42
config
batch 16, seed 0

run1/6step2180/7274epoch0/2config1/5seed0batch16modelModernBERT-base 149Mrev8949b90items58183+500

Loss train 0.1946 ▼ −2.9760 lowest 0.0180 at step 1830
Timebase
Range
Smoothing
218 readings, 120 min/div120 min/div60 min/div
0123400:00120:00240:00360:00phase starting, 0.0 s to 68.8 sphase training, 68.8 s to 135:42training−2.9760since first3.1707Lowest last train loss across this project's 6 runs, run 6lowest last 0.1056, run 6failed here7274 steps0.19460123400:00120:00240:00360:00phase starting, 0.0 s to 68.8 sphase training, 68.8 s to 135:42training3.1707Lowest last train loss across this project's 6 runs, run 6lowest last 0.1056, run 6failed here7274 steps0.1946▼ −2.97600123400:0060:00120:00180:00240:00300:00360:00420:00phase starting, 0.0 s to 68.8 sphase training, 68.8 s to 135:42training3.1707Lowest last train loss across this project's 6 runs, run 6lowest last 0.1056, run 6failed here7274 steps0.1946▼ −2.9760

Elapsed

8142.3 sto last reading

8142.3 s to last reading

Set-up

58183 train, 500 eval

synthetic, batch size 16, epochs 2, max length 512, precision bf16-mixed, loss ce, optimizer adamw, scheduler linear, seed 0

Progress

30.0%

2180 of 7274 steps, 0 of 2 epochs

Pace

3.72 s

per step

16.1 steps a minute

4.3 items a second

Learning rate

3.72e-5

▲ +3.61e-5 since first

On one axis

phase starting, 0.0 s to 68.8 sphase training, 68.8 s to 135:42traininglowest last 0.1056Train loss0.1946▼ −2.97604.000.003.17070.1946last ▼ −2.9760Learning rate3.72e-5▲ +3.61e-56.0e-50.0e+01.15e-63.72e-5last ▲ +3.61e-5Power, whole chip, CPU and GPU49 W▼ −4 W150 W0 W53 W49 Wlast ▼ −4 WHottest, one board zone97 °C▲ +28 °C100 °C60 °C69 °C97 °Clast ▲ +28 °C00:0060:00120:00180:00240:00300:00360:00420:00
GPUpower: whole chip, CPU and GPUtemperature: one board zone

GPU 0

Wwhole chip, CPU and GPU

49 W whole chip, CPU and GPU

use0%
memory, whole GPU1.1 GB

peak 25.0 GB, whole GPU

97 °C one board zone

history 0 to 133 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 ◂ older

1 of 6 failed at step 2180 of 7274, -, config 1 of 5 (batch 16), seed 0, Fri 20 Nov, 07:59 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, full pool, synthetic58183 train500 eval
Config
batch size 16, epochs 2, 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
9.7 swent public
68.8 sphase training

Readings

218 rows as a table

rows 1 to 100 of 218

StepSecondsTrain lossLearning rate
1068.83.17071.15e-6
20104.82.50932.29e-6
30160.51.52673.44e-6
40213.01.54574.59e-6
50259.51.46925.73e-6
60287.11.49496.88e-6
70332.12.18388.03e-6
80369.91.77399.17e-6
90444.31.59151.03e-5
100481.01.48491.15e-5
110525.21.67361.26e-5
120575.41.46621.38e-5
130611.60.95471.49e-5
140656.61.12881.61e-5
150684.21.01301.72e-5
160737.61.22791.83e-5
170806.31.65891.95e-5
180859.42.09562.06e-5
190898.31.57962.18e-5
200926.91.46352.29e-5
210963.31.55392.41e-5
2201028.71.02142.52e-5
2301072.21.17182.64e-5
2401119.40.47792.75e-5
2501145.91.08452.87e-5
2601173.41.33832.98e-5
2701221.01.03183.10e-5
2801249.80.90003.21e-5
2901286.81.29293.33e-5
3001332.21.84293.44e-5
3101377.31.21413.56e-5
3201403.50.91283.67e-5
3301421.21.31793.78e-5
3401454.90.93873.90e-5
3501482.00.56044.01e-5
3601537.60.72594.13e-5
3701555.20.76914.24e-5
3801583.50.86164.36e-5
3901610.80.61924.47e-5
4001664.41.84174.59e-5
4101739.10.78144.70e-5
4201756.70.91754.82e-5
4301812.31.14054.93e-5
4401850.00.96985.00e-5
4501885.70.36064.99e-5
4601912.60.57434.98e-5
4701958.00.35344.98e-5
4801984.40.74304.97e-5
4902032.70.40114.96e-5
5002067.81.03654.95e-5
5102076.30.45784.95e-5
5202122.30.80474.94e-5
5302168.00.25924.93e-5
5402194.30.63844.92e-5
5502249.10.68524.92e-5
5602284.80.72094.91e-5
5702320.90.62874.90e-5
5802367.30.93004.89e-5
5902419.40.49474.89e-5
6002436.20.40584.88e-5
6102480.40.66164.87e-5
6202517.50.70754.87e-5
6302553.20.21644.86e-5
6402580.20.09694.85e-5
6502600.70.74504.84e-5
6602637.10.30304.84e-5
6702673.10.48944.83e-5
6802699.80.12834.82e-5
6902754.60.48874.81e-5
7002791.80.45984.81e-5
7102820.00.59904.80e-5
7202846.50.30814.79e-5
7302883.11.27614.79e-5
7402917.60.60804.78e-5
7502935.40.97294.77e-5
7603007.90.16864.76e-5
7703035.20.28384.76e-5
7803054.31.07724.75e-5
7903090.30.44964.74e-5
8003146.80.42284.73e-5
8103199.60.19184.73e-5
8203239.80.55494.72e-5
8303273.70.47224.71e-5
8403302.20.44244.70e-5
8503337.70.46454.70e-5
8603393.00.70434.69e-5
8703420.20.76964.68e-5
8803449.40.33534.68e-5
8903477.10.47194.67e-5
9003504.40.80754.66e-5
9103553.30.11184.65e-5
9203598.50.32804.65e-5
9303626.90.36344.64e-5
9403664.60.30654.63e-5
9503691.81.13264.62e-5
9603767.10.36344.62e-5
9703802.70.41304.61e-5
9803832.50.46714.60e-5
9903850.70.60524.59e-5
10003897.20.85604.59e-5

GPU readings

218 rows as a table

rows 1 to 100 of 197

StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
100052.611.168.8
200053.311.168.8
300053.111.170.8
400053.012.969.8
500053.114.968.8
600051.714.967.8
700054.116.968.8
8002450.616.968.8
900051.616.967.8
1000052.616.967.8
1100054.916.967.8
1200052.416.969.8
1300053.216.967.8
14002353.416.967.8
15001254.416.968.8
1600056.016.970.8
17001353.116.968.8
1800052.616.967.8
1900150.916.968.8
2000053.316.967.8
2100052.816.968.8
2200053.316.968.8
2300053.416.967.8
24002054.416.968.8
2500053.116.971.8
2600054.116.969.8
2700053.116.967.8
2800055.416.968.8
2900053.216.969.8
3000053.916.967.8
3100053.416.966.8
3300053.216.968.8
3400053.016.968.8
3500052.616.968.8
3600053.116.969.8
3800054.416.968.8
3900054.616.967.8
4000055.516.970.8
4100053.016.967.8
4200053.716.968.8
4300053.216.969.8
4400053.216.967.8
46001654.416.969.8
4700054.016.968.8
4800053.616.970.8
4900053.416.967.8
50002452.616.966.8
5100052.616.967.8
5200053.216.967.8
5300053.316.969.8
5400054.016.970.8
5500055.016.968.8
5600053.316.968.8
5700052.616.971.8
5800053.216.968.8
5900252.916.967.8
6000053.416.968.8
6100052.516.968.8
6200054.116.967.8
6300052.716.966.8
6400154.416.968.8
6500054.016.968.8
6600054.216.967.8
6700052.716.968.8
6800052.816.968.8
69002254.316.969.8
7100053.316.968.8
7200053.516.967.8
7300053.416.968.8
7400053.616.965.8
7500051.216.968.8
76001152.316.969.8
7700054.916.968.8
7800053.016.968.8
7900054.516.969.8
8000053.216.968.8
8100052.816.966.8
8200053.316.968.8
8300053.016.967.8
8400054.916.968.8
8500053.116.968.8
8600053.716.968.8
8700053.116.966.8
8800053.716.968.8
9000051.518.967.8
91001854.218.968.8
9200053.418.967.8
9300052.618.967.8
9400054.818.968.8
9500052.918.968.8
9600053.118.968.8
9700053.718.969.8
9800253.018.968.8
9900053.318.966.8
10000654.921.068.8
10100054.221.068.8
102001554.221.069.8
10400254.421.069.8
10500053.921.069.8
10700054.121.070.8

Worked out

ReadoutValueWorked out from
Run complete30.0%2180 steps over 7274
Step time3.72 s8073.5 s over 2170 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute16.160 s over the step time of 3.72 s
Train loss since first reading−2.97600.1946 at step 2180 minus 3.1707 at step 10
Lowest train loss0.0180Lowest of 218 readings, at step 1830
Lowest last train loss0.1056Lowest last train loss across this project's 6 runs, run 6
Items per second4.31 step over 3.72 s, times batch size 16
Learning rate since first reading+3.61e-53.72e-5 minus 1.15e-6
Peak memory per GPU, whole GPU25.0 GBHighest sensor reading of any GPU, GPU 0, counting every process on it