CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Sun 22 Nov, 10:05 UTC
%
0.8860443 of 500 eval, 57 wrong
▲ +6.40 pts across 3 evals▲ +3.40 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Familiar decider: ettin-encoder-68m

finishedRadeon 8060S

One of the small models that decide whether Familiar, a desktop companion, reacts to something on the PC, and how. Trained on the same 6,000 synthetic decision points as every candidate, two ways: the reaction itself, and the plain-case intent that rules adjust for her traits.

started
Sun 22 Nov, 09:56 UTC
ran for
07:39
config
learning rate 0.00005, batch 16, seed 0

run3/10step1125/1125epoch3/3config3/8seed0lr0.00005batch16modelettin-encoder-68m 68Mrevac19ae4items6000+500

Loss train 0.1353 ▼ −1.3023 lowest 0.0554 at step 1000 run before, run 2 ▼ −0.1701 step 1125
Timebase
Range
Smoothing
Reference
113 readings, 2 min/div2 min/div1 min/div
00.511.522.500:0002:0004:0006:00phase starting, 0.0 s to 15.8 sphase training, 15.8 s to 07:39epoch 1, 02:42epoch 2, 05:04epoch 3, 07:34trainingepoch 1epoch 2−1.3023since first1.4377Lowest last train loss across this project's 10 runs, run 9lowest last 0.0119, run 90.135300.511.522.500:0002:0004:0006:00phase starting, 0.0 s to 15.8 sphase training, 15.8 s to 07:39epoch 1, 02:42epoch 2, 05:04epoch 3, 07:34trainingepoch 1epoch 21.4377Lowest last train loss across this project's 10 runs, run 9lowest last 0.0119, run 90.1353▼ −1.302300.511.522.500:0001:0002:0003:0004:0005:0006:0007:00phase starting, 0.0 s to 15.8 sphase training, 15.8 s to 07:39epoch 1, 02:42epoch 2, 05:04epoch 3, 07:34trainingepoch 1epoch 21.4377Lowest last train loss across this project's 10 runs, run 9lowest last 0.0119, run 90.1353▼ −1.3023

Eval measures

Elapsed

145.8 s

epoch interval

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

0.39 s per step, 152.8 steps a minute

Eval accuracy by eval

+0.0340

▲ on the eval before

eval 1 0.8220, eval 3 0.8860

Against the other runs

4 of 10

by last eval accuracy

best so far 0.9120, run 8

▲ +0.0060 against run 2

Set-up

6000 train, 500 eval

synthetic, learning rate 0.00005, batch size 16, epochs 3, max length 512, precision bf16-mixed, loss ce, optimizer adamw, scheduler linear, seed 0

Learning rate

0.00e+0

▼ −7.46e-6 since first

On one axis

phase starting, 0.0 s to 15.8 sphase training, 15.8 s to 07:39E1E2E3traininglowest last 0.0119Train loss0.1353▼ −1.30232.000.001.43770.1353last ▼ −1.3023best 0.9120, run 8Last-value range, 10 runs, 0.8584 to 0.9120Eval accuracy0.8860▲ +0.06400.950.800.82200.8860last ▲ +0.0640Learning rate0.00e+0▼ −7.46e-66.0e-50.0e+07.46e-6last 0.00e+0Power, whole chip, CPU and GPU116 W▼ −3 W120 W114 W119 Wlast 116 WHottest, one board zone92 °C▼ −3 °C95 °C80 °C95 °Clast 92 °C00:0001:0002:0003:0004:0005:0006:0007:00
GPUpower: whole chip, CPU and GPUtemperature: one board zone

GPU 0

Wwhole chip, CPU and GPU

116 W whole chip, CPU and GPU

use85%
memory, whole GPU0.8 GB

peak 9.7 GB, whole GPU

92 °C one board zone

history 0 to 120 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
0.950.850.950.85finished, eval accuracy 0.8780, config 1 of 8 (learning rate 0.0001, batch 16), seed 0, Sun 22 Nov, 09:36 UTCfinished, eval accuracy 0.8800, config 2 of 8 (learning rate 0.00002, batch 16), seed 0, Sun 22 Nov, 09:46 UTCfinished, eval accuracy 0.8860, config 3 of 8 (learning rate 0.00005, batch 16), seed 0, Sun 22 Nov, 09:56 UTCfinished, eval accuracy 0.8831, config 4 of 8 (learning rate 0.0001, batch 16), seed 0, Sun 22 Nov, 10:05 UTCfinished, eval accuracy 0.9034, config 5 of 8 (learning rate 0.00002, batch 16), seed 0, Sun 22 Nov, 10:14 UTCfinished, eval accuracy 0.8809, config 6 of 8 (learning rate 0.00005, batch 16), seed 0, Sun 22 Nov, 10:24 UTCfinished, eval accuracy 0.9060, config 7 of 8 (batch 16), seed 1 of 2, Sun 22 Nov, 12:18 UTCfinished, eval accuracy 0.9120, config 7 of 8 (batch 16), seed 2 of 2, Sun 22 Nov, 12:28 UTCfinished, eval accuracy 0.8831, config 8 of 8 (batch 16), seed 1 of 2, Sun 22 Nov, 12:38 UTCfinished, eval accuracy 0.8584, config 8 of 8 (batch 16), seed 2 of 2, Sun 22 Nov, 12:47 UTC0.950.85finished, eval accuracy 0.8780, config 1 of 8 (learning rate 0.0001, batch 16), seed 0, Sun 22 Nov, 09:36 UTCfinished, eval accuracy 0.8800, config 2 of 8 (learning rate 0.00002, batch 16), seed 0, Sun 22 Nov, 09:46 UTCfinished, eval accuracy 0.8860, config 3 of 8 (learning rate 0.00005, batch 16), seed 0, Sun 22 Nov, 09:56 UTCfinished, eval accuracy 0.8831, config 4 of 8 (learning rate 0.0001, batch 16), seed 0, Sun 22 Nov, 10:05 UTCfinished, eval accuracy 0.9034, config 5 of 8 (learning rate 0.00002, batch 16), seed 0, Sun 22 Nov, 10:14 UTCfinished, eval accuracy 0.8809, config 6 of 8 (learning rate 0.00005, batch 16), seed 0, Sun 22 Nov, 10:24 UTCfinished, eval accuracy 0.9060, config 7 of 8 (batch 16), seed 1 of 2, Sun 22 Nov, 12:18 UTCfinished, eval accuracy 0.9120, config 7 of 8 (batch 16), seed 2 of 2, Sun 22 Nov, 12:28 UTCfinished, eval accuracy 0.8831, config 8 of 8 (batch 16), seed 1 of 2, Sun 22 Nov, 12:38 UTCfinished, eval accuracy 0.8584, config 8 of 8 (batch 16), seed 2 of 2, Sun 22 Nov, 12:47 UTC

3 of 10 finished, 0.8860, config 3 of 8 (learning rate 0.00005, batch 16), seed 0, Sun 22 Nov, 09:56 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Sun 22 Nov, 12:47 UTC| finished0.85841032 of 1032batch 16, seed 2 of 2
Sun 22 Nov, 12:38 UTC| finished0.88311032 of 1032batch 16, seed 1 of 2
Sun 22 Nov, 12:28 UTC| finished0.91201125 of 1125batch 16, seed 2 of 2
Sun 22 Nov, 12:18 UTC| finished0.90601125 of 1125batch 16, seed 1 of 2
Sun 22 Nov, 10:24 UTC| finished0.88091032 of 1032learning rate 0.00005, batch 16, seed 0
Sun 22 Nov, 10:14 UTC| finished0.90341032 of 1032learning rate 0.00002, batch 16, seed 0
Sun 22 Nov, 10:05 UTC| finished0.88311032 of 1032learning rate 0.0001, batch 16, seed 0
Sun 22 Nov, 09:56 UTC| finished0.88601125 of 1125learning rate 0.00005, batch 16, seed 0
Sun 22 Nov, 09:46 UTC| finished0.88001125 of 1125learning rate 0.00002, batch 16, seed 0
Sun 22 Nov, 09:36 UTC| finished0.87801125 of 1125learning rate 0.0001, batch 16, seed 0

Plate

Model
jhu-clsp/ettin-encoder-68m, 68M parameters revision ac19ae4
Data
Familiar decisions, shared draw, synthetic6000 train500 eval
Config
learning rate 0.00005, 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
15.8 sphase training
160.3 seval 1
162.6 sepoch 1
297.4 swent public
304.5 sepoch 2
309.5 seval 2
454.2 sepoch 3
459.2 seval 3

Readings

114 rows as a table

rows 1 to 100 of 114

StepSecondsTrain lossEval accuracyLearning rate
1015.81.4377-7.46e-6
2019.51.4666-1.49e-5
3023.31.0765-2.24e-5
4027.11.3382-2.99e-5
5030.81.9825-3.73e-5
6034.61.8504-4.48e-5
7038.41.3316-4.99e-5
8042.21.3643-4.94e-5
9046.01.6849-4.89e-5
10049.81.5719-4.84e-5
11053.61.3355-4.80e-5
12057.60.7574-4.75e-5
13061.31.4710-4.70e-5
14065.21.2447-4.66e-5
15068.90.9370-4.61e-5
16072.71.1545-4.56e-5
17076.61.3139-4.51e-5
18080.30.8567-4.47e-5
19084.21.0699-4.42e-5
20088.10.6961-4.37e-5
21091.90.5742-4.32e-5
22095.80.8418-4.28e-5
23099.60.7681-4.23e-5
240103.50.3482-4.18e-5
250107.30.8643-4.14e-5
260111.10.4554-4.09e-5
270114.90.6506-4.04e-5
280118.81.2216-3.99e-5
290122.71.2260-3.95e-5
300126.50.8648-3.90e-5
310130.40.3179-3.85e-5
320134.20.2471-3.80e-5
330138.10.2580-3.76e-5
340141.90.6937-3.71e-5
350145.70.4039-3.66e-5
360149.60.9436-3.62e-5
370153.40.5153-3.57e-5
375160.3-0.8220-
380162.60.5757-3.52e-5
390166.40.1919-3.47e-5
400170.30.7225-3.43e-5
410174.10.6030-3.38e-5
420178.00.4859-3.33e-5
430181.80.5476-3.28e-5
440185.60.2752-3.24e-5
450189.40.8192-3.19e-5
460193.20.4731-3.14e-5
470197.00.1729-3.10e-5
480200.80.4617-3.05e-5
490204.70.2594-3.00e-5
500208.50.4703-2.95e-5
510212.30.3925-2.91e-5
520216.20.6974-2.86e-5
530220.10.3290-2.81e-5
540223.90.5559-2.76e-5
550227.81.2448-2.72e-5
560231.60.1541-2.67e-5
570235.50.1877-2.62e-5
580239.40.4551-2.58e-5
590243.20.2992-2.53e-5
600247.00.6182-2.48e-5
610250.90.3686-2.43e-5
620254.70.4238-2.39e-5
630258.50.2630-2.34e-5
640262.40.7414-2.29e-5
650266.20.6069-2.24e-5
660270.10.1494-2.20e-5
670273.90.2313-2.15e-5
680277.70.3088-2.10e-5
690281.50.2041-2.06e-5
700285.40.3828-2.01e-5
710289.20.4724-1.96e-5
720293.00.6275-1.91e-5
730296.90.2323-1.87e-5
740300.60.4849-1.82e-5
750309.50.63900.85201.77e-5
760313.90.1806-1.72e-5
770317.80.2650-1.68e-5
780321.60.4396-1.63e-5
790325.50.2732-1.58e-5
800329.30.2661-1.54e-5
810333.20.4574-1.49e-5
820337.10.4368-1.44e-5
830340.90.1397-1.39e-5
840344.70.4390-1.35e-5
850348.50.3289-1.30e-5
860352.20.1636-1.25e-5
870356.10.1756-1.21e-5
880359.90.2406-1.16e-5
890363.80.3916-1.11e-5
900367.50.3445-1.06e-5
910371.40.2704-1.02e-5
920375.20.1127-9.69e-6
930379.10.2588-9.22e-6
940382.90.4778-8.74e-6
950386.80.1205-8.27e-6
960390.60.1885-7.80e-6
970394.50.1397-7.33e-6
980398.40.1682-6.85e-6
990402.20.5182-6.38e-6

GPU readings

114 rows as a table
StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
20069118.98.494.8
1100100116.89.784.8
1900100115.59.785.8
280075117.79.786.8
360073118.89.787.8
4300100116.19.787.8
5200100116.19.788.8
6000100116.89.788.8
6900100117.39.788.8
7600100116.49.791.8
840072119.79.788.8
9300100117.39.789.8
1010072119.19.789.8
1090073120.09.789.8
1125085116.30.891.8

Worked out

ReadoutValueWorked out from
Run complete100%1125 steps over 1125
Step time0.39 s429.9 s over 1095 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute152.860 s over the step time of 0.39 s
Epoch interval145.8 sMean of 2 gaps between epoch marks
Train loss since first reading−1.30230.1353 at step 1125 minus 1.4377 at step 10
Lowest train loss0.0554Lowest of 113 readings, at step 1000
Eval accuracy, as shown88.6%0.8860 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.06400.8860 minus 0.8220
Eval change+0.0340Eval 3 at 0.8860 minus eval 2 at 0.8520
Wrong answers57500 eval items minus 443 correct. 443 over 500 is the only count that gives 0.8860
Rank4 of 10By last eval accuracy, across this project's 10 runs with an eval
Best so far0.9120Highest last eval accuracy, run 8
Against the run before+0.00600.8860 minus 0.8800 on run 2
Lowest last train loss0.0119Lowest last train loss across this project's 10 runs, run 9
Items per second40.81 step over 0.39 s, times batch size 16
Train loss against run 2−0.17010.1353 minus 0.3055 on run 2, both at step 1125
Learning rate since first reading−7.46e-60.00e+0 minus 7.46e-6
Peak memory per GPU, whole GPU9.7 GBHighest sensor reading of any GPU, GPU 0, counting every process on it