CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Sun 22 Nov, 13:14 UTC
%
0.9140457 of 500 eval, 43 wrong
▲ +3.60 pts across 3 evals▲ +2.00 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Familiar decider: decider-nano's encoder

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, 12:56 UTC
ran for
14:00
config
batch 16, seed 1 of 2

run7/10step1125/1125epoch3/3config7/8seed1/2batch16modeltasksource-decider-nano 149Mrev2c7c01fitems6000+500

Loss train 0.0018 ▼ −2.9213 lowest 0.0018 at step 1125 run before, run 6
Timebase
Range
Smoothing
Reference
113 readings, 5 min/div5 min/div2 min/div
0123400:0005:0010:00phase starting, 0.0 s to 22.9 sphase training, 22.9 s to 14:00epoch 1, 04:54epoch 2, 09:16epoch 3, 13:51trainingepoch 1epoch 2−2.9213since first2.9231Lowest last train loss across this project's 10 runs, run 7lowest last 0.0018, run 70.00180123400:0005:0010:00phase starting, 0.0 s to 22.9 sphase training, 22.9 s to 14:00epoch 1, 04:54epoch 2, 09:16epoch 3, 13:51trainingepoch 1epoch 22.9231Lowest last train loss across this project's 10 runs, run 7lowest last 0.0018, run 70.0018▼ −2.92130123400:0002:0004:0006:0008:0010:0012:0014:00phase starting, 0.0 s to 22.9 sphase training, 22.9 s to 14:00epoch 1, 04:54epoch 2, 09:16epoch 3, 13:51trainingepoch 1epoch 22.9231Lowest last train loss across this project's 10 runs, run 7lowest last 0.0018, run 70.0018▼ −2.9213

Eval measures

Elapsed

268.8 s

epoch interval

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

0.71 s per step, 84.5 steps a minute

Eval accuracy by eval

+0.0200

▲ on the eval before

eval 1 0.8780, eval 3 0.9140

Against the other runs

4 of 10

by last eval accuracy

best so far 0.9360, run 2

▲ +0.0196 against run 6

Set-up

6000 train, 500 eval

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

Seed spread, observed

0.0040
0.900.95

2 finished seeds

Learning rate

0.00e+0

▼ −2.99e-6 since first

On one axis

phase starting, 0.0 s to 22.9 sphase training, 22.9 s to 14:00E1E2E3traininglowest last 0.0018Train loss0.0018▼ −2.92133.000.002.92310.0018last ▼ −2.9213best 0.9360, run 2Last-value range, 10 runs, 0.8854 to 0.9360Eval accuracy0.9140▲ +0.03600.940.860.87800.9140last ▲ +0.0360Learning rate0.00e+0▼ −2.99e-62.0e-50.0e+02.99e-6last 0.00e+0Power, whole chip, CPU and GPU117 W▼ −14 W140 W110 W131 Wlast 117 WHottest, one board zone90 °C▼ −8 °C100 °C80 °C98 °Clast 90 °C00:0002:0004:0006:0008:0010:0012:0014:00
GPUpower: whole chip, CPU and GPUtemperature: one board zone

GPU 0

Wwhole chip, CPU and GPU

117 W whole chip, CPU and GPU

use88%
memory, whole GPU1.1 GB

peak 17.8 GB, whole GPU

90 °C one board zone

history 0 to 132 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 0.9100, config 1 of 8 (learning rate 0.0001, batch 16), seed 0, Sun 22 Nov, 10:33 UTCfinished, eval accuracy 0.9360, config 2 of 8 (learning rate 0.00002, batch 16), seed 0, Sun 22 Nov, 10:51 UTCfinished, eval accuracy 0.9220, config 3 of 8 (learning rate 0.00005, batch 16), seed 0, Sun 22 Nov, 11:09 UTCfinished, eval accuracy 0.8966, config 4 of 8 (learning rate 0.0001, batch 16), seed 0, Sun 22 Nov, 11:27 UTCfinished, eval accuracy 0.8944, config 5 of 8 (learning rate 0.00002, batch 16), seed 0, Sun 22 Nov, 11:44 UTCfinished, eval accuracy 0.8944, config 6 of 8 (learning rate 0.00005, batch 16), seed 0, Sun 22 Nov, 12:01 UTCfinished, eval accuracy 0.9140, config 7 of 8 (batch 16), seed 1 of 2, Sun 22 Nov, 12:56 UTCfinished, eval accuracy 0.9180, config 7 of 8 (batch 16), seed 2 of 2, Sun 22 Nov, 13:14 UTCfinished, eval accuracy 0.8966, config 8 of 8 (batch 16), seed 1 of 2, Sun 22 Nov, 13:32 UTCfinished, eval accuracy 0.8854, config 8 of 8 (batch 16), seed 2 of 2, Sun 22 Nov, 13:49 UTC0.950.85finished, eval accuracy 0.9100, config 1 of 8 (learning rate 0.0001, batch 16), seed 0, Sun 22 Nov, 10:33 UTCfinished, eval accuracy 0.9360, config 2 of 8 (learning rate 0.00002, batch 16), seed 0, Sun 22 Nov, 10:51 UTCfinished, eval accuracy 0.9220, config 3 of 8 (learning rate 0.00005, batch 16), seed 0, Sun 22 Nov, 11:09 UTCfinished, eval accuracy 0.8966, config 4 of 8 (learning rate 0.0001, batch 16), seed 0, Sun 22 Nov, 11:27 UTCfinished, eval accuracy 0.8944, config 5 of 8 (learning rate 0.00002, batch 16), seed 0, Sun 22 Nov, 11:44 UTCfinished, eval accuracy 0.8944, config 6 of 8 (learning rate 0.00005, batch 16), seed 0, Sun 22 Nov, 12:01 UTCfinished, eval accuracy 0.9140, config 7 of 8 (batch 16), seed 1 of 2, Sun 22 Nov, 12:56 UTCfinished, eval accuracy 0.9180, config 7 of 8 (batch 16), seed 2 of 2, Sun 22 Nov, 13:14 UTCfinished, eval accuracy 0.8966, config 8 of 8 (batch 16), seed 1 of 2, Sun 22 Nov, 13:32 UTCfinished, eval accuracy 0.8854, config 8 of 8 (batch 16), seed 2 of 2, Sun 22 Nov, 13:49 UTC

7 of 10 finished, 0.9140, config 7 of 8 (batch 16), seed 1 of 2, Sun 22 Nov, 12:56 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Sun 22 Nov, 13:49 UTC| finished0.88541032 of 1032batch 16, seed 2 of 2
Sun 22 Nov, 13:32 UTC| finished0.89661032 of 1032batch 16, seed 1 of 2
Sun 22 Nov, 13:14 UTC| finished0.91801125 of 1125batch 16, seed 2 of 2
Sun 22 Nov, 12:56 UTC| finished0.91401125 of 1125batch 16, seed 1 of 2
Sun 22 Nov, 12:01 UTC| finished0.89441032 of 1032learning rate 0.00005, batch 16, seed 0
Sun 22 Nov, 11:44 UTC| finished0.89441032 of 1032learning rate 0.00002, batch 16, seed 0
Sun 22 Nov, 11:27 UTC| finished0.89661032 of 1032learning rate 0.0001, batch 16, seed 0
Sun 22 Nov, 11:09 UTC| finished0.92201125 of 1125learning rate 0.00005, batch 16, seed 0
Sun 22 Nov, 10:51 UTC| finished0.93601125 of 1125learning rate 0.00002, batch 16, seed 0
Sun 22 Nov, 10:33 UTC| finished0.91001125 of 1125learning rate 0.0001, batch 16, seed 0

Plate

Model
tasksource/tasksource-decider-nano, 149M parameters revision 2c7c01f
Data
Familiar decisions, shared draw, synthetic6000 train500 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
12.6 swent public
22.9 sphase training
289.9 seval 1
294.4 sepoch 1
556.4 sepoch 2
565.2 seval 2
832.0 sepoch 3
840.8 seval 3

Seeds

SeedStatusEval accuracy
seed 1| finished0.9140
seed 2| finished0.9180

observed spread 0.0040, 2 finished seeds

Readings

114 rows as a table

rows 1 to 100 of 114

StepSecondsTrain lossEval accuracyLearning rate
1022.92.9231-2.99e-6
2029.91.0618-5.97e-6
3037.00.9990-8.96e-6
4044.01.6194-1.19e-5
5051.21.5714-1.49e-5
6058.32.3572-1.79e-5
7065.41.0457-1.99e-5
8072.51.2612-1.98e-5
9079.61.3727-1.96e-5
10086.71.1472-1.94e-5
11093.71.9953-1.92e-5
120100.81.3218-1.90e-5
130108.01.0923-1.88e-5
140115.00.8111-1.86e-5
150122.10.7995-1.84e-5
160129.10.9946-1.82e-5
170136.30.8590-1.81e-5
180143.50.6932-1.79e-5
190150.61.3924-1.77e-5
200157.60.4398-1.75e-5
210164.70.2711-1.73e-5
220171.60.4620-1.71e-5
230178.80.4332-1.69e-5
240185.90.4823-1.67e-5
250192.80.7155-1.65e-5
260199.90.8840-1.64e-5
270207.00.3216-1.62e-5
280214.00.7923-1.60e-5
290221.00.4045-1.58e-5
300228.00.2608-1.56e-5
310235.10.9147-1.54e-5
320242.20.2131-1.52e-5
330249.20.6253-1.50e-5
340256.40.5188-1.48e-5
350263.50.3717-1.47e-5
360270.50.0806-1.45e-5
370277.50.2240-1.43e-5
375289.9-0.8780-
380294.40.3173-1.41e-5
390301.50.4758-1.39e-5
400308.60.4132-1.37e-5
410315.60.2535-1.35e-5
420322.70.4776-1.33e-5
430329.70.4472-1.31e-5
440336.80.3015-1.29e-5
450343.91.0682-1.28e-5
460351.00.4763-1.26e-5
470358.00.1151-1.24e-5
480365.00.4892-1.22e-5
490372.10.2324-1.20e-5
500379.30.1419-1.18e-5
510386.50.3967-1.16e-5
520393.60.4298-1.14e-5
530400.70.0550-1.12e-5
540407.80.0959-1.11e-5
550414.90.2912-1.09e-5
560422.00.7046-1.07e-5
570429.10.5088-1.05e-5
580436.20.9361-1.03e-5
590443.10.0967-1.01e-5
600450.20.0880-9.92e-6
610457.20.0750-9.74e-6
620464.30.3467-9.55e-6
630471.30.2671-9.36e-6
640478.40.0308-9.17e-6
650485.50.3807-8.98e-6
660492.50.0211-8.79e-6
670499.60.1254-8.60e-6
680506.60.6840-8.41e-6
690513.80.7553-8.22e-6
700520.80.5825-8.03e-6
710528.10.4372-7.84e-6
720535.20.5907-7.66e-6
730542.20.2869-7.47e-6
740549.30.3999-7.28e-6
750565.20.17690.89407.09e-6
760573.30.4294-6.90e-6
770580.40.0711-6.71e-6
780587.50.2120-6.52e-6
790594.60.3558-6.33e-6
800601.70.1993-6.14e-6
810608.70.2061-5.95e-6
820615.80.2626-5.77e-6
830622.90.7836-5.58e-6
840630.00.4100-5.39e-6
850637.00.6174-5.20e-6
860644.10.7950-5.01e-6
870651.10.0297-4.82e-6
880658.10.5615-4.63e-6
890665.30.1648-4.44e-6
900672.30.0215-4.25e-6
910679.30.1751-4.06e-6
920686.30.0316-3.88e-6
930693.50.2083-3.69e-6
940700.50.0765-3.50e-6
950707.70.0965-3.31e-6
960714.70.3793-3.12e-6
970721.80.3310-2.93e-6
980729.00.3646-2.74e-6
990736.20.0341-2.55e-6

GPU readings

114 rows as a table
StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
10068131.110.397.8
40066119.614.383.8
900100116.316.085.8
140071115.916.086.8
180067119.817.887.8
230066120.217.887.8
270071119.917.888.8
320068120.317.888.8
370066120.817.888.8
400071120.417.888.8
440073119.817.888.8
490071120.517.888.8
540067116.617.889.8
580070119.017.889.8
630071118.317.889.8
670067118.917.889.8
7200100117.317.889.8
750067131.917.892.8
800073119.717.888.8
8400100115.717.888.8
8900100115.917.888.8
9300100115.817.887.8
980072119.217.888.8
1020070119.717.888.8
10700100115.717.888.8
1120070119.517.888.8
1125088116.81.189.8

Worked out

ReadoutValueWorked out from
Run complete100%1125 steps over 1125
Step time0.71 s777.4 s over 1095 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute84.560 s over the step time of 0.71 s
Epoch interval268.8 sMean of 2 gaps between epoch marks
Train loss since first reading−2.92130.0018 at step 1125 minus 2.9231 at step 10
Lowest train loss0.0018Lowest of 113 readings, at step 1125
Eval accuracy, as shown91.4%0.9140 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.03600.9140 minus 0.8780
Eval change+0.0200Eval 3 at 0.9140 minus eval 2 at 0.8940
Wrong answers43500 eval items minus 457 correct. 457 over 500 is the only count that gives 0.9140
Rank4 of 10By last eval accuracy, across this project's 10 runs with an eval
Best so far0.9360Highest last eval accuracy, run 2
Against the run before+0.01960.9140 minus 0.8944 on run 6
Lowest last train loss0.0018Lowest last train loss across this project's 10 runs, run 7
Seed spread, observed0.0040High 0.9180 minus low 0.9140 across 2 finished seeds
Items per second22.51 step over 0.71 s, times batch size 16
Learning rate since first reading−2.99e-60.00e+0 minus 2.99e-6
Peak memory per GPU, whole GPU17.8 GBHighest sensor reading of any GPU, GPU 0, counting every process on it