CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Sat 21 Nov, 22:15 UTC
%
0.6427286 of 445 eval, 159 wrong
▲ +3.60 pts across 3 evals▲ +0.00 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Familiar decider: Laya's encoder

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, 21:36 UTC
ran for
25:50
config
batch 16, seed 1 of 2

run5/6step1032/1032epoch3/3config5/5seed1/2batch16modellaya 395Mrev7b928d8items5503+445

Loss train 0.9209 ▼ −0.5620 lowest 0.5404 at step 890 run before, run 4 ▲ +0.4475 step 1032
Timebase
Range
Smoothing
Reference
104 readings, 10 min/div5 min/div5 min/div
012300:0010:0020:00phase starting, 0.0 s to 36.7 sphase training, 36.7 s to 25:50epoch 1, 09:01epoch 2, 17:25epoch 3, 25:35trainingepoch 1epoch 21.4828Lowest last train loss across this project's 6 runs, run 1lowest last 0.0241, run 10.9209012300:0005:0010:0015:0020:0025:00phase starting, 0.0 s to 36.7 sphase training, 36.7 s to 25:50epoch 1, 09:01epoch 2, 17:25epoch 3, 25:35trainingepoch 1epoch 21.4828Lowest last train loss across this project's 6 runs, run 1lowest last 0.0241, run 10.9209▼ −0.5620012300:0005:0010:0015:0020:0025:00phase starting, 0.0 s to 36.7 sphase training, 36.7 s to 25:50epoch 1, 09:01epoch 2, 17:25epoch 3, 25:35trainingepoch 1epoch 21.4828Lowest last train loss across this project's 6 runs, run 1lowest last 0.0241, run 10.9209▼ −0.5620

Eval measures

Elapsed

497.1 s

epoch interval

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

1.44 s per step, 41.8 steps a minute

Eval accuracy by eval

+0.0000

▲ on the eval before

eval 1 0.6067, eval 3 0.6427

Against the other runs

5 of 6

by last eval accuracy

best so far 0.9580, run 1

▼ −0.2045 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

0.2045
0.600.85

2 finished seeds

Learning rate

0.00e+0

▼ −3.28e-6 since first

On one axis

phase starting, 0.0 s to 36.7 sphase training, 36.7 s to 25:50E1E2E3traininglowest last 0.0241Train loss0.9209▼ −0.56203.000.001.48280.9209last ▼ −0.5620best 0.9580, run 1Last-value range, 6 runs, 0.5618 to 0.9580Eval accuracy0.6427▲ +0.03601.000.400.60670.6427last ▲ +0.0360Learning rate0.00e+0▼ −3.28e-62.0e-50.0e+03.28e-6last 0.00e+0Power, whole chip, CPU and GPU118 W▲ +2 W125 W110 W116 Wlast 118 WHottest, one board zone92 °C0 °C100 °C85 °C92 °Clast 92 °C00:0005:0010:0015:0020:0025:00
GPUpower: whole chip, CPU and GPUtemperature: one board zone

GPU 0

Wwhole chip, CPU and GPU

118 W whole chip, CPU and GPU

use98%
memory, whole GPU2.1 GB

peak 24.1 GB, whole GPU

92 °C one board zone

history 0 to 123 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, 0.6427, config 5 of 5 (batch 16), seed 1 of 2, Sat 21 Nov, 21:36 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Sat 21 Nov, 22:15 UTC| finished0.84721032 of 1032batch 16, seed 2 of 2
Sat 21 Nov, 21:36 UTC| finished0.64271032 of 1032batch 16, seed 1 of 2
Sat 21 Nov, 17:47 UTC| finished0.84721032 of 1032learning rate 0.00005, batch 16, seed 0
Sat 21 Nov, 17:08 UTC| finished0.88991032 of 1032learning rate 0.00002, batch 16, seed 0
Sat 21 Nov, 16:30 UTC| finished0.56181032 of 1032learning rate 0.0001, batch 16, seed 0
Sat 21 Nov, 08:25 UTC| finished0.95807274 of 7274batch 16, seed 0

Plate

Model
convaiinnovations/laya, 395M parameters revision 7b928d8
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
4.2 swent public
36.7 sphase training
530.6 seval 1
541.1 sepoch 1
1040.6 seval 2
1045.5 sepoch 2
1535.3 sepoch 3
1550.7 seval 3

Seeds

SeedStatusEval accuracy
seed 1| finished0.6427
seed 2| finished0.8472

observed spread 0.2045, 2 finished seeds

Readings

106 rows as a table

rows 1 to 100 of 106

StepSecondsTrain lossEval accuracyLearning rate
1036.71.4828-3.28e-6
2051.01.7978-6.56e-6
3065.21.4614-9.84e-6
4079.61.4462-1.31e-5
5094.00.9455-1.64e-5
60108.31.2479-1.97e-5
70122.71.7492-1.98e-5
80137.01.8939-1.96e-5
90151.20.9137-1.94e-5
100165.51.0253-1.92e-5
110179.71.7857-1.90e-5
120194.01.7881-1.88e-5
130208.40.9448-1.86e-5
140223.01.5798-1.84e-5
150237.31.8640-1.82e-5
160251.61.0538-1.80e-5
170265.91.3612-1.78e-5
180280.12.0154-1.75e-5
190294.61.6988-1.73e-5
200309.01.8939-1.71e-5
210323.41.4131-1.69e-5
220337.61.5881-1.67e-5
230351.91.8359-1.65e-5
240366.21.5060-1.63e-5
250380.61.4477-1.61e-5
260394.91.4159-1.59e-5
270409.21.6448-1.57e-5
280423.62.1109-1.55e-5
290438.01.0892-1.53e-5
300452.51.2091-1.51e-5
310466.62.7719-1.49e-5
320481.01.7618-1.47e-5
330495.32.3197-1.45e-5
340509.60.7997-1.43e-5
344530.6-0.6067-
350541.11.7758-1.40e-5
360555.51.2668-1.38e-5
370569.81.5114-1.36e-5
380584.11.6452-1.34e-5
390598.51.1493-1.32e-5
400612.71.0248-1.30e-5
410627.02.0084-1.28e-5
420641.51.5388-1.26e-5
430655.81.3932-1.24e-5
440670.21.0084-1.22e-5
450684.61.4393-1.20e-5
460698.71.5275-1.18e-5
470713.11.8516-1.16e-5
480727.42.1696-1.14e-5
490741.71.1088-1.12e-5
500756.01.8593-1.10e-5
510770.31.9121-1.08e-5
520784.60.9738-1.05e-5
530798.82.4165-1.03e-5
540813.21.4214-1.01e-5
550827.51.2476-9.93e-6
560841.91.2768-9.72e-6
570856.21.3574-9.52e-6
580870.50.8408-9.31e-6
590884.81.6362-9.10e-6
600899.31.1407-8.90e-6
610913.71.5710-8.69e-6
620928.11.2427-8.49e-6
630942.50.7408-8.28e-6
640956.81.5170-8.07e-6
650971.02.0965-7.87e-6
660985.31.2669-7.66e-6
670999.51.7056-7.46e-6
6801013.81.4644-7.25e-6
6881040.6-0.6427-
6901045.51.9987-7.04e-6
7001059.50.9313-6.84e-6
7101073.80.9959-6.63e-6
7201088.01.4583-6.43e-6
7301102.31.3947-6.22e-6
7401116.62.1286-6.01e-6
7501130.91.6923-5.81e-6
7601145.30.9456-5.60e-6
7701159.71.5272-5.40e-6
7801174.11.2561-5.19e-6
7901188.50.9776-4.98e-6
8001202.81.2960-4.78e-6
8101217.21.6188-4.57e-6
8201231.41.0989-4.37e-6
8301245.60.7813-4.16e-6
8401260.00.8024-3.95e-6
8501274.21.4601-3.75e-6
8601288.71.2993-3.54e-6
8701302.81.9740-3.34e-6
8801317.21.4416-3.13e-6
8901331.50.5404-2.92e-6
9001345.91.5327-2.72e-6
9101360.10.6843-2.51e-6
9201374.41.3468-2.31e-6
9301389.11.0973-2.10e-6
9401403.51.5559-1.89e-6
9501417.91.1048-1.69e-6
9601432.31.4351-1.48e-6
9701446.72.1529-1.28e-6
9801461.01.8126-1.07e-6

GPU readings

106 rows as a table
StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
10019115.617.391.8
200100114.717.387.8
4000122.920.488.8
600100115.124.189.8
9000116.824.189.8
1100100122.624.189.8
13000115.424.189.8
1500100118.224.190.8
1800100114.424.190.8
2000100118.024.190.8
2200100114.224.190.8
24000123.324.190.8
2700100115.724.190.8
29000122.324.190.8
3100100117.824.190.8
3400100114.124.190.8
3500100120.324.195.8
3700100114.824.189.8
39000121.924.189.8
4100100114.824.189.8
44000121.624.189.8
4600100116.924.189.8
48000116.324.189.8
5000100123.224.189.8
5300100114.124.189.8
55000122.524.189.8
5700100114.424.189.8
59000119.624.189.8
6200100114.224.189.8
64000121.924.189.8
6600100117.924.189.8
6800100120.024.189.8
69000122.624.190.8
7200100117.724.189.8
7400100113.424.189.8
7600100123.324.189.8
79000114.524.189.8
8100100118.324.189.8
8300100114.924.189.8
85000121.824.189.8
8800100115.724.189.8
90000120.524.189.8
9200100117.724.189.8
94000120.724.189.8
9700100114.924.190.8
9900100114.424.190.8
10100100119.924.190.8
1032098117.82.191.8

Worked out

ReadoutValueWorked out from
Run complete100%1032 steps over 1032
Step time1.44 s1448.0 s over 1008 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute41.860 s over the step time of 1.44 s
Epoch interval497.1 sMean of 2 gaps between epoch marks
Train loss since first reading−0.56200.9209 at step 1032 minus 1.4828 at step 10
Lowest train loss0.5404Lowest of 104 readings, at step 890
Eval accuracy, as shown64.3%0.6427 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.03600.6427 minus 0.6067
Eval change+0.0000Eval 3 at 0.6427 minus eval 2 at 0.6427
Wrong answers159445 eval items minus 286 correct. 286 over 445 is the only count that gives 0.6427
Rank5 of 6By last eval accuracy, across this project's 6 runs with an eval
Best so far0.9580Highest last eval accuracy, run 1
Against the run before−0.20450.6427 minus 0.8472 on run 4
Lowest last train loss0.0241Lowest last train loss across this project's 6 runs, run 1
Seed spread, observed0.2045High 0.8472 minus low 0.6427 across 2 finished seeds
Items per second11.11 step over 1.44 s, times batch size 16
Train loss against run 4+0.44750.9209 minus 0.4733 on run 4, both at step 1032
Learning rate since first reading−3.28e-60.00e+0 minus 3.28e-6
Peak memory per GPU, whole GPU24.1 GBHighest sensor reading of any GPU, GPU 0, counting every process on it