CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Sat 21 Nov, 22:54 UTC
%
0.8472377 of 445 eval, 68 wrong
▲ +4.72 pts across 3 evals▲ +2.25 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, 22:15 UTC
ran for
25:54
config
batch 16, seed 2 of 2

run6/6step1032/1032epoch3/3config5/5seed2/2batch16modellaya 395Mrev7b928d8items5503+445

Loss train 0.2218 ▼ −2.0868 lowest 0.0582 at step 760 run before, run 5 ▼ −0.6991 step 1032
Timebase
Range
Smoothing
Reference
104 readings, 10 min/div5 min/div5 min/div
012300:0010:0020:00phase starting, 0.0 s to 37.0 sphase training, 37.0 s to 25:54epoch 1, 09:02epoch 2, 17:27epoch 3, 25:38trainingepoch 1epoch 2−2.0868since first2.3086Lowest last train loss across this project's 6 runs, run 1lowest last 0.0241, run 10.2218012300:0005:0010:0015:0020:0025:00phase starting, 0.0 s to 37.0 sphase training, 37.0 s to 25:54epoch 1, 09:02epoch 2, 17:27epoch 3, 25:38trainingepoch 1epoch 22.3086Lowest last train loss across this project's 6 runs, run 1lowest last 0.0241, run 10.2218▼ −2.0868012300:0005:0010:0015:0020:0025:00phase starting, 0.0 s to 37.0 sphase training, 37.0 s to 25:54epoch 1, 09:02epoch 2, 17:27epoch 3, 25:38trainingepoch 1epoch 22.3086Lowest last train loss across this project's 6 runs, run 1lowest last 0.0241, run 10.2218▼ −2.0868

Eval measures

Elapsed

498.3 s

epoch interval

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

1.45 s per step, 41.3 steps a minute

Eval accuracy by eval

+0.0225

▲ on the eval before

eval 1 0.8000, eval 3 0.8472

Against the other runs

4 of 6

by last eval accuracy

best so far 0.9580, run 1

▲ +0.2045 against run 5

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 2

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 37.0 sphase training, 37.0 s to 25:54E1E2E3traininglowest last 0.0241Train loss0.2218▼ −2.08683.000.002.30860.2218last ▼ −2.0868best 0.9580, run 1Last-value range, 6 runs, 0.5618 to 0.9580Eval accuracy0.8472▲ +0.04721.000.400.80000.8472last ▲ +0.0472Learning rate0.00e+0▼ −3.28e-62.0e-50.0e+03.28e-6last 0.00e+0Power, whole chip, CPU and GPU118 W▲ +12 W130 W100 W106 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

use99%
memory, whole GPU2.1 GB

peak 29.0 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 newer ▸

6 of 6 finished, 0.8472, config 5 of 5 (batch 16), seed 2 of 2, Sat 21 Nov, 22:15 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 2
Hardware
Radeon 8060S, ROCm

Log

Since startMark
0.0 sphase starting
10.2 swent public
37.0 sphase training
531.4 seval 1
542.2 sepoch 1
1042.8 seval 2
1047.7 sepoch 2
1538.8 sepoch 3
1554.2 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
1037.02.3086-3.28e-6
2051.22.3790-6.56e-6
3065.61.7462-9.84e-6
4079.91.2691-1.31e-5
5094.31.6433-1.64e-5
60108.51.5471-1.97e-5
70122.71.8716-1.98e-5
80137.01.6844-1.96e-5
90151.51.5975-1.94e-5
100166.01.9999-1.92e-5
110180.41.4806-1.90e-5
120194.81.2893-1.88e-5
130209.21.2322-1.86e-5
140223.41.0669-1.84e-5
150237.81.8967-1.82e-5
160252.21.1570-1.80e-5
170266.60.6502-1.78e-5
180281.20.6136-1.75e-5
190295.60.6204-1.73e-5
200309.81.0107-1.71e-5
210324.20.3390-1.69e-5
220338.60.9567-1.67e-5
230353.00.7723-1.65e-5
240367.00.7318-1.63e-5
250381.40.7470-1.61e-5
260395.50.7574-1.59e-5
270409.80.6087-1.57e-5
280424.40.7078-1.55e-5
290438.70.5851-1.53e-5
300453.20.9679-1.51e-5
310467.40.5371-1.49e-5
320481.70.5175-1.47e-5
330496.10.4870-1.45e-5
340510.40.3625-1.43e-5
344531.4-0.8000-
350542.20.9704-1.40e-5
360556.50.1949-1.38e-5
370570.80.5028-1.36e-5
380585.10.2532-1.34e-5
390599.50.6860-1.32e-5
400613.80.1501-1.30e-5
410628.00.6025-1.28e-5
420642.40.4269-1.26e-5
430656.81.1103-1.24e-5
440671.10.5759-1.22e-5
450685.40.2988-1.20e-5
460699.80.4291-1.18e-5
470714.20.2577-1.16e-5
480728.60.1369-1.14e-5
490742.90.2447-1.12e-5
500757.20.2295-1.10e-5
510771.50.6797-1.08e-5
520785.80.4941-1.05e-5
530800.10.6619-1.03e-5
540814.70.3548-1.01e-5
550829.00.6514-9.93e-6
560843.30.4788-9.72e-6
570857.70.2761-9.52e-6
580872.30.4452-9.31e-6
590886.70.3815-9.10e-6
600901.00.6561-8.90e-6
610915.40.9247-8.69e-6
620929.70.7777-8.49e-6
630944.00.1584-8.28e-6
640958.40.5107-8.07e-6
650972.70.3200-7.87e-6
660987.20.3470-7.66e-6
6701001.60.4415-7.46e-6
6801016.00.0879-7.25e-6
6881042.8-0.8247-
6901047.70.2691-7.04e-6
7001062.20.5938-6.84e-6
7101076.40.1453-6.63e-6
7201090.70.0743-6.43e-6
7301105.20.2717-6.22e-6
7401119.40.1812-6.01e-6
7501133.70.3640-5.81e-6
7601148.10.0582-5.60e-6
7701162.40.4536-5.40e-6
7801176.70.0766-5.19e-6
7901191.00.1299-4.98e-6
8001205.30.1927-4.78e-6
8101219.70.1712-4.57e-6
8201234.20.4733-4.37e-6
8301248.60.1744-4.16e-6
8401263.00.0635-3.95e-6
8501277.20.2293-3.75e-6
8601291.60.1102-3.54e-6
8701306.10.2738-3.34e-6
8801320.50.4715-3.13e-6
8901334.90.0730-2.92e-6
9001349.30.1580-2.72e-6
9101363.50.1570-2.51e-6
9201378.00.1899-2.31e-6
9301392.30.1969-2.10e-6
9401406.70.2371-1.89e-6
9501421.30.3221-1.69e-6
9601435.50.1644-1.48e-6
9701449.80.2754-1.28e-6
9801464.10.0646-1.07e-6

GPU readings

106 rows as a table
StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
10031106.017.091.8
2000118.622.787.8
400100118.125.688.8
600100113.925.689.8
900100121.929.089.8
1100100114.429.089.8
1300100120.929.089.8
1500100114.029.090.8
1800100117.329.090.8
2000100113.529.090.8
2200100118.429.090.8
2400100115.329.090.8
27000119.029.090.8
2900100114.729.090.8
31000118.529.090.8
3300100118.229.090.8
34400123.129.095.8
3700100116.129.090.8
39000116.129.090.8
4100100120.029.090.8
4400100114.429.090.8
4600100122.129.090.8
4800100115.429.090.8
50000120.129.090.8
5300100117.929.090.8
55000119.629.091.8
5700100116.529.090.8
59000121.929.090.8
6200100115.629.090.8
64000120.429.090.8
6600100116.029.090.8
68000121.129.090.8
6900100115.229.092.8
72000121.629.090.8
7400100115.929.090.8
76000115.629.090.8
7800100117.829.090.8
8100100113.229.090.8
8300100117.729.090.8
85000115.229.090.8
8700100115.729.090.8
90000120.829.090.8
9200100116.629.090.8
94000119.429.090.8
9600100117.429.090.8
99000114.229.090.8
10100100117.729.090.8
1032099117.92.191.8

Worked out

ReadoutValueWorked out from
Run complete100%1032 steps over 1032
Step time1.45 s1473.0 s over 1014 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute41.360 s over the step time of 1.45 s
Epoch interval498.3 sMean of 2 gaps between epoch marks
Train loss since first reading−2.08680.2218 at step 1032 minus 2.3086 at step 10
Lowest train loss0.0582Lowest of 104 readings, at step 760
Eval accuracy, as shown84.7%0.8472 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.04720.8472 minus 0.8000
Eval change+0.0225Eval 3 at 0.8472 minus eval 2 at 0.8247
Wrong answers68445 eval items minus 377 correct. 377 over 445 is the only count that gives 0.8472
Rank4 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+0.20450.8472 minus 0.6427 on run 5
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 second11.01 step over 1.45 s, times batch size 16
Train loss against run 5−0.69910.2218 minus 0.9209 on run 5, both at step 1032
Learning rate since first reading−3.28e-60.00e+0 minus 3.28e-6
Peak memory per GPU, whole GPU29.0 GBHighest sensor reading of any GPU, GPU 0, counting every process on it