CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Thu 19 Nov, 12:06 UTC
%
0.9000450 of 500 eval, 50 wrong
▲ +34.40 pts across 2 evals▲ +34.40 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Familiar decider: OneJev 0.8B

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.

started
Thu 19 Nov, 07:46 UTC
ran for
259:22
config
learning rate 0.000005, batch 8, seed 0 of 3

run1/3step1446/1446epoch3/3seed0/3lr0.000005batch8modelOneJev1.6-0.8B 0.8Brevc89933citems6000+500

Loss train 0.0002 ▼ −1.1689 eval 0.3603 lowest 0.0001 at step 1000
Timebase
Range
Smoothing
290 readings, 120 min/div60 min/div30 min/div
00.511.500:00120:00240:00phase starting, 0.0 s to 06:42phase training, 06:42 to 259:22epoch 1, 06:42epoch 2, 93:01epoch 3, 173:48trainingepoch 2epoch 3−1.1689since first1.1690Lowest last train loss across this project's 3 runs, run 1lowest last 0.0002, run 10.000200.511.500:0060:00120:00180:00240:00phase starting, 0.0 s to 06:42phase training, 06:42 to 259:22epoch 1, 06:42epoch 2, 93:01epoch 3, 173:48trainingepoch 2epoch 31.1690Lowest last train loss across this project's 3 runs, run 1lowest last 0.0002, run 1eval 0.36030.0002▼ −1.168900.511.500:0030:0060:0090:00120:00150:00180:00210:00240:00phase starting, 0.0 s to 06:42phase training, 06:42 to 259:22epoch 1, 06:42epoch 2, 93:01epoch 3, 173:48trainingepoch 2epoch 31.1690Lowest last train loss across this project's 3 runs, run 1lowest last 0.0002, run 1eval 0.36030.0002▼ −1.1689

Eval measures

Eval losslower is better
0.3603▼ −1.0101 since first
02 →
ECEcalibration error, lower is better
0.0673▼ −0.0451 since first
01

Elapsed

5013.2 s

epoch interval

1446 of 1446 steps, 3 of 3 epochs, 100%

10.4 s per step, 5.8 steps a minute

Eval accuracy by eval

+0.3440

▲ on the eval before

eval 1 0.5560, eval 2 0.9000

Against the other runs

3 of 3

by last eval accuracy

best so far 0.9120, run 2

Train minus eval

−0.3601

step 1446

train 0.0002, eval 0.3603

Set-up

6000 train, 500 eval

synthetic, learning rate 0.000005, batch size 8, grad accum 2, epochs 3, optimizer adamw, scheduler cosine, seed 0

Seed spread, observed

0.0120
0.900.95

3 finished seeds

Learning rate

0.00e+0

▼ −5.81e-7 since first

On one axis

phase starting, 0.0 s to 06:42phase training, 06:42 to 259:22E2E3traininglowest last 0.0002Train loss0.0002▼ −1.16891.500.001.16900.0002last ▼ −1.1689best 0.9120, run 2Last-value range, 3 runs, 0.9000 to 0.9120Eval accuracy0.9000▲ +0.34401.000.400.55600.9000last ▲ +0.3440Learning rate0.00e+0▼ −5.81e-76.0e-60.0e+05.81e-7last 0.00e+0Power, whole chip, CPU and GPU53 W▲ +1 W55 W52 W53 Wlast 53 WHottest, one board zone69 °C▲ +1 °C70 °C67 °C68 °Clast 69 °C00:0030:0060:0090:00120:00150:00180:00210:00240:00
GPUpower: whole chip, CPU and GPUtemperature: one board zone

GPU 0

Wwhole chip, CPU and GPU

53 W whole chip, CPU and GPU

use2%
memory, whole GPU28.8 GB

peak 47.1 GB, whole GPU

69 °C one board zone

history 0 to 55 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 3 finished, 0.9000, config 1 of 1 (learning rate 0.000005, batch 8), seed 0 of 3, Thu 19 Nov, 07:46 UTC

◂
Runs as a table
StartedStatusEval accuracyStepsConfig
Thu 19 Nov, 16:26 UTC| finished0.91001446 of 1446learning rate 0.000005, batch 8, seed 2 of 3
Thu 19 Nov, 12:06 UTC| finished0.91201446 of 1446learning rate 0.000005, batch 8, seed 1 of 3
Thu 19 Nov, 07:46 UTC| finished0.90001446 of 1446learning rate 0.000005, batch 8, seed 0 of 3

Plate

Model
OmniJev/OneJev1.6-0.8B, 0.8B parameters revision c89933c
Data
Familiar decisions, shared draw, synthetic6000 train500 eval
Config
learning rate 0.000005, batch size 8, grad accum 2, epochs 3, optimizer adamw, scheduler cosine, seed 0
Hardware
Radeon 8060S, ROCm

Log

Since startMark
0.0 sphase starting
6.6 swent public
311.8 seval 1
402.2 sphase training
402.2 sepoch 1
5581.2 sepoch 2
10428.7 sepoch 3
15562.4 seval 2

Seeds

SeedStatusEval accuracy
seed 0| finished0.9000
seed 1| finished0.9120
seed 2| finished0.9100

observed spread 0.0120, 3 finished seeds

Readings

291 rows as a table

rows 1 to 100 of 291

StepSecondsTrain lossEval lossEval accuracyECELearning rate
0311.8-1.37040.55600.1124-
5402.21.1690---5.81e-7
10474.71.0891---1.16e-6
15556.70.4793---1.74e-6
20628.71.1058---2.33e-6
25703.81.0500---2.91e-6
30783.90.7279---3.49e-6
35850.50.5699---4.07e-6
40909.50.9047---4.65e-6
45977.80.7288---5.00e-6
501038.40.7085---5.00e-6
551097.80.7507---5.00e-6
601161.30.8121---5.00e-6
651216.90.4212---5.00e-6
701274.90.7079---5.00e-6
751326.00.8343---4.99e-6
801386.10.5302---4.99e-6
851444.60.4071---4.99e-6
901499.40.9121---4.99e-6
951549.41.0514---4.98e-6
1001600.20.4042---4.98e-6
1051659.60.4534---4.98e-6
1101712.60.4019---4.97e-6
1151762.50.4160---4.97e-6
1201812.80.5798---4.96e-6
1251868.00.4886---4.96e-6
1301923.70.7564---4.95e-6
1351978.90.3599---4.95e-6
1402029.10.6216---4.94e-6
1452080.00.4608---4.94e-6
1502135.10.5566---4.93e-6
1552186.41.0272---4.92e-6
1602237.90.4468---4.91e-6
1652291.60.3114---4.91e-6
1702347.80.8700---4.90e-6
1752402.30.7735---4.89e-6
1802457.90.6258---4.88e-6
1852508.10.6778---4.87e-6
1902564.40.3511---4.87e-6
1952614.00.4308---4.86e-6
2002663.80.6331---4.85e-6
2052714.40.7890---4.84e-6
2102764.90.4295---4.83e-6
2152814.50.4324---4.82e-6
2202864.60.7023---4.81e-6
2252914.50.9014---4.80e-6
2302964.90.3195---4.78e-6
2353018.10.2222---4.77e-6
2403068.90.4086---4.76e-6
2453122.90.2560---4.75e-6
2503173.50.4100---4.74e-6
2553223.80.5123---4.72e-6
2603274.00.6740---4.71e-6
2653324.20.5302---4.70e-6
2703378.90.4670---4.68e-6
2753433.20.3908---4.67e-6
2803482.50.4563---4.66e-6
2853532.30.2938---4.64e-6
2903582.80.8238---4.63e-6
2953633.60.2082---4.61e-6
3003688.90.5166---4.60e-6
3053743.30.6407---4.58e-6
3103794.40.5407---4.57e-6
3153843.90.1736---4.55e-6
3203892.60.4540---4.53e-6
3253947.80.3041---4.52e-6
3303999.10.3926---4.50e-6
3354054.40.2954---4.48e-6
3404105.00.8557---4.47e-6
3454155.60.5777---4.45e-6
3504204.70.6123---4.43e-6
3554254.70.6064---4.41e-6
3604304.70.4284---4.40e-6
3654360.20.3023---4.38e-6
3704410.50.3924---4.36e-6
3754461.70.2925---4.34e-6
3804512.00.3449---4.32e-6
3854561.70.2613---4.30e-6
3904611.50.2143---4.28e-6
3954661.60.2327---4.26e-6
4004711.81.0338---4.24e-6
4054762.70.8334---4.22e-6
4104811.90.2659---4.20e-6
4154865.50.5452---4.18e-6
4204917.20.1753---4.16e-6
4254967.10.5525---4.14e-6
4305023.10.4257---4.12e-6
4355075.00.0747---4.10e-6
4405124.70.4945---4.08e-6
4455174.80.1100---4.05e-6
4505226.20.5593---4.03e-6
4555281.60.1999---4.01e-6
4605332.30.4309---3.99e-6
4655381.20.4827---3.96e-6
4705431.20.3624---3.94e-6
4755480.60.0754---3.92e-6
4805531.70.2788---3.90e-6
4855581.20.4915---3.87e-6
4905630.90.1239---3.85e-6
4955680.90.7591---3.83e-6

GPU readings

291 rows as a table

rows 1 to 100 of 290

StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
004452.639.867.8
50053.238.768.8
1004852.839.568.8
1503052.825.968.8
2002953.534.167.8
250053.040.568.8
300053.145.167.8
3503553.028.568.8
4001453.137.668.8
450253.041.468.8
500052.637.868.8
550052.940.068.8
600053.141.967.8
6502053.144.667.8
700052.942.467.8
750152.941.967.8
800053.140.468.8
850053.045.068.8
900053.046.968.8
9502053.346.968.8
1000052.844.168.8
10501553.238.668.8
1100053.044.968.8
1150253.246.867.8
1200052.937.668.8
1250052.941.467.8
1300053.039.768.8
1350052.942.267.8
1400053.342.268.8
1450052.842.268.8
1500052.942.268.8
1550053.142.267.8
1600052.844.068.8
1650052.944.768.8
1700053.039.468.8
1750052.844.268.8
1800053.644.268.8
1850153.242.067.8
1900053.042.068.8
1950053.042.068.8
2000053.042.068.8
2050053.242.068.8
2100352.942.068.8
2150053.142.068.8
2200053.142.068.8
2250053.042.068.8
2300053.042.068.8
23501653.145.668.8
2400053.146.668.8
2450653.146.968.8
2500053.146.968.8
2550052.846.968.8
2600153.046.968.8
2650053.138.469.8
2700153.344.868.8
2750053.144.868.8
2800053.144.868.8
2850053.044.868.8
2900053.044.868.8
2950153.241.968.8
3000053.242.368.8
3050053.042.368.8
3100753.242.368.8
3150053.042.368.8
3200153.141.668.8
32502352.941.668.8
3300053.141.668.8
3350053.144.168.8
3400053.144.168.8
3450052.845.168.8
3500153.045.168.8
3550052.745.168.8
3600052.942.168.8
3650153.042.168.8
3700153.142.168.8
3750052.942.168.8
3800152.842.168.8
3850052.842.168.8
3900552.942.168.8
3950052.842.168.8
4000053.042.168.8
4050153.142.168.8
4100053.045.468.8
4150053.046.368.8
4200153.146.368.8
4250053.040.469.8
4300053.142.368.8
4350052.942.368.8
4400052.942.368.8
4450053.442.368.8
4500052.842.168.8
4550153.242.168.8
4600052.842.168.8
4650152.942.168.8
4700053.142.168.8
4750053.043.168.8
4800053.142.568.8
4850053.042.568.8
4900052.742.568.8
4950053.042.568.8

Worked out

ReadoutValueWorked out from
Run complete100%1446 steps over 1446
Step time10.4 s14967.9 s over 1445 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute5.860 s over the step time of 10.36 s
Epoch interval5013.2 sMean of 2 gaps between epoch marks
Train loss since first reading−1.16890.0002 at step 1446 minus 1.1690 at step 5
Lowest train loss0.0001Lowest of 290 readings, at step 1000
Eval accuracy, as shown90.0%0.9000 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.34400.9000 minus 0.5560
Eval loss since first eval−1.01010.3603 minus 1.3704
ECE since first eval−0.04510.0673 minus 0.1124
Eval change+0.3440Eval 2 at 0.9000 minus eval 1 at 0.5560
Train minus eval loss−0.36010.0002 minus 0.3603, both at step 1446
Wrong answers50500 eval items minus 450 correct. 450 over 500 is the only count that gives 0.9000
Rank3 of 3By last eval accuracy, across this project's 3 runs with an eval
Best so far0.9120Highest last eval accuracy, run 2
Lowest last train loss0.0002Lowest last train loss across this project's 3 runs, run 1
Seed spread, observed0.0120High 0.9120 minus low 0.9000 across 3 finished seeds
Items per second0.81 step over 10.36 s, times batch size 8
Learning rate since first reading−5.81e-70.00e+0 minus 5.81e-7
Peak memory per GPU, whole GPU47.1 GBHighest sensor reading of any GPU, GPU 0, counting every process on it