CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Sun 22 Nov, 01:31 UTC
%
0.561573 of 130 eval, 57 wrong
▲ +5.38 pts across 2 evals▲ +5.38 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Claviger: OneJev 0.8B

finishedRadeon 8060S

A small model that reads one Discord message and decides whether an agent should reply, and which one. OneJev 0.8B fine-tuned on invented chat, beside the Laya model.

started
Sun 22 Nov, 00:55 UTC
ran for
36:01
config
learning rate 0.00005, batch 8, seed 0

run6/7step339/339epoch3/3config4/4seed0lr0.00005batch8modelOneJev1.6-0.8B 0.8Brevc89933citems1800+130

Loss train 0.9522 ▲ +0.0711 eval 0.9727 lowest 0.8084 at step 40 run before, run 5 ▲ +0.9513 step 339
Timebase
Range
Smoothing
Reference
68 readings, 10 min/div10 min/div5 min/div
00.511.500:0010:0020:0030:00phase starting, 0.0 s to 90.4 sphase training, 90.4 s to 36:01epoch 1, 90.4 sepoch 2, 14:00epoch 3, 25:07trainingepoch 2epoch 3+0.0711since first0.8811Lowest last train loss across this project's 6 runs, run 2lowest last 0.0000, run 20.952200.511.500:0010:0020:0030:00phase starting, 0.0 s to 90.4 sphase training, 90.4 s to 36:01epoch 1, 90.4 sepoch 2, 14:00epoch 3, 25:07trainingepoch 2epoch 30.8811Lowest last train loss across this project's 6 runs, run 2lowest last 0.0000, run 2eval 0.97270.9522▲ +0.071100.511.500:0005:0010:0015:0020:0025:0030:0035:00phase starting, 0.0 s to 90.4 sphase training, 90.4 s to 36:01epoch 1, 90.4 sepoch 2, 14:00epoch 3, 25:07trainingepoch 2epoch 30.8811Lowest last train loss across this project's 6 runs, run 2lowest last 0.0000, run 2eval 0.97270.9522▲ +0.0711

Eval measures

Eval losslower is better
0.9727▼ −0.0899 since first
02 →
ECEcalibration error, lower is better
0.1650▼ −0.0386 since first
01

Elapsed

708.4 s

epoch interval

339 of 339 steps, 3 of 3 epochs, 100%

6.19 s per step, 9.7 steps a minute

Eval accuracy by eval

+0.0538

▲ on the eval before

eval 1 0.5077, eval 2 0.5615

Against the other runs

5 of 7

by last eval accuracy

best so far 0.9385, run 7

▼ −0.3154 against run 5

Train minus eval

−0.0205

step 339

train 0.9522, eval 0.9727

Set-up

1800 train, 130 eval

synthetic, learning rate 0.00005, batch size 8, grad accum 2, epochs 3, optimizer adamw, scheduler cosine, seed 0

Learning rate

0.00e+0

▼ −2.50e-5 since first

On one axis

phase starting, 0.0 s to 90.4 sphase training, 90.4 s to 36:01E1E2E3traininglowest last 0.0000Train loss0.9522▲ +0.07111.500.000.88110.9522last ▲ +0.0711best 0.9385, run 7Last-value range, 7 runs, 0.3000 to 0.9385Eval accuracy0.5615▲ +0.05381.000.250.50770.5615last ▲ +0.0538Learning rate0.00e+0▼ −2.50e-56.0e-50.0e+02.50e-5last 0.00e+0Power, whole chip, CPU and GPU118 W▲ +15 W130 W90 W103 Wlast 118 WHottest, one board zone90 °C▼ −8 °C100 °C85 °C98 °Clast 90 °C00:0005:0010:0015:0020:0025:0030:0035:00
GPUpower: whole chip, CPU and GPUtemperature: one board zone

GPU 0

Wwhole chip, CPU and GPU

118 W whole chip, CPU and GPU

use100%
memory, whole GPU38.1 GB

peak 41.7 GB, whole GPU

90 °C one board zone

history 0 to 122 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
0.950.300.950.30finished, eval accuracy 0.9000, config 1 of 4 (learning rate 0.000005, batch 8), seed 0 of 3, Thu 19 Nov, 22:36 UTCfinished, eval accuracy 0.9154, config 1 of 4 (learning rate 0.000005, batch 8), seed 1 of 3, Thu 19 Nov, 23:24 UTCfailed, eval accuracy 0.5077, config 1 of 4 (learning rate 0.000005, batch 8), seed 2 of 3, Fri 20 Nov, 00:08 UTCfinished, eval accuracy 0.3000, config 2 of 4 (learning rate 0.0001, batch 8), seed 0, Sat 21 Nov, 23:42 UTCfinished, eval accuracy 0.8769, config 3 of 4 (learning rate 0.00002, batch 8), seed 0, Sun 22 Nov, 00:18 UTCfinished, eval accuracy 0.5615, config 4 of 4 (learning rate 0.00005, batch 8), seed 0, Sun 22 Nov, 00:55 UTCfinished, eval accuracy 0.9385, config 1 of 4 (learning rate 0.000005, batch 8), seed 2 of 3, Sun 22 Nov, 01:31 UTC0.950.30finished, eval accuracy 0.9000, config 1 of 4 (learning rate 0.000005, batch 8), seed 0 of 3, Thu 19 Nov, 22:36 UTCfinished, eval accuracy 0.9154, config 1 of 4 (learning rate 0.000005, batch 8), seed 1 of 3, Thu 19 Nov, 23:24 UTCfailed, eval accuracy 0.5077, config 1 of 4 (learning rate 0.000005, batch 8), seed 2 of 3, Fri 20 Nov, 00:08 UTCfinished, eval accuracy 0.3000, config 2 of 4 (learning rate 0.0001, batch 8), seed 0, Sat 21 Nov, 23:42 UTCfinished, eval accuracy 0.8769, config 3 of 4 (learning rate 0.00002, batch 8), seed 0, Sun 22 Nov, 00:18 UTCfinished, eval accuracy 0.5615, config 4 of 4 (learning rate 0.00005, batch 8), seed 0, Sun 22 Nov, 00:55 UTCfinished, eval accuracy 0.9385, config 1 of 4 (learning rate 0.000005, batch 8), seed 2 of 3, Sun 22 Nov, 01:31 UTC

6 of 7 finished, 0.5615, config 4 of 4 (learning rate 0.00005, batch 8), seed 0, Sun 22 Nov, 00:55 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Sun 22 Nov, 01:31 UTC| finished0.9385339 of 339learning rate 0.000005, batch 8, seed 2 of 3
Sun 22 Nov, 00:55 UTC| finished0.5615339 of 339learning rate 0.00005, batch 8, seed 0
Sun 22 Nov, 00:18 UTC| finished0.8769339 of 339learning rate 0.00002, batch 8, seed 0
Sat 21 Nov, 23:42 UTC| finished0.3000339 of 339learning rate 0.0001, batch 8, seed 0
Fri 20 Nov, 00:08 UTC× failed0.50770learning rate 0.000005, batch 8, seed 2 of 3
Thu 19 Nov, 23:24 UTC| finished0.9154339 of 339learning rate 0.000005, batch 8, seed 1 of 3
Thu 19 Nov, 22:36 UTC| finished0.9000339 of 339learning rate 0.000005, batch 8, seed 0 of 3

Plate

Model
OmniJev/OneJev1.6-0.8B, 0.8B parameters revision c89933c
Data
Synthetic Discord triage set, 2,000-row draw, synthetic1800 train130 eval
Config
learning rate 0.00005, 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
8.4 swent public
40.0 seval 1
90.4 sphase training
90.4 sepoch 1
840.2 sepoch 2
1507.2 sepoch 3
2161.9 seval 2

Readings

69 rows as a table
StepSecondsTrain lossEval lossEval accuracyECELearning rate
040.0-1.06260.50770.2036-
590.40.8811---2.50e-5
10132.81.2637---5.00e-5
15171.50.9330---5.00e-5
20209.71.0433---4.99e-5
25250.50.8696---4.97e-5
30288.71.0660---4.95e-5
35328.91.2466---4.93e-5
40360.90.8084---4.90e-5
45394.41.0171---4.86e-5
50428.61.1040---4.82e-5
55462.91.0604---4.77e-5
60496.41.2160---4.72e-5
65533.30.9604---4.66e-5
70563.21.0501---4.60e-5
75593.31.2753---4.53e-5
80622.20.9840---4.46e-5
85658.90.9686---4.39e-5
90692.50.9332---4.31e-5
95722.51.0765---4.22e-5
100751.71.0883---4.13e-5
105783.31.0885---4.04e-5
110814.31.0476---3.94e-5
115840.21.0110---3.85e-5
120869.41.0043---3.74e-5
125896.90.9435---3.64e-5
130927.11.0436---3.53e-5
135956.41.1061---3.42e-5
140989.21.1220---3.31e-5
1451018.01.0303---3.19e-5
1501046.70.9912---3.08e-5
1551075.01.0946---2.96e-5
1601103.51.1181---2.85e-5
1651131.41.1572---2.73e-5
1701160.40.9687---2.61e-5
1751190.91.0485---2.49e-5
1801217.81.0357---2.37e-5
1851246.81.0659---2.25e-5
1901275.71.0530---2.13e-5
1951305.21.0524---2.01e-5
2001333.91.0888---1.90e-5
2051364.71.0417---1.78e-5
2101393.11.1022---1.67e-5
2151424.51.0362---1.56e-5
2201452.70.9893---1.45e-5
2251481.30.9614---1.34e-5
2301507.20.9311---1.24e-5
2351535.31.0511---1.13e-5
2401563.90.9276---1.04e-5
2451591.41.0832---9.41e-6
2501622.20.9207---8.50e-6
2551651.61.1106---7.62e-6
2601681.20.9514---6.78e-6
2651711.11.0777---5.99e-6
2701739.60.9771---5.23e-6
2751768.80.9698---4.53e-6
2801796.50.9316---3.86e-6
2851825.30.9689---3.25e-6
2901853.20.9956---2.69e-6
2951882.90.9250---2.17e-6
3001911.71.0879---1.71e-6
3051942.60.9337---1.31e-6
3101970.90.9873---9.52e-7
3152000.90.9819---6.54e-7
3202028.11.0111---4.10e-7
3252055.90.8832---2.23e-7
3302085.10.9050---9.23e-8
3352113.80.9137---1.82e-8
3392161.90.95220.97270.56150.16500.00e+0

GPU readings

69 rows as a table
StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
00100103.118.197.8
5024120.629.397.8
10065117.431.397.8
150100104.117.797.8
2000119.426.997.8
25052109.329.297.8
3000119.929.297.8
3500118.931.697.8
4000121.731.197.8
45073107.235.397.8
5000118.631.396.8
55038114.730.397.8
6007121.430.097.8
70091117.533.097.8
750100112.626.494.8
80068118.033.295.8
8500117.833.994.8
95064119.833.294.8
100063118.234.894.8
10500118.234.196.8
11000117.933.096.8
115010097.410.793.8
1200100118.634.894.8
1250100113.034.496.8
13000117.936.595.8
1350100121.027.395.8
140062118.541.794.8
14500118.441.795.8
15000118.241.794.8
1600100117.941.795.8
165091120.541.795.8
17000121.827.395.8
1750100114.934.193.8
18000117.634.194.8
18500120.434.195.8
190055117.536.394.8
19500118.936.394.8
2000100118.037.495.8
210093118.337.493.8
215058121.229.796.8
22000117.633.693.8
22500117.937.894.8
2300100117.837.895.8
235058120.937.894.8
2400100114.937.893.8
24507296.724.095.8
250063118.334.295.8
2600100113.932.995.8
26500118.737.595.8
2700100111.137.595.8
27500118.034.295.8
280094119.234.294.8
28500117.734.394.8
290065118.031.995.8
29500117.637.395.8
3000100112.933.993.8
3050100117.435.094.8
3150100114.334.596.8
32000118.034.592.8
3250100118.235.594.8
3300100115.438.093.8
33500118.738.195.8
3390100118.038.189.8

Worked out

ReadoutValueWorked out from
Run complete100%339 steps over 339
Step time6.19 s2073.8 s over 335 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute9.760 s over the step time of 6.19 s
Epoch interval708.4 sMean of 2 gaps between epoch marks
Train loss since first reading+0.07110.9522 at step 339 minus 0.8811 at step 5
Lowest train loss0.8084Lowest of 68 readings, at step 40
Eval accuracy, as shown56.2%0.5615 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.05380.5615 minus 0.5077
Eval loss since first eval−0.08990.9727 minus 1.0626
ECE since first eval−0.03860.1650 minus 0.2036
Eval change+0.0538Eval 2 at 0.5615 minus eval 1 at 0.5077
Train minus eval loss−0.02050.9522 minus 0.9727, both at step 339
Wrong answers57130 eval items minus 73 correct. 73 over 130 is the only count that gives 0.5615
Rank5 of 7By last eval accuracy, across this project's 7 runs with an eval
Best so far0.9385Highest last eval accuracy, run 7
Against the run before−0.31540.5615 minus 0.8769 on run 5
Lowest last train loss0.0000Lowest last train loss across this project's 6 runs, run 2
Items per second1.31 step over 6.19 s, times batch size 8
Train loss against run 5+0.95130.9522 minus 0.0009 on run 5, both at step 339
Learning rate since first reading−2.50e-50.00e+0 minus 2.50e-5
Peak memory per GPU, whole GPU41.7 GBHighest sensor reading of any GPU, GPU 0, counting every process on it