CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Tue 10 Nov, 08:01 UTC
%
0.8000104 of 130 eval, 26 wrong
▲ +29.23 pts across 2 evals▲ +29.23 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Claviger: OneJev 0.8B (backfilled)

finishedRadeon 8060S

A small model that reads one Discord message and decides whether an agent should reply, and which one. Backfilled: recorded after the fact from the run's own logs and records. Times inside a run are estimated, and there are no GPU sensor readings.

started
Tue 10 Nov, 07:42 UTC
ran for
17:52
config
learning rate 0.000005, batch 8, seed 0 of 3

run1/6step150/150epoch5/5config1/4seed0/3lr0.000005batch8modelOneJev1.6-0.8B 0.8Brevc89933citems479+130

Loss train 0.0000 ▼ −1.3381 eval 1.6473 lowest 0.0000 at step 83
Timebase
Range
Smoothing
150 readings, 5 min/div5 min/div2 min/div
00.511.5200:0005:0010:0015:00phase starting, 0.0 s to 7.1 sphase training, 7.1 s to 17:52epoch 1, 7.1 sepoch 3, 07:16epoch 4, 10:50epoch 5, 14:25trainingE3E4E5−1.3381since first1.3381Lowest last train loss across this project's 6 runs, run 5lowest last 0.0000, run 50.000000.511.5200:0005:0010:0015:00phase starting, 0.0 s to 7.1 sphase training, 7.1 s to 17:52epoch 1, 7.1 sepoch 2, 03:41epoch 3, 07:16epoch 4, 10:50epoch 5, 14:25trainingepoch 2epoch 3epoch 4epoch 51.3381Lowest last train loss across this project's 6 runs, run 5lowest last 0.0000, run 5eval 1.64730.0000▼ −1.338100.511.5200:0002:0004:0006:0008:0010:0012:0014:0016:00phase starting, 0.0 s to 7.1 sphase training, 7.1 s to 17:52epoch 1, 7.1 sepoch 2, 03:41epoch 3, 07:16epoch 4, 10:50epoch 5, 14:25trainingepoch 2epoch 3epoch 4epoch 51.3381Lowest last train loss across this project's 6 runs, run 5lowest last 0.0000, run 5eval 1.64730.0000▼ −1.3381

Eval measures

Eval losslower is better
1.6473▲ +0.5847 since first
02 →
ECEcalibration error, lower is better
0.1925▼ −0.0111 since first
01

Elapsed

214.5 s

epoch interval

150 of 150 steps, 5 of 5 epochs, 100%

7.15 s per step, 8.4 steps a minute

Eval accuracy by eval

+0.2923

▲ on the eval before

eval 1 0.5077, eval 2 0.8000

Against the other runs

3 of 6

by last eval accuracy

best so far 0.8615, run 6

Train minus eval

−1.6473

step 150

train 0.0000, eval 1.6473

Set-up

479 train, 130 eval

synthetic, learning rate 0.000005, batch size 8, grad accum 2, epochs 5, optimizer adamw, seed 0

Seed spread, observed

0.0615
0.800.90

3 finished seeds

Learning rate

0.00e+0

▼ −1.25e-6 since first

On one axis

phase starting, 0.0 s to 7.1 sphase training, 7.1 s to 17:52E2E3E4E5traininglowest last 0.0000Train loss0.0000▼ −1.33812.000.001.33810.0000last ▼ −1.3381best 0.8615, run 6Last-value range, 6 runs, 0.5615 to 0.8615last-value range, n 6Eval accuracy0.8000▲ +0.29231.000.400.50770.8000last ▲ +0.2923Learning rate0.00e+0▼ −1.25e-66.0e-60.0e+01.25e-60.00e+0last ▼ −1.25e-600:0002:0004:0006:0008:0010:0012:0014:0016:00

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 6 finished, 0.8000, config 1 of 4 (learning rate 0.000005, batch 8), seed 0 of 3, Tue 10 Nov, 07:42 UTC

◂
Runs as a table
StartedStatusEval accuracyStepsConfig
Tue 10 Nov, 09:19 UTC| finished0.8615150 of 150learning rate 0.000005, batch 8, seed 2 of 3
Tue 10 Nov, 08:59 UTC| finished0.8154150 of 150learning rate 0.000005, batch 8, seed 1 of 3
Tue 10 Nov, 08:40 UTC| finished0.5615150 of 150learning rate 0.00005, batch 8, seed 0
Tue 10 Nov, 08:21 UTC| finished0.7769150 of 150learning rate 0.00002, batch 8, seed 0
Tue 10 Nov, 08:01 UTC| finished0.7923150 of 150learning rate 0.00001, batch 8, seed 0
Tue 10 Nov, 07:42 UTC| finished0.8000150 of 150learning 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, mixed rosters, synthetic479 train130 eval
Config
learning rate 0.000005, batch size 8, grad accum 2, epochs 5, optimizer adamw, seed 0
Hardware
Radeon 8060S, ROCm

Log

Since startMark
0.0 sphase starting
0.0 seval 1
7.1 sphase training
7.1 sepoch 1
221.6 sepoch 2
436.1 sepoch 3
650.5 sepoch 4
865.0 sepoch 5
1072.3 seval 2
197 h 9 min laterwent public

Seeds

SeedStatusEval accuracy
seed 0| finished0.8000
seed 1| finished0.8154
seed 2| finished0.8615

observed spread 0.0615, 3 finished seeds

Readings

151 rows as a table

rows 1 to 100 of 151

StepSecondsTrain lossEval lossEval accuracyECELearning rate
00.0-1.06260.50770.2036-
17.11.3381---1.25e-6
214.30.7801---2.50e-6
321.40.8108---3.75e-6
428.60.6146---5.00e-6
535.70.7017---5.00e-6
642.90.6792---5.00e-6
750.00.7296---4.99e-6
857.20.6400---4.99e-6
964.30.5494---4.99e-6
1071.50.5852---4.98e-6
1178.60.5479---4.97e-6
1285.80.4713---4.96e-6
1392.90.4431---4.95e-6
14100.10.2661---4.94e-6
15107.20.8296---4.93e-6
16114.41.2756---4.92e-6
17121.51.7324---4.90e-6
18128.70.0676---4.89e-6
19135.81.6610---4.87e-6
20143.01.0162---4.85e-6
21150.10.7377---4.83e-6
22157.30.4564---4.81e-6
23164.40.7171---4.79e-6
24171.60.6172---4.77e-6
25178.70.6961---4.75e-6
26185.90.4815---4.73e-6
27193.00.6626---4.70e-6
28200.20.4575---4.67e-6
29207.30.4474---4.65e-6
30214.50.7909---4.62e-6
31221.60.3245---4.59e-6
32228.80.1647---4.56e-6
33235.90.3062---4.53e-6
34243.10.1584---4.50e-6
35250.20.4116---4.46e-6
36257.40.4367---4.43e-6
37264.50.4148---4.40e-6
38271.60.2613---4.36e-6
39278.80.2655---4.32e-6
40285.90.2651---4.29e-6
41293.10.4007---4.25e-6
42300.20.3510---4.21e-6
43307.40.2913---4.17e-6
44314.50.3210---4.13e-6
45321.70.2023---4.09e-6
46328.80.0387---4.05e-6
47336.00.1091---4.00e-6
48343.10.2250---3.96e-6
49350.30.0944---3.92e-6
50357.40.3514---3.87e-6
51364.60.2569---3.83e-6
52371.70.4178---3.78e-6
53378.90.3040---3.73e-6
54386.00.3517---3.69e-6
55393.20.0514---3.64e-6
56400.30.4425---3.59e-6
57407.50.1280---3.54e-6
58414.60.1372---3.49e-6
59421.80.1248---3.44e-6
60428.90.0832---3.39e-6
61436.10.0893---3.34e-6
62443.20.0188---3.29e-6
63450.40.0536---3.24e-6
64457.50.0554---3.19e-6
65464.70.0203---3.14e-6
66471.80.0119---3.09e-6
67479.00.2521---3.03e-6
68486.10.0064---2.98e-6
69493.30.1168---2.93e-6
70500.40.0010---2.88e-6
71507.60.0024---2.82e-6
72514.70.0569---2.77e-6
73521.90.0036---2.71e-6
74529.00.0002---2.66e-6
75536.10.0006---2.61e-6
76543.30.0001---2.55e-6
77550.40.0001---2.50e-6
78557.60.0001---2.45e-6
79564.70.0004---2.39e-6
80571.90.5331---2.34e-6
81579.00.8235---2.29e-6
82586.20.0006---2.23e-6
83593.30.0000---2.18e-6
84600.50.2526---2.12e-6
85607.60.2237---2.07e-6
86614.81.0371---2.02e-6
87621.90.0000---1.97e-6
88629.10.0017---1.91e-6
89636.20.0001---1.86e-6
90643.40.0001---1.81e-6
91650.50.0000---1.76e-6
92657.70.0000---1.71e-6
93664.80.0000---1.66e-6
94672.00.0000---1.61e-6
95679.10.0000---1.56e-6
96686.30.0000---1.51e-6
97693.40.0000---1.46e-6
98700.60.0000---1.41e-6
99707.70.0002---1.36e-6

Worked out

ReadoutValueWorked out from
Run complete100%150 steps over 150
Step time7.15 s1065.2 s over 149 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute8.460 s over the step time of 7.15 s
Epoch interval214.5 sMean of 4 gaps between epoch marks
Train loss since first reading−1.33810.0000 at step 150 minus 1.3381 at step 1
Lowest train loss0.0000Lowest of 150 readings, at step 83
Eval accuracy, as shown80.0%0.8000 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.29230.8000 minus 0.5077
Eval loss since first eval+0.58471.6473 minus 1.0626
ECE since first eval−0.01110.1925 minus 0.2036
Eval change+0.2923Eval 2 at 0.8000 minus eval 1 at 0.5077
Train minus eval loss−1.64730.0000 minus 1.6473, both at step 150
Wrong answers26130 eval items minus 104 correct. 104 over 130 is the only count that gives 0.8000
Rank3 of 6By last eval accuracy, across this project's 6 runs with an eval
Best so far0.8615Highest last eval accuracy, run 6
Lowest last train loss0.0000Lowest last train loss across this project's 6 runs, run 5
Seed spread, observed0.0615High 0.8615 minus low 0.8000 across 3 finished seeds
Items per second1.11 step over 7.15 s, times batch size 8
Learning rate since first reading−1.25e-60.00e+0 minus 1.25e-6