CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Tue 10 Nov, 10:30 UTC
%
0.7846102 of 130 eval, 28 wrong
▲ +16.15 pts across 2 evals▲ +16.15 pts on the eval before
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Claviger: Kev-4B (LoRA) (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, 10:04 UTC
ran for
20:25
config
learning rate 0.00002, batch 1, seed 1 of 3

run2/3step120/120epoch2/2seed1/3lr0.00002batch1modelkev-4b 4Brev6cfce5citems479+130

Loss train 0.0329 ▼ −0.6204 eval 0.6547 lowest 0.0011 at step 112 run before, run 1 ▼ −0.0957 step 120
Timebase
Range
Smoothing
Reference
120 readings, 10 min/div5 min/div5 min/div
00.511.5200:0010:0020:00phase starting, 0.0 s to 10.2 sphase training, 10.2 s to 20:25epoch 1, 10.2 sepoch 2, 10:23trainingepoch 20.6533Lowest last train loss across this project's 3 runs, run 3lowest last 0.0102, run 30.032900.511.5200:0005:0010:0015:0020:00phase starting, 0.0 s to 10.2 sphase training, 10.2 s to 20:25epoch 1, 10.2 sepoch 2, 10:23trainingepoch 20.6533Lowest last train loss across this project's 3 runs, run 3lowest last 0.0102, run 3eval 0.65470.0329▼ −0.620400.511.5200:0005:0010:0015:0020:00phase starting, 0.0 s to 10.2 sphase training, 10.2 s to 20:25epoch 1, 10.2 sepoch 2, 10:23trainingepoch 20.6533Lowest last train loss across this project's 3 runs, run 3lowest last 0.0102, run 3eval 0.65470.0329▼ −0.6204

Eval measures

Eval losslower is better
0.6547▼ −0.1482 since first
02 →
ECEcalibration error, lower is better
0.1599▲ +0.0741 since first
01

Elapsed

612.8 s

epoch interval

10.2 s to epoch 1, 612.8 s to epoch 2, 602.5 s to eval 2

120 of 120 steps, 2 of 2 epochs, 100%

10.2 s per step, 5.9 steps a minute

Eval accuracy by eval

+0.1615

▲ on the eval before

eval 1 0.6231, eval 2 0.7846

Against the other runs

3 of 3

by last eval accuracy

best so far 0.8308, run 1

▼ −0.0462 against run 1

Train minus eval

−0.6218

step 120

train 0.0329, eval 0.6547

Set-up

479 train, 130 eval

synthetic, learning rate 0.00002, batch size 1, grad accum 8, epochs 2, optimizer adamw, seed 1

Seed spread, observed

0.0462
0.750.85

3 finished seeds

Learning rate

4.31e-9

▼ −1.18e-6 since first

On one axis

phase starting, 0.0 s to 10.2 sphase training, 10.2 s to 20:25E2traininglowest last 0.0102Train loss0.0329▼ −0.62042.000.000.65330.0329last ▼ −0.6204best 0.8308, run 1Last-value range, 3 runs, 0.7846 to 0.8308last-value range, n 3Eval accuracy0.7846▲ +0.16150.900.600.62310.7846last ▲ +0.1615Learning rate4.31e-9▼ −1.18e-62.0e-50.0e+01.19e-64.31e-9last ▼ −1.18e-600:0005:0010:0015:0020:00

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

2 of 3 finished, 0.7846, config 1 of 1 (learning rate 0.00002, batch 1), seed 1 of 3, Tue 10 Nov, 10:04 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Tue 10 Nov, 10:30 UTC| finished0.7923120 of 120learning rate 0.00002, batch 1, seed 2 of 3
Tue 10 Nov, 10:04 UTC| finished0.7846120 of 120learning rate 0.00002, batch 1, seed 1 of 3
Tue 10 Nov, 09:38 UTC| finished0.8308120 of 120learning rate 0.00002, batch 1, seed 0 of 3

Plate

Model
jaredpalmer/kev-4b, 4B parameters revision 6cfce5c
Data
Synthetic Discord triage set, mixed rosters, synthetic479 train130 eval
Config
learning rate 0.00002, batch size 1, grad accum 8, epochs 2, optimizer adamw, seed 1
Hardware
Radeon 8060S, ROCm

Log

Since startMark
0.0 sphase starting
0.0 seval 1
10.2 sphase training
10.2 sepoch 1
623.0 sepoch 2
1225.5 seval 2
194 h 47 min laterwent public

Seeds

SeedStatusEval accuracy
seed 0| finished0.8308
seed 1| finished0.7846
seed 2| finished0.7923

observed spread 0.0462, 3 finished seeds

Readings

121 rows as a table

rows 1 to 100 of 121

StepSecondsTrain lossEval lossEval accuracyECELearning rate
00.0-0.80290.62310.0858-
110.20.6533---1.19e-6
220.41.1269---2.32e-6
330.60.8462---4.11e-6
440.90.8272---6.41e-6
551.11.1196---9.03e-6
661.30.7511---1.18e-5
771.51.0033---1.44e-5
881.70.5940---1.67e-5
991.91.0145---1.85e-5
10102.11.2676---1.96e-5
11112.30.6899---2.00e-5
12122.60.7456---2.00e-5
13132.80.4762---2.00e-5
14143.00.4868---2.00e-5
15153.20.3005---1.99e-5
16163.40.3522---1.99e-5
17173.60.7966---1.98e-5
18183.80.2708---1.98e-5
19194.00.3741---1.97e-5
20204.20.4664---1.97e-5
21214.50.4927---1.96e-5
22224.70.4153---1.95e-5
23234.90.2792---1.94e-5
24245.10.7616---1.93e-5
25255.31.0587---1.92e-5
26265.50.3542---1.91e-5
27275.71.3055---1.89e-5
28285.91.0341---1.88e-5
29296.21.6473---1.87e-5
30306.40.6024---1.85e-5
31316.60.3780---1.84e-5
32326.80.4507---1.82e-5
33337.00.2168---1.80e-5
34347.20.6059---1.78e-5
35357.40.1786---1.77e-5
36367.60.2932---1.75e-5
37377.90.2655---1.73e-5
38388.10.0708---1.71e-5
39398.30.4201---1.69e-5
40408.50.2478---1.66e-5
41418.70.4114---1.64e-5
42428.90.5729---1.62e-5
43439.10.6896---1.60e-5
44449.30.3531---1.57e-5
45459.60.3731---1.55e-5
46469.80.2539---1.52e-5
47480.00.8122---1.50e-5
48490.20.1118---1.47e-5
49500.40.3492---1.45e-5
50510.60.7312---1.42e-5
51520.80.2831---1.40e-5
52531.10.2975---1.37e-5
53541.30.8927---1.34e-5
54551.50.1630---1.31e-5
55561.70.2432---1.29e-5
56571.90.2889---1.26e-5
57582.10.6269---1.23e-5
58592.30.2288---1.20e-5
59602.50.2162---1.17e-5
60612.80.5958---1.14e-5
61623.00.4508---1.12e-5
62633.20.2938---1.09e-5
63643.40.5872---1.06e-5
64653.60.0735---1.03e-5
65663.80.7171---1.00e-5
66674.00.1248---9.71e-6
67684.20.3308---9.42e-6
68694.40.2930---9.13e-6
69704.70.1517---8.84e-6
70714.90.2384---8.55e-6
71725.10.0628---8.26e-6
72735.30.0694---7.98e-6
73745.50.0561---7.69e-6
74755.70.1541---7.41e-6
75765.90.0162---7.13e-6
76776.10.1532---6.85e-6
77786.40.1133---6.58e-6
78796.60.0371---6.31e-6
79806.80.0132---6.04e-6
80817.00.0274---5.77e-6
81827.20.0802---5.51e-6
82837.40.0965---5.25e-6
83847.60.2828---5.00e-6
84857.80.6673---4.75e-6
85868.10.1497---4.50e-6
86878.30.3425---4.26e-6
87888.50.3733---4.03e-6
88898.70.2429---3.80e-6
89908.90.0109---3.57e-6
90919.10.0129---3.35e-6
91929.31.2615---3.14e-6
92939.60.1913---2.93e-6
93949.80.4914---2.73e-6
94960.00.0191---2.53e-6
95970.20.2053---2.34e-6
96980.40.0434---2.16e-6
97990.60.4673---1.98e-6
981000.80.1261---1.81e-6
991011.00.0284---1.65e-6

Worked out

ReadoutValueWorked out from
Run complete100%120 steps over 120
Step time10.2 s1215.3 s over 119 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute5.960 s over the step time of 10.21 s
Epoch interval612.8 sThe one gap between epoch marks
Train loss since first reading−0.62040.0329 at step 120 minus 0.6533 at step 1
Lowest train loss0.0011Lowest of 120 readings, at step 112
Eval accuracy, as shown78.5%0.7846 times 100, to one decimal. Changes on that plate are percentage points
Eval accuracy since first eval+0.16150.7846 minus 0.6231
Eval loss since first eval−0.14820.6547 minus 0.8029
ECE since first eval+0.07410.1599 minus 0.0858
Eval change+0.1615Eval 2 at 0.7846 minus eval 1 at 0.6231
Train minus eval loss−0.62180.0329 minus 0.6547, both at step 120
Wrong answers28130 eval items minus 102 correct. 102 over 130 is the only count that gives 0.7846
Rank3 of 3By last eval accuracy, across this project's 3 runs with an eval
Best so far0.8308Highest last eval accuracy, run 1
Against the run before−0.04620.7846 minus 0.8308 on run 1
Lowest last train loss0.0102Lowest last train loss across this project's 3 runs, run 3
Seed spread, observed0.0462High 0.8308 minus low 0.7846 across 3 finished seeds
Items per second0.11 step over 10.21 s, times batch size 1
Train loss against run 1−0.09570.0329 minus 0.1287 on run 1, both at step 120
Learning rate since first reading−1.18e-64.31e-9 minus 1.19e-6