CITADEL
StandbyNothing is live right now.
Eval accuracy
%
no eval

Claviger router: Kev-4B (LoRA) (backfilled)

finishedRadeon 8060S

A model that routes a request to the one destination, of many, that should take it. 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, 20:40 UTC
ran for
113:48
config
learning rate 0.00002, batch 1, seed 1 of 3

run2/3step84/84epoch1/1seed1/3lr0.00002batch1modelkev-4b 4Brev6cfce5c

Loss train 0.0007 ▼ −0.1296 lowest 0.0000 at step 20 run before, run 1 ▲ +0.0007 step 84
Timebase
Range
Smoothing
Reference
84 readings, 30 min/div30 min/div30 min/div
012300:0030:0060:0090:00phase starting, 0.0 s to 81.3 sphase training, 81.3 s to 113:48epoch 1, 81.3 straining0.1303Lowest last train loss across this project's 3 runs, run 1lowest last 0.0000, run 10.0007012300:0030:0060:0090:00phase starting, 0.0 s to 81.3 sphase training, 81.3 s to 113:48epoch 1, 81.3 straining0.1303Lowest last train loss across this project's 3 runs, run 1lowest last 0.0000, run 10.0007▼ −0.1296012300:0030:0060:0090:00phase starting, 0.0 s to 81.3 sphase training, 81.3 s to 113:48epoch 1, 81.3 straining0.1303Lowest last train loss across this project's 3 runs, run 1lowest last 0.0000, run 10.0007▼ −0.1296

Elapsed

81.3 s to epoch 1, 6747.2 s to last reading

84 of 84 steps, 1 of 1 epochs, 100%

81.3 s per step, 0.7 steps a minute

Set-up

666 train

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

Learning rate

8.71e-9

▼ −1.64e-6 since first

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, -, config 1 of 1 (learning rate 0.00002, batch 1), seed 1 of 3, Tue 10 Nov, 20:40 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Tue 10 Nov, 22:34 UTC| finished-84 of 84learning rate 0.00002, batch 1, seed 2 of 3
Tue 10 Nov, 20:40 UTC| finished-84 of 84learning rate 0.00002, batch 1, seed 1 of 3
Tue 10 Nov, 18:45 UTC| finished-84 of 84learning rate 0.00002, batch 1, seed 0 of 3

Plate

Model
jaredpalmer/kev-4b, 4B parameters revision 6cfce5c
Data
Synthetic routing worlds, synthetic
Config
learning rate 0.00002, batch size 1, grad accum 8, epochs 1, optimizer adamw, seed 1
Hardware
Radeon 8060S, ROCm

Log

Since startMark
0.0 sphase starting
81.3 sphase training
81.3 sepoch 1
184 h 12 min laterwent public

Seeds

SeedStatusEval accuracy
seed 0| finished-
seed 1| finished-
seed 2| finished-

Readings

84 rows as a table
StepSecondsTrain lossLearning rate
181.30.13031.65e-6
2162.60.22614.06e-6
3243.90.00587.59e-6
4325.20.00701.16e-5
5406.50.01051.54e-5
6487.80.23961.84e-5
7569.00.00051.99e-5
8650.30.00172.00e-5
9731.60.00012.00e-5
10812.90.00281.99e-5
11894.20.00021.99e-5
12975.50.00021.98e-5
131056.80.00301.97e-5
141138.10.65071.96e-5
151219.40.02081.95e-5
161300.70.00001.94e-5
171382.00.00031.92e-5
181463.20.00041.90e-5
191544.50.00081.89e-5
201625.80.00001.87e-5
211707.10.00001.84e-5
221788.40.00001.82e-5
231869.70.00001.80e-5
241951.00.00271.77e-5
252032.30.04641.74e-5
262113.60.00001.72e-5
272194.90.00031.69e-5
282276.20.00021.66e-5
292357.50.00001.62e-5
302438.80.00001.59e-5
312520.00.00001.56e-5
322601.30.27561.52e-5
332682.60.00001.49e-5
342763.90.00041.45e-5
352845.20.00001.41e-5
362926.50.00001.37e-5
373007.80.00011.33e-5
383089.10.00001.29e-5
393170.40.00131.25e-5
403251.70.00001.21e-5
413333.00.00121.17e-5
423414.20.00001.13e-5
433495.50.00001.09e-5
443576.80.00001.05e-5
453658.10.00061.01e-5
463739.40.00009.67e-6
473820.70.00009.25e-6
483902.00.00008.84e-6
493983.30.00008.43e-6
504064.60.00008.02e-6
514145.90.07757.61e-6
524227.20.03047.21e-6
534308.50.00016.82e-6
544389.80.00006.42e-6
554471.00.00006.04e-6
564552.30.00005.66e-6
574633.60.00005.29e-6
584714.90.00004.93e-6
594796.20.00004.57e-6
604877.50.00144.23e-6
614958.80.00003.90e-6
625040.10.00003.57e-6
635121.40.00063.26e-6
645202.70.00002.96e-6
655284.00.00002.67e-6
665365.20.00002.39e-6
675446.50.00002.13e-6
685527.80.00001.88e-6
695609.10.00001.65e-6
705690.40.00111.42e-6
715771.70.00001.22e-6
725853.00.00001.03e-6
735934.30.00018.51e-7
746015.60.00006.91e-7
756096.90.00005.48e-7
766178.20.00004.20e-7
776259.50.00003.09e-7
786340.80.00122.15e-7
796422.00.00001.38e-7
806503.30.00007.77e-8
816584.60.00003.46e-8
826665.90.00008.71e-9
836747.20.00008.00e-11
846828.50.00078.71e-9

Worked out

ReadoutValueWorked out from
Run complete100%84 steps over 84
Step time81.3 s6747.2 s over 83 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute0.760 s over the step time of 81.29 s
Train loss since first reading−0.12960.0007 at step 84 minus 0.1303 at step 1
Lowest train loss0.0000Lowest of 84 readings, at step 20
Lowest last train loss0.0000Lowest last train loss across this project's 3 runs, run 1
Items per second0.01 step over 81.29 s, times batch size 1
Train loss against run 1+0.00070.0007 minus 0.0000 on run 1, both at step 84
Learning rate since first reading−1.64e-68.71e-9 minus 1.65e-6