CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Thu 19 Nov, 22:36 UTC
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Claviger

finishedRadeon 8060S

A small model that reads one Discord message and decides whether an agent should reply, and which one. Fine-tuned from Laya on invented chat.

started
Thu 19 Nov, 22:21 UTC
ran for
14:42
config
batch 8, seed 2 of 3

run17/22step336/339epoch2/3config7/11seed2/3batch8modellaya 421Mrev7b928d8items1800+130

Loss train 0.0000 ▼ −0.5109 eval 0.4675 lowest -0.2792 at step 138 run before, run 16 ▼ −0.2394 step 336 Laya's objective: soft cross-entropy plus a policy-gradient term, so it can go below zero.
Timebase
Range
Smoothing
Reference
66 readings, 5 min/div5 min/div2 min/div
-10123400:0005:0010:00phase starting, 0.0 s to 45.3 sphase training, 45.3 s to 14:34phase eval, 14:34 to 14:42epoch 1, 04:57epoch 2, 09:18epoch 3, 13:38trainingepoch 1epoch 2E30.5109Lowest last train loss across this project's 22 runs, run 19lowest last -0.0937, run 19339 steps0.0000-10123400:0005:0010:00phase starting, 0.0 s to 45.3 sphase training, 45.3 s to 14:34phase eval, 14:34 to 14:42epoch 1, 04:57epoch 2, 09:18epoch 3, 13:38trainingepoch 1epoch 2E30.5109Lowest last train loss across this project's 22 runs, run 19lowest last -0.0937, run 19eval 0.4675339 steps0.0000▼ −0.5109-10123400:0002:0004:0006:0008:0010:0012:0014:00phase starting, 0.0 s to 45.3 sphase training, 45.3 s to 14:34phase eval, 14:34 to 14:42epoch 1, 04:57epoch 2, 09:18epoch 3, 13:38trainingepoch 1epoch 2epoch 30.5109Lowest last train loss across this project's 22 runs, run 19lowest last -0.0937, run 19eval 0.4675339 stepsrun 160.0000▼ −0.5109

Eval measures

Eval losslower is better
0.4675
02 →
ECEcalibration error, lower is better
0.1020
01
Brierlower is better
0.2253
02 →

Elapsed

260.9 s

epoch interval

336 of 339 steps, 2 of 3 epochs, 99.1%

2.31 s per step, 25.9 steps a minute

Against the other runs

4 of 22

by last eval accuracy

best so far 0.9231, run 19

▲ +0.0000 against run 16

Train minus eval

−0.4675

step 336

train 0.0000, eval 0.4675

Set-up

1800 train, 130 eval

synthetic, batch size 8, grad accum 2, epochs 3, loss rlcd, seed 2

Seed spread, observed

0.0077
0.850.90

3 finished seeds

Samples per second

1.89

▼ −5.11 since first

Learning rate

1.00e-1

▲ +1.00e-1 since first

GPU memory, this run

4.9 GB

On one axis

phase starting, 0.0 s to 45.3 sphase training, 45.3 s to 14:34phase eval, 14:34 to 14:42E1E2E3traininglowest last -0.0937Train loss0.0000▼ −0.51094.00-2.000.5109Samples per second1.89▼ −5.11806.99Learning rate1.00e-1▲ +1.00e-11.0e-10.0e+02.50e-5Power, whole chip, CPU and GPU53 W▲ +0 W55 W52 W53 WHottest, one board zone71 °C▲ +3 °C72 °C64 °C68 °C00:0002:0004:0006:0008:0010:0012:0014:00
GPUpower: whole chip, CPU and GPUtemperature: one board zone

GPU 0

Wwhole chip, CPU and GPU

53 W whole chip, CPU and GPU

use0%
memory, whole GPU10.1 GB

this run's peak 8.0 GB

71 °C one board zone

history 0 to 54 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
0.950.750.950.75finished, eval accuracy 0.8077, config 1 of 11 (batch 8), seed 2, Mon 2 Nov, 13:19 UTCfinished, eval accuracy 0.8231, config 2 of 11 (batch 8), seed 2 of 3, Mon 2 Nov, 14:03 UTCfinished, eval accuracy 0.8231, config 3 of 11 (batch 8), seed 2, Mon 2 Nov, 14:09 UTCfinished, eval accuracy 0.7846, config 2 of 11 (batch 8), seed 0 of 3, Mon 2 Nov, 14:16 UTCfinished, eval accuracy 0.8385, config 2 of 11 (batch 8), seed 1 of 3, Mon 2 Nov, 14:22 UTCfinished, eval accuracy 0.7923, config 4 of 11 (batch 8), seed 0 of 3, Mon 2 Nov, 18:15 UTCfinished, eval accuracy 0.8000, config 4 of 11 (batch 8), seed 1 of 3, Mon 2 Nov, 18:22 UTCfinished, eval accuracy 0.8154, config 4 of 11 (batch 8), seed 2 of 3, Mon 2 Nov, 18:28 UTCfinished, eval accuracy 0.7923, config 5 of 11 (batch 8), seed 0 of 3, Mon 2 Nov, 18:46 UTCfinished, eval accuracy 0.8385, config 5 of 11 (batch 8), seed 1 of 3, Mon 2 Nov, 18:52 UTCfinished, eval accuracy 0.8000, config 5 of 11 (batch 8), seed 2 of 3, Mon 2 Nov, 18:59 UTCfinished, eval accuracy 0.8231, config 6 of 11 (batch 8), seed 0 of 3, Tue 3 Nov, 00:59 UTCfinished, eval accuracy 0.8231, config 6 of 11 (batch 8), seed 1 of 3, Tue 3 Nov, 01:13 UTCfinished, eval accuracy 0.8000, config 6 of 11 (batch 8), seed 2 of 3, Tue 3 Nov, 01:26 UTCfinished at step 336 of 339, eval accuracy 0.8769, config 7 of 11 (batch 8), seed 0 of 3, Thu 19 Nov, 21:49 UTCfinished at step 336 of 339, eval accuracy 0.8846, config 7 of 11 (batch 8), seed 1 of 3, Thu 19 Nov, 22:06 UTCfinished at step 336 of 339, eval accuracy 0.8846, config 7 of 11 (batch 8), seed 2 of 3, Thu 19 Nov, 22:21 UTCfinished at step 336 of 339, eval accuracy 0.8615, config 8 of 11 (learning rate 0.0001, batch 8), seed 0, Sat 21 Nov, 23:04 UTCfinished at step 336 of 339, eval accuracy 0.9231, config 9 of 11 (learning rate 0.00002, batch 8), seed 0, Sat 21 Nov, 23:17 UTCfinished at step 336 of 339, eval accuracy 0.8538, config 10 of 11 (learning rate 0.00005, batch 8), seed 0, Sat 21 Nov, 23:29 UTCfinished at step 336 of 339, eval accuracy 0.9231, config 11 of 11 (learning rate 0.00002, batch 8), seed 1 of 2, Sun 22 Nov, 03:49 UTCfinished at step 336 of 339, eval accuracy 0.8846, config 11 of 11 (learning rate 0.00002, batch 8), seed 2 of 2, Sun 22 Nov, 04:01 UTC0.950.75finished, eval accuracy 0.8077, config 1 of 11 (batch 8), seed 2, Mon 2 Nov, 13:19 UTCfinished, eval accuracy 0.8231, config 2 of 11 (batch 8), seed 2 of 3, Mon 2 Nov, 14:03 UTCfinished, eval accuracy 0.8231, config 3 of 11 (batch 8), seed 2, Mon 2 Nov, 14:09 UTCfinished, eval accuracy 0.7846, config 2 of 11 (batch 8), seed 0 of 3, Mon 2 Nov, 14:16 UTCfinished, eval accuracy 0.8385, config 2 of 11 (batch 8), seed 1 of 3, Mon 2 Nov, 14:22 UTCfinished, eval accuracy 0.7923, config 4 of 11 (batch 8), seed 0 of 3, Mon 2 Nov, 18:15 UTCfinished, eval accuracy 0.8000, config 4 of 11 (batch 8), seed 1 of 3, Mon 2 Nov, 18:22 UTCfinished, eval accuracy 0.8154, config 4 of 11 (batch 8), seed 2 of 3, Mon 2 Nov, 18:28 UTCfinished, eval accuracy 0.7923, config 5 of 11 (batch 8), seed 0 of 3, Mon 2 Nov, 18:46 UTCfinished, eval accuracy 0.8385, config 5 of 11 (batch 8), seed 1 of 3, Mon 2 Nov, 18:52 UTCfinished, eval accuracy 0.8000, config 5 of 11 (batch 8), seed 2 of 3, Mon 2 Nov, 18:59 UTCfinished, eval accuracy 0.8231, config 6 of 11 (batch 8), seed 0 of 3, Tue 3 Nov, 00:59 UTCfinished, eval accuracy 0.8231, config 6 of 11 (batch 8), seed 1 of 3, Tue 3 Nov, 01:13 UTCfinished, eval accuracy 0.8000, config 6 of 11 (batch 8), seed 2 of 3, Tue 3 Nov, 01:26 UTCfinished at step 336 of 339, eval accuracy 0.8769, config 7 of 11 (batch 8), seed 0 of 3, Thu 19 Nov, 21:49 UTCfinished at step 336 of 339, eval accuracy 0.8846, config 7 of 11 (batch 8), seed 1 of 3, Thu 19 Nov, 22:06 UTCfinished at step 336 of 339, eval accuracy 0.8846, config 7 of 11 (batch 8), seed 2 of 3, Thu 19 Nov, 22:21 UTCfinished at step 336 of 339, eval accuracy 0.8615, config 8 of 11 (learning rate 0.0001, batch 8), seed 0, Sat 21 Nov, 23:04 UTCfinished at step 336 of 339, eval accuracy 0.9231, config 9 of 11 (learning rate 0.00002, batch 8), seed 0, Sat 21 Nov, 23:17 UTCfinished at step 336 of 339, eval accuracy 0.8538, config 10 of 11 (learning rate 0.00005, batch 8), seed 0, Sat 21 Nov, 23:29 UTCfinished at step 336 of 339, eval accuracy 0.9231, config 11 of 11 (learning rate 0.00002, batch 8), seed 1 of 2, Sun 22 Nov, 03:49 UTCfinished at step 336 of 339, eval accuracy 0.8846, config 11 of 11 (learning rate 0.00002, batch 8), seed 2 of 2, Sun 22 Nov, 04:01 UTC

17 of 22 finished at step 336 of 339, 0.8846, config 7 of 11 (batch 8), seed 2 of 3, Thu 19 Nov, 22:21 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Sun 22 Nov, 04:01 UTC| finished0.8846336 of 339learning rate 0.00002, batch 8, seed 2 of 2
Sun 22 Nov, 03:49 UTC| finished0.9231336 of 339learning rate 0.00002, batch 8, seed 1 of 2
Sat 21 Nov, 23:29 UTC| finished0.8538336 of 339learning rate 0.00005, batch 8, seed 0
Sat 21 Nov, 23:17 UTC| finished0.9231336 of 339learning rate 0.00002, batch 8, seed 0
Sat 21 Nov, 23:04 UTC| finished0.8615336 of 339learning rate 0.0001, batch 8, seed 0
Thu 19 Nov, 22:21 UTC| finished0.8846336 of 339batch 8, seed 2 of 3
Thu 19 Nov, 22:06 UTC| finished0.8846336 of 339batch 8, seed 1 of 3
Thu 19 Nov, 21:49 UTC| finished0.8769336 of 339batch 8, seed 0 of 3
Tue 3 Nov, 01:26 UTC| finished0.8000150 of 150batch 8, seed 2 of 3
Tue 3 Nov, 01:13 UTC| finished0.8231150 of 150batch 8, seed 1 of 3
Tue 3 Nov, 00:59 UTC| finished0.8231150 of 150batch 8, seed 0 of 3
Mon 2 Nov, 18:59 UTC| finished0.8000150 of 150batch 8, seed 2 of 3
Mon 2 Nov, 18:52 UTC| finished0.8385150 of 150batch 8, seed 1 of 3
Mon 2 Nov, 18:46 UTC| finished0.7923150 of 150batch 8, seed 0 of 3
Mon 2 Nov, 18:28 UTC| finished0.8154150 of 150batch 8, seed 2 of 3
Mon 2 Nov, 18:22 UTC| finished0.8000150 of 150batch 8, seed 1 of 3
Mon 2 Nov, 18:15 UTC| finished0.7923150 of 150batch 8, seed 0 of 3
Mon 2 Nov, 14:22 UTC| finished0.8385150 of 150batch 8, seed 1 of 3
Mon 2 Nov, 14:16 UTC| finished0.7846150 of 150batch 8, seed 0 of 3
Mon 2 Nov, 14:09 UTC| finished0.8231150 of 150batch 8, seed 2
Mon 2 Nov, 14:03 UTC| finished0.8231150 of 150batch 8, seed 2 of 3
Mon 2 Nov, 13:19 UTC| finished0.8077150 of 150batch 8, seed 2

Plate

Model
convaiinnovations/laya, 421M parameters revision 7b928d8
Data
Synthetic Discord triage set, 2,000-row draw, synthetic1800 train130 eval
Config
batch size 8, grad accum 2, epochs 3, loss rlcd, seed 2
Hardware
Radeon 8060S, ROCm

Log

Since startMark
0.0 sphase starting
14.6 swent public
45.3 sphase training
297.0 sepoch 1
558.5 sepoch 2
818.8 sepoch 3
874.0 sphase eval
874.0 seval 1

Seeds

SeedStatusEval accuracy
seed 0| finished0.8769
seed 1| finished0.8846
seed 2| finished0.8846

observed spread 0.0077, 3 finished seeds

Readings

67 rows as a table
StepSecondsTrain lossEval lossEval accuracyECEBrierSamples per secondLearning rateGPU GB, this run
045.1-----6.9942.50e-56.5
556.50.5109----7.0512.50e-56.5
1067.50.6555----7.2452.49e-56.5
1579.50.7720----6.6822.48e-56.5
2091.50.6481----6.6622.47e-56.5
25103.70.6074----6.5542.46e-56.5
30115.2-0.0133----6.9792.44e-56.5
35127.10.5085----6.7282.42e-56.5
40138.70.4149----6.8712.40e-56.5
45150.50.7315----6.7902.38e-56.5
50163.00.2640----6.3812.35e-56.5
55174.60.6905----6.9252.33e-56.5
60186.20.3082----6.8842.30e-56.5
65198.10.3724----6.7432.26e-56.5
70209.30.1560----7.1172.23e-56.5
75220.50.3385----7.1312.19e-56.5
80232.50.2969----6.6742.15e-56.5
85244.10.7657----6.8942.11e-56.5
90255.6-0.2090----6.9962.07e-56.5
95267.20.8787----6.8962.03e-56.5
100279.20.4089----6.6271.98e-56.5
105290.90.5200----6.8791.94e-56.5
110304.81.4797----5.7301.89e-56.5
118316.6-0.1627----6.8141.84e-56.5
123327.90.2009----7.0601.79e-56.5
128339.70.4638----6.7701.74e-56.5
133351.10.2906----7.0411.69e-56.5
138363.0-0.2792----6.7241.63e-56.5
143374.41.1844----6.9991.58e-56.5
148386.43.2357----6.6931.53e-56.5
153398.1-0.0399----6.7801.47e-56.5
158409.00.7163----7.3511.42e-56.5
163420.51.1008----6.9481.36e-56.5
168432.30.1431----6.7901.31e-56.5
173443.80.0020----6.9651.25e-56.5
178455.4-0.2285----6.8961.19e-56.5
183466.40.5570----7.2921.14e-56.5
188478.10.6049----6.8511.08e-56.5
193489.72.1410----6.8851.03e-56.5
198501.4-0.0125----6.8199.76e-66.5
203512.9-0.0255----6.9919.23e-66.5
208523.7-0.0215----7.3868.70e-66.5
213534.8-0.0020----7.1858.19e-66.5
218546.00.0061----7.1857.69e-66.5
223571.20.1865----5.6616.71e-66.5
231583.40.0000----6.5746.25e-66.5
236594.5-0.0813----7.2225.79e-66.5
241605.70.0000----7.1585.36e-66.5
246616.61.4201----7.3224.94e-66.5
251628.10.0000----6.9214.53e-66.5
256639.2-0.0408----7.2314.15e-66.5
261651.00.0000----6.8063.78e-66.5
266662.70.2341----6.7883.44e-66.5
271673.40.0000----7.4833.11e-66.5
276685.30.0000----6.7682.81e-66.5
281696.40.0321----7.1962.53e-66.5
286707.30.0000----7.3442.27e-66.5
291719.20.0843----6.7282.03e-66.5
296730.30.0000----7.1751.82e-66.5
301741.90.0000----6.8731.63e-66.5
306753.40.6366----6.9611.46e-66.5
311764.50.0000----7.2181.32e-66.5
316776.30.0250----6.7941.21e-66.5
321787.70.0000----7.0221.12e-66.5
326799.40.0677----6.8611.05e-66.5
331810.8-0.1984----6.9891.01e-66.5
336882.60.00000.46750.88460.10200.22531.8871.00e-14.9

GPU readings

67 rows as a table
StepGPUUse %Power W, whole chip, CPU and GPUMemory GB, whole GPUTemp C, one board zone
00053.03.667.8
1004753.79.665.8
2001553.09.765.8
3502353.19.765.8
500653.010.066.8
650953.310.066.8
800152.610.066.8
9501853.110.066.8
10501153.010.066.8
11803654.110.066.8
1330552.710.066.8
14803052.910.066.8
1630653.510.066.8
1780852.910.066.8
19301953.110.066.8
2080652.910.066.8
22303753.010.066.8
2360752.910.066.8
25102252.910.066.8
2610953.310.066.8
27602153.110.066.8
29101852.410.166.8
30601453.410.166.8
32102052.610.166.8
3360053.110.170.8

Worked out

ReadoutValueWorked out from
Run complete99.1%336 steps over 339
Step time2.31 s765.7 s over 331 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute25.960 s over the step time of 2.31 s
Epoch interval260.9 sMean of 2 gaps between epoch marks
Train loss since first reading−0.51090.0000 at step 336 minus 0.5109 at step 5
Lowest train loss-0.2792Lowest of 66 readings, at step 138
Eval accuracy, as shown88.5%0.8846 times 100, to one decimal. Changes on that plate are percentage points
Train minus eval loss−0.46750.0000 minus 0.4675, both at step 336
Wrong answers15130 eval items minus 115 correct. 115 over 130 is the only count that gives 0.8846
Rank4 of 22By last eval accuracy, across this project's 22 runs with an eval
Best so far0.9231Highest last eval accuracy, run 19
Against the run before+0.00000.8846 minus 0.8846 on run 16
Lowest last train loss-0.0937Lowest last train loss across this project's 22 runs, run 19
Seed spread, observed0.0077High 0.8846 minus low 0.8769 across 3 finished seeds
Train loss against run 16−0.23940.0000 minus 0.2394 on run 16, both at step 336
Samples per second since first reading−5.111.89 minus 6.99
Learning rate since first reading+1.00e-11.00e-1 minus 2.50e-5
This run's peak memory per GPU8.0 GBAs the run reports it: its own memory, not the whole GPU's