AlphaGitHub
Auto-refresh · every 20 s

All training runs / Baguette / RTX 3070

Baguette Training monitor

320.96M parameters · PleIAs/SYNTH · Alpha native execution on RTX 3070

The retained run's first 22,100 updates ran on a rented RTX 3090. It resumed on a rented RTX 3070 from the durable checkpoint at update 22,100, and every later update runs there.

pausedRTX 3070 · 580.126.20
Snapshot 18:39:33 UTC
Recorded update
65,999
Invocation 32 · pass 0
Training loss
3.597
Pre-update loss · lower is better
Training throughput
Not recorded
Effective wall-clock rate
GPU memory
0.0 GiB
Device-wide usage · 7.7 GiB visible

Time to the Chinchilla token target

Live planning estimate · updates with the training observations

4.21% of token target270,331,904 / 6,419,139,840 tokens
Tokens remaining6,148,807,936
Last measured training rate 7,923 tokens/s
Estimated training time remaining8d 23h after training resumes
Recent rate spread 8d 22h–9d 2h
Projected target dateWaiting for training to resume
8d 23h training time at 7,923 tokens/s

263,819,264 distinct target positions · 6,512,640 repeated exposures. Future pauses and evaluation/storage gaps extend calendar time. The performance scenario is hypothetical.

Chinchilla scaling research · This token target is a planning milestone. Training continues under the project’s data coverage and quality goals.

Training loss

Cross-entropy per token · current invocation

Update 65,999Loss 3.596920-record mean 3.6350
2.673.333.994.665.3264,00064,50065,00065,49965,999
2,000 records · hover or use arrow keys to inspect

Checkpoint playground

Inspect recorded model responses across checkpoints and six prompt categories.

6 recorded tests

Conversation

Needs quality review

Prompt

User

Say hello and introduce yourself in one sentence.

Exact formatted input
<|im_start|>user
Say hello and introduce yourself in one sentence.<|im_end|>
<|im_start|>assistant
<think>

Model response

64 generated tokens · 05/10/2026, 14:34:29 UTC

### 1. Query Decomposition

"Would you explain why the term was the first set in one set in one set?"

→ Core question: Why did the term "in the first set" mean?
→ "What was the main number"
→ "What was the main number"

Evaluation notes

This response has not received a quality review. A successful inference run confirms execution, not answer quality.

Generation starts inside the released template’s thinking trace. The response may include reasoning before an answer. Repetition, mistakes and incomplete output are shown as generated.

Alpha-native greedy argmax; bounded 64-token generation; released tokenizer 0.20.3; baguette-prompts/1 diagnostic suite

Verified evidence and timing

Generation speed: No generation-only host or device timer was recorded for this prediction.

Checkpoint SHA-256
1e65aef5420761394ed1f1b11716e11e66b0c31d8df99f1350c90c14f58eae70
Prompt token record SHA-256
8e982823c044d7d1f739316425ca6c86a5f0da58c23a2e16331b8ea7d2b3c133
Prompt text SHA-256
65770472270c346c56dc15a45338c2d9f99bcee4e263705122fb14117095b529
Generated record SHA-256
94cd2f6a94b15cadc40a701b0da6a0a5b5c2d44dc75b3680ae669c9c41bc01b9
Decoded text SHA-256
175eb914344a833edcd4d6c938a771fd9616b262b50ce734e7d36e06c2c7fdbe

Training energy

Last verified benchmark · 100 steps per run · checkpoint 22,100 · 04/10/2026, 03:04:52 UTC

Candidate energy
0.02791 J/token
Combined energy ÷ combined target tokens
Baseline energy
0.03003 J/token
Same checkpoint, tokens and precision
Energy change
7.05% less
Two runs per implementation; observed range, not a confidence interval

Diagnostic replay, separate from production progress. Fresh-process measurements include warm-up and checkpoint load/save. Upload/readback is outside the measured window. CPU, host RAM and PSU losses are excluded. NVIDIA GeForce RTX 3090 · driver 580.126.20 · 350 W power limit.

Energy per training token J/token
baseline-1: 0.029863 J/token0.02986Baseline 1candidate-1: 0.026205 J/token0.02621Candidate 1candidate-2: 0.029620 J/token0.02962Candidate 2baseline-2: 0.030194 J/token0.03019Baseline 20
Total GPU energy kJ
baseline-1: 12.232 kJ12.23Baseline 1candidate-1: 10.734 kJ10.73Candidate 1candidate-2: 12.132 kJ12.13Candidate 2baseline-2: 12.367 kJ12.37Baseline 20
Mean GPU power, lower bound W
baseline-1: 238.78 W238.78Baseline 1candidate-1: 241.60 W241.6Candidate 1candidate-2: 243.88 W243.88Candidate 2baseline-2: 243.33 W243.33Baseline 20
Mean GPU power, upper bound W
baseline-1: 238.81 W238.81Baseline 1candidate-1: 241.68 W241.68Candidate 1candidate-2: 243.93 W243.93Candidate 2baseline-2: 243.38 W243.38Baseline 20
Counter-query boundary overhead ms
baseline-1: 5.8607 ms5.86Baseline 1candidate-1: 13.529 ms13.53Candidate 1candidate-2: 9.6324 ms9.63Candidate 2baseline-2: 9.5306 ms9.53Baseline 20
Fresh-process duration s
baseline-1: 51.219 s51.22Baseline 1candidate-1: 44.413 s44.41Candidate 1candidate-2: 49.736 s49.74Candidate 2baseline-2: 50.815 s50.82Baseline 20
Exact energy windows and evidence
RunTarget tokensJ/tokenMean power boundsSealed AOS source
baseline-1409,6000.02986271238.785–238.812 W6122d2a9b0674c57
candidate-1409,6000.02620545241.604–241.678 W6dd757b3eaf48eea
candidate-2409,6000.02961952243.885–243.932 Wbbed5fa2301cffc9
baseline-2409,6000.03019390243.334–243.379 Wd134621b556190d1

Counter-query timing bounds describe the sample interval, not sensor accuracy. Comparison c5cd5bf060632bb4 · archived evidence 83a7c81557e48727.

Where device time goes

Last recorded launch intervals · 01/10/2026, 21:49:26 UTC

Compute and setup
656.10 ms
Sum of last static intervals; includes setup
Checkpoint copies
7.181 s
Recorded once; separate from repeated work
Whole diagnostic process
91.38 s
100 updates, startup and checkpoint I/O

These are last occurrences of static launches, not averages across 100 updates. Startup and checkpoint work may run once. Diagnostic replay from checkpoint 22,100; separate from production progress. NVIDIA GeForce RTX 3090 · driver 580.126.20 · 350 W cap. 0.06159 GPU J/token for this profiled process.

Compute and setup, by component

Largest recorded groups first. Percentages use only the timed intervals below; an untimed launch is not zero.

  1. grouped attention key backward120.77 ms
    160 static launches18.4% of recorded time in this group
  2. grouped attention query backward87.89 ms
    80 static launches13.4% of recorded time in this group
  3. grouped attention forward65.32 ms
    80 static launches10.0% of recorded time in this group
  4. restore forward workspace31.99 ms
    79 static launches4.9% of recorded time in this group
  5. save forward workspace31.93 ms
    79 static launches4.9% of recorded time in this group
  6. feed down weight pullback12.26 ms
    80 static launches1.9% of recorded time in this group
  7. feed gate weight pullback11.91 ms
    80 static launches1.8% of recorded time in this group
  8. feed up weight pullback11.91 ms
    80 static launches1.8% of recorded time in this group
  9. parameter update11.31 ms
    1 static launches1.7% of recorded time in this group
  10. norm input10.87 ms
    160 static launches1.7% of recorded time in this group
  11. feed reduction residual10.80 ms
    80 static launches1.6% of recorded time in this group
  12. feed gate gate pullback10.54 ms
    80 static launches1.6% of recorded time in this group
  13. feed gate input pullback10.36 ms
    80 static launches1.6% of recorded time in this group
  14. feed up input pullback10.36 ms
    80 static launches1.6% of recorded time in this group
  15. input norm input pullback10.04 ms
    80 static launches1.5% of recorded time in this group
  16. feed norm input pullback9.56 ms
    80 static launches1.5% of recorded time in this group
  17. feed down input pullback9.54 ms
    80 static launches1.5% of recorded time in this group
  18. feed up forward9.43 ms
    80 static launches1.4% of recorded time in this group
  19. feed gate forward9.42 ms
    80 static launches1.4% of recorded time in this group
  20. tied vocabulary forward8.29 ms
    1 static launches1.3% of recorded time in this group
  21. projection weight pullback7.98 ms
    80 static launches1.2% of recorded time in this group
  22. feed gated product7.34 ms
    80 static launches1.1% of recorded time in this group
  23. attention output weight pullback7.34 ms
    80 static launches1.1% of recorded time in this group
  24. feed gate up pullback7.22 ms
    80 static launches1.1% of recorded time in this group
  25. projection input pullback6.79 ms
    80 static launches1.0% of recorded time in this group
  26. projection forward6.35 ms
    80 static launches1.0% of recorded time in this group
  27. tied vocabulary input pullback5.84 ms
    1 static launches0.9% of recorded time in this group
  28. norm feed5.63 ms
    80 static launches0.9% of recorded time in this group
  29. embedding gradient scatter5.13 ms
    1 static launches0.8% of recorded time in this group
  30. feed norm gain reduction4.90 ms
    80 static launches0.7% of recorded time in this group
  31. attention output residual4.82 ms
    80 static launches0.7% of recorded time in this group
  32. input norm gain reduction4.73 ms
    80 static launches0.7% of recorded time in this group
  33. attention output input pullback4.40 ms
    80 static launches0.7% of recorded time in this group
  34. tied vocabulary weight pullback4.34 ms
    1 static launches0.7% of recorded time in this group
  35. feed gate gradient to bfloat164.10 ms
    80 static launches0.6% of recorded time in this group
  36. feed gated to bfloat164.08 ms
    80 static launches0.6% of recorded time in this group
  37. feed up gradient to bfloat163.94 ms
    80 static launches0.6% of recorded time in this group
  38. query rows to columns3.73 ms
    80 static launches0.6% of recorded time in this group
  39. output gradient rows to columns3.73 ms
    80 static launches0.6% of recorded time in this group
  40. feed branch gradient sum3.08 ms
    80 static launches0.5% of recorded time in this group
  41. gradient apply scalar2.95 ms
    1 static launches0.4% of recorded time in this group
  42. block input gradient sum2.85 ms
    80 static launches0.4% of recorded time in this group
  43. attention residual gradient sum2.85 ms
    80 static launches0.4% of recorded time in this group
  44. exact vocabulary logit gradient2.84 ms
    1 static launches0.4% of recorded time in this group
  45. projection gradient to bfloat162.68 ms
    80 static launches0.4% of recorded time in this group
  46. query inverse rotate2.48 ms
    80 static launches0.4% of recorded time in this group
  47. exact vocabulary loss rows2.43 ms
    1 static launches0.4% of recorded time in this group
  48. scaled gradient square2.40 ms
    1 static launches0.4% of recorded time in this group
  49. attention gradient token to head2.36 ms
    80 static launches0.4% of recorded time in this group
  50. last input norm to bfloat162.05 ms
    80 static launches0.3% of recorded time in this group
  51. query rotate1.94 ms
    80 static launches0.3% of recorded time in this group
  52. residual gradient to bfloat161.90 ms
    80 static launches0.3% of recorded time in this group
  53. norm feed to bfloat161.89 ms
    80 static launches0.3% of recorded time in this group
  54. norm input to bfloat161.88 ms
    80 static launches0.3% of recorded time in this group
  55. logit gradient to bfloat161.85 ms
    1 static launches0.3% of recorded time in this group
  56. attention residual gradient to bfloat161.83 ms
    80 static launches0.3% of recorded time in this group
  57. value columns to rows1.78 ms
    80 static launches0.3% of recorded time in this group
  58. key rows to columns1.77 ms
    80 static launches0.3% of recorded time in this group
  59. output gradient to bfloat161.66 ms
    80 static launches0.3% of recorded time in this group
  60. value gradient group reduction1.63 ms
    80 static launches0.2% of recorded time in this group
  61. key gradient group reduction1.58 ms
    80 static launches0.2% of recorded time in this group
  62. value transpose1.45 ms
    80 static launches0.2% of recorded time in this group
  63. value inverse transpose1.33 ms
    80 static launches0.2% of recorded time in this group
  64. key inverse rotate1.14 ms
    80 static launches0.2% of recorded time in this group
  65. feed down weight to bfloat161.01 ms
    80 static launches0.2% of recorded time in this group
  66. feed gate weight to bfloat161.00 ms
    80 static launches0.2% of recorded time in this group
  67. feed up weight to bfloat160.99 ms
    80 static launches0.2% of recorded time in this group
  68. key rotate0.99 ms
    80 static launches0.2% of recorded time in this group
  69. projection weight to bfloat160.83 ms
    80 static launches0.1% of recorded time in this group
  70. attention output weight to bfloat160.71 ms
    80 static launches0.1% of recorded time in this group
  71. rotary table0.43 ms
    32 static launches0.1% of recorded time in this group
  72. tied embedding to bfloat160.27 ms
    1 static launches0.0% of recorded time in this group
  73. final norm input pullback0.12 ms
    1 static launches0.0% of recorded time in this group
  74. final norm0.07 ms
    1 static launches0.0% of recorded time in this group
  75. final norm gain reduction0.06 ms
    1 static launches0.0% of recorded time in this group
  76. final norm to bfloat160.02 ms
    1 static launches0.0% of recorded time in this group
  77. embed0.02 ms
    1 static launches0.0% of recorded time in this group
  78. gradient reduction zero0.02 ms
    1 static launches0.0% of recorded time in this group
  79. gradient partial level 10.02 ms
    1 static launches0.0% of recorded time in this group
  80. token window copy0.01 ms
    1 static launches0.0% of recorded time in this group
  81. loss mean readback0.01 ms
    1 static launches0.0% of recorded time in this group
  82. loss row partial sum0.01 ms
    1 static launches0.0% of recorded time in this group
  83. next token logit0.01 ms
    1 static launches0.0% of recorded time in this group
  84. gradient norm square0.01 ms
    1 static launches0.0% of recorded time in this group
  85. loss mean final sum0.01 ms
    1 static launches0.0% of recorded time in this group
  86. gradient partial level 20.01 ms
    1 static launches0.0% of recorded time in this group
  87. gradient bank padding0.01 ms
    1 static launches0.0% of recorded time in this group
  88. gradient clip scalar0.01 ms
    1 static launches0.0% of recorded time in this group

Checkpoint copies

Separate timing group: these transfers occurred once in this diagnostic, not once per training update.

  1. checkpoint second moment f322147.73 ms
    1,226 static launches29.9% of recorded time in this group
  2. checkpoint first moment f322057.50 ms
    1,226 static launches28.7% of recorded time in this group
  3. checkpoint master f321949.90 ms
    1,226 static launches27.2% of recorded time in this group
  4. checkpoint weights bf161025.56 ms
    614 static launches14.3% of recorded time in this group

Not yet measured: per-update device-time distributions, host work by phase, transfer overlap and idle gaps. This profile does not establish efficiency against measured hardware ceilings.

All 92 device regions and exact evidence
ComponentDevice regionStatic launchesLast intervals, ms
checkpoint second moment f32copy-words1,2262147.732512
checkpoint first moment f32copy-words1,2262057.499872
checkpoint master f32copy-words1,2261949.897888
checkpoint weights bf16copy-words6141025.555968
grouped attention key backwardgrouped-attention-key-backward160120.770080
grouped attention query backwardgrouped-attention-query-backward8087.891392
grouped attention forwardgrouped-attention-forward8065.319424
restore forward workspacecopy-words7931.990816
save forward workspacecopy-words7931.926016
feed down weight pullbackfeed-reduction-weight-pullback8012.263360
feed gate weight pullbackfeed-expansion-weight-pullback8011.914560
feed up weight pullbackfeed-expansion-weight-pullback8011.905728
parameter updateparameter-update111.308128
norm inputmean-square-forward16010.872960
feed reduction residualfeed-reduction-forward8010.798656
feed gate gate pullbackfeed-gate-gate-pullback8010.536320
feed gate input pullbackfeed-expansion-input-pullback8010.360736
feed up input pullbackfeed-expansion-input-pullback8010.358560
input norm input pullbackmean-square-input-backward8010.040672
feed norm input pullbackmean-square-input-backward809.561632
feed down input pullbackfeed-reduction-input-pullback809.535264
feed up forwardfeed-expansion-forward809.430720
feed gate forwardfeed-expansion-forward809.421056
tied vocabulary forwardvocabulary-forward18.291360
projection weight pullbackprojection-weight-pullback807.975296
feed gated productfeed-gate-forward807.342720
attention output weight pullbackattention-output-weight-pullback807.342080
feed gate up pullbackfeed-gate-up-pullback807.221152
projection input pullbackprojection-input-pullback806.792384
projection forwardprojection-forward806.348160
tied vocabulary input pullbackvocabulary-input-pullback15.836352
norm feedmean-square-forward805.633856
embedding gradient scatterembedding-gradient-scatter15.131872
feed norm gain reductionstrided-column-sum804.898208
attention output residualattention-output-forward804.820288
input norm gain reductionstrided-column-sum804.732800
attention output input pullbackattention-output-input-pullback804.396704
tied vocabulary weight pullbackvocabulary-weight-pullback14.336448
feed gate gradient to bfloat16float32-to-bfloat16804.095456
feed gated to bfloat16float32-to-bfloat16804.079488
feed up gradient to bfloat16float32-to-bfloat16803.937440
query rows to columnsattention-head-rows-to-columns803.733408
output gradient rows to columnsattention-head-rows-to-columns803.728192
feed branch gradient sumelementwise-sum803.079200
gradient apply scalargradient-apply-scalar12.946592
block input gradient sumelementwise-sum802.854112
attention residual gradient sumelementwise-sum802.849664
exact vocabulary logit gradientexact-vocabulary-logit-gradient12.840032
projection gradient to bfloat16float32-to-bfloat16802.677824
query inverse rotatequery-inverse-rotate802.476256
exact vocabulary loss rowsexact-vocabulary-loss-rows12.427712
scaled gradient squarescaled-gradient-square12.402624
attention gradient token to headattention-token-to-head802.362080
last input norm to bfloat16float32-to-bfloat16802.045440
query rotatequery-rotate801.943616
residual gradient to bfloat16float32-to-bfloat16801.903424
norm feed to bfloat16float32-to-bfloat16801.885824
norm input to bfloat16float32-to-bfloat16801.882816
logit gradient to bfloat16float32-to-bfloat1611.846816
attention residual gradient to bfloat16float32-to-bfloat16801.830304
value columns to rowsattention-head-columns-to-rows801.779776
key rows to columnsattention-head-rows-to-columns801.767808
output gradient to bfloat16float32-to-bfloat16801.664512
value gradient group reductiongrouped-attention-gradient-reduction801.632544
key gradient group reductiongrouped-attention-gradient-reduction801.578560
value transposevalue-transpose801.452032
value inverse transposevalue-inverse-transpose801.325472
key inverse rotatekey-inverse-rotate801.140224
feed down weight to bfloat16float32-to-bfloat16801.010208
feed gate weight to bfloat16float32-to-bfloat16800.995872
feed up weight to bfloat16float32-to-bfloat16800.992320
key rotatekey-rotate800.988704
projection weight to bfloat16float32-to-bfloat16800.829504
attention output weight to bfloat16float32-to-bfloat16800.708672
rotary tablerotary-table320.433504
tied embedding to bfloat16float32-to-bfloat1610.266400
final norm input pullbackmean-square-input-backward10.124480
final normmean-square-forward10.067968
final norm gain reductionstrided-column-sum10.062208
final norm to bfloat16float32-to-bfloat1610.023328
embedembedding-gather10.022624
gradient reduction zerofill-words10.019200
gradient partial level 1gradient-partial-sum10.016992
token window copycopy-words10.009952
loss mean readbackcopy-words10.007744
loss row partial sumloss-row-partial-sum10.007360
next token logittarget-logit-gather10.007328
gradient norm squaregradient-partial-sum10.007296
loss mean final sumloss-mean-final-sum10.007264
gradient partial level 2gradient-partial-sum10.007200
gradient bank paddingfill-words10.006816
gradient clip scalargradient-clip-scalar10.006784
Profiled ELF SHA-256
abe73980e91ab78c76094c2b50a8642a5430d24f6681f42aebf0e911b6376a6f
Sealed AOS stream SHA-256
5d8ec45e1ba31f7b1a5c85533b06cc334fb7062b26df29abfa1c1537c8909a95
Verified output checkpoint SHA-256
da3d238cfe3d8a4a748136d51d5456e2dd308cb0cad61c6e9ecb47f840319ea0
Archived evidence SHA-256
d3f31a85487cce46e9b8792ca20f13a0b47595aaa2fdcf87eddec602810af2b9

All training metrics

Every tracked measurement, with its recording window and evidence.

265 measurements · all charts shown

Data coverage

Confirmed update

65,999updates

10033,05065,99910065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

Invocation

32invocations

0163210065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

Corpus pass

0passes

010065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

Uploaded shards

500shards

50010065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

Verified shard bodies

500shards

50010065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

Eligible rows

Not recorded

Only the first source-row prefix was decoded; whole-corpus eligibility is not known.
0 recorded observations

Consumed rows

Not recorded

Source-row attribution for confirmed GPU updates is not yet published.
0 recorded observations

Rejected rows

1rows

11001,999

Latest unavailable: The input manifest links selected windows; it does not contain a reconciled full-prefix rejection count.. Showing retained history.

45 recorded observationsUpdate 1,999 · civil clock
2026-09-30T09:25:17.749Z

Eligible tokens

Not recorded

Only the first source-row prefix was tokenized; whole-corpus eligible tokens are not known.
0 recorded observations

Consumed tokens

264Mtokens

410K132M264M10065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

Repeated tokens

7Mtokens

03M7M10065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

Learning

Training mean loss

3.6nats/token

2.853.884.9264,00065,999
600 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Independent evaluation loss

7.2nats/token

7.28.068.914,00060,000

Latest unavailable: Not emitted by this Alpha qualification run.. Showing retained history.

7 recorded observationsUpdate 60,000 · civil clock
2026-10-05T14:32:46.805647+00:00

Training perplexity

36.49ratio

21.921,1102,1991,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Learning rate

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Gradient norm

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Nonfinite values at verified checkpoint

0values

010064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Loss statistics

Loss records

2,000records

01,0002,00010065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

Real SYNTH target tokens evaluated

8Mtokens

04M8M10065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

First training loss

3.77nats/token

3.316.69.891,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Minimum training loss

2.85nats/token

2.854.796.731,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Maximum training loss

5.14nats/token

4.096.999.891,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Median training loss

3.52nats/token

3.55.657.81,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

10th percentile training loss

3.3nats/token

3.285.227.161,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

90th percentile training loss

3.82nats/token

3.796.449.091,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Mean training loss

3.55nats/token

3.525.757.971,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Training loss standard deviation

0.239nats/token

0.1880.4540.721,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Loss change since first record

-0.171nats/token

-2.19-0.3891.411,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Lowest-loss pre-update index

64,944updates

1,08233,01364,9441,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Highest-loss pre-update index

65,788updates

10032,94465,7881,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Model

Training target top-1 accuracy

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Training target top-5 accuracy

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Held-out target top-1 accuracy

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Maximum target-row logit

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Logit RMS

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Attention entropy

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Maximum attention probability

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Residual stream RMS

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Attention output RMS

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Feed-forward output RMS

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Maximum absolute gradient

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Nonfinite gradients

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Gradient clip coefficient

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Master parameter L2 norm

477L2

470.2475.2480.210064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Parameter update L2 norm

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Update to weight norm ratio

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

First-moment L2 norm

0.226L2

0.2050.5150.82510064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Second-moment L2 norm

0.001L2

0.0010.0010.00210064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Nonfinite BF16 parameter values

0values

010064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Master parameter RMS

0.027F32 value

0.0260.0270.02710064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Largest absolute master parameter

1.47F32 value

11.241.4710064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

BF16 parameter L2 norm

477L2

470.2475.2480.210064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

BF16 parameter RMS

0.027F32 value

0.0260.0270.02710064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Largest absolute BF16 parameter

1.48F32 value

11.241.4810064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

First-moment RMS

1.26e-5F32 value

1.15e-52.87e-54.60e-510064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Largest absolute first moment

0.002F32 value

0.0020.0310.0610064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Second-moment RMS

2.92e-8F32 value

2.92e-86.90e-81.09e-710064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Largest absolute second moment

1.37e-5F32 value

1.23e-50010064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Last 100 pre-update losses: mean

3.53nats/token

3.515.367.21,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Previous 100 pre-update losses: mean

3.52nats/token

3.515.47.281,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Change between 100-loss windows

0.016nats/token

-0.102-0.0060.0891,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Last 100 pre-update losses: minimum

3.11nats/token

2.854.796.731,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Last 100 pre-update losses: maximum

4.61nats/token

45.887.761,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Last 100 pre-update losses: standard deviation

0.229nats/token

0.1830.3010.4191,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Largest adjacent pre-update loss rise

1.49nats/token

0.5711.662.751,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Largest adjacent pre-update loss fall

-1.58nats/token

-3.47-2.18-0.8931,15265,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:39:59.000Z

Tied embedding gradient L2 norm

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Attention projection gradient L2 norm

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Feed-forward gradient L2 norm

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Normalization gain gradient L2 norm

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Nonfinite activations

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Nonfinite logits

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Nonfinite optimizer updates

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Maximum causal attention score

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Minimum causal attention score

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Maximum attention log-normalizer

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Minimum attention log-normalizer

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Largest absolute residual activation

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Normalization input RMS

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Normalization output RMS

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Feed-forward gate saturation

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Mean target-token probability

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

Next-token prediction entropy

Not recorded

No sealed native AOS record for this diagnostic in the current qualification invocation.
0 recorded observations

GPU & speed

Effective training speed

4,718tokens/s

1,5495,1358,7221,18165,999

Latest unavailable: A second advancing publisher poll is needed; this run has no per-update device timer.. Showing retained history.

120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:40:15.062Z

Synthetic fixture speed

Not recorded

This run uses real SYNTH-prefix tokens, not the synthetic fixture.
0 recorded observations

Synthetic update interval

Not recorded

This run uses real SYNTH-prefix tokens, not the synthetic fixture.
0 recorded observations

Device execution speed

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Update wall time

868.1ms

469.61,5572,6441,18165,999

Latest unavailable: A second advancing publisher poll is needed; this run has no per-update device timer.. Showing retained history.

120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T15:40:15.062Z

Host wait

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Data fetch

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

GPU utilization

0%

049.59910065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:28Z

VRAM used

0.001GiB

0.0014.448.8810065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:28Z

VRAM peak

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

VRAM free

7.66GiB

7.6611.1714.6810065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:28Z

Power draw

16.07W

15.31130.624610065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:28Z

GPU temperature

30°C

30445810065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:28Z

S3 throughput

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Device timing

Launches

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Timed launches

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Untimed launches

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Forward time

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Backward time

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Optimizer time

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Checkpoint time

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Data time

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Reliability

Stalled time

Not recorded

Not emitted by this Alpha qualification run.
0 recorded observations

Failed attempts

0attempts

00.5110065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

AOS telemetry gaps

0gaps

010065,999
120 recorded observationsUpdate 65,999 · civil clock
2026-10-05T18:39:33.645Z

Host phases

request-host:begin

0ms

010065,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:request-checked

3,130ms

3,1303,1393,14710065,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:begin to request-host:request-checked

3,130ms

3,1303,1393,14710065,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:channel-ready

3,172ms

3,1723,4333,6951,15265,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:request-checked to request-host:channel-ready

41.71ms

41.71265.7489.71,15265,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:uploaded

3,253ms

3,2533,5153,7781,15265,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:channel-ready to request-host:uploaded

80.38ms

80.3881.5882.781,15265,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:learner-scalars

3,255ms

3,2553,5163,7781,15265,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:uploaded to request-host:learner-scalars

2.08ms

0.0761.082.081,15265,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:done

1Mms

1M2M3M1,99965,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

request-host:learner-scalars to request-host:done

1Mms

1M2M3M1,99965,999
120 recorded observationsUpdate 65,999 · host clock
2026-10-05T18:39:33.645Z

Checkpoint

master minimum

-0.185F32 value

-0.185-0.142-0.09910064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

master maximum

1.47F32 value

11.241.4710064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

master nonfinite count

0values

010064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

bf16 minimum

-0.185F32 value

-0.185-0.142-0.09910064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

bf16 maximum

1.48F32 value

11.241.4810064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

bf16 nonfinite count

0values

010064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

firstMoment minimum

-0.002F32 value

-0.06-0.031-0.00110064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

firstMoment maximum

0.002F32 value

0.0020.030.05910064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

firstMoment nonfinite count

0values

010064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

secondMoment minimum

5.87e-18F32 value

2.84e-188.38e-161.67e-1510064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

secondMoment maximum

1.37e-5F32 value

1.23e-50010064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

secondMoment nonfinite count

0values

010064,065
30 recorded observationsUpdate 64,065 · civil clock
2026-10-05T15:22:24.474Z

Inference

why-diagnostic generated tokens

64tokens

6412,100
One recorded observation · trend unavailableUpdate 12,100 · civil clock
2026-09-29T22:19:55.000Z

why-diagnostic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

6416,000
One recorded observation · trend unavailableUpdate 16,000 · civil clock
2026-09-30T00:37:42.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

6416,000
One recorded observation · trend unavailableUpdate 16,000 · civil clock
2026-09-30T00:37:42.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

6416,000
One recorded observation · trend unavailableUpdate 16,000 · civil clock
2026-09-30T00:37:42.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

6416,000
One recorded observation · trend unavailableUpdate 16,000 · civil clock
2026-09-30T00:37:42.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

6416,000
One recorded observation · trend unavailableUpdate 16,000 · civil clock
2026-09-30T00:37:42.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

6416,000
One recorded observation · trend unavailableUpdate 16,000 · civil clock
2026-09-30T00:37:42.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

6418,000
One recorded observation · trend unavailableUpdate 18,000 · civil clock
2026-09-30T01:48:28.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

6418,000
One recorded observation · trend unavailableUpdate 18,000 · civil clock
2026-09-30T01:48:28.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

6418,000
One recorded observation · trend unavailableUpdate 18,000 · civil clock
2026-09-30T01:48:28.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

6418,000
One recorded observation · trend unavailableUpdate 18,000 · civil clock
2026-09-30T01:48:28.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

6418,000
One recorded observation · trend unavailableUpdate 18,000 · civil clock
2026-09-30T01:48:28.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

6418,000
One recorded observation · trend unavailableUpdate 18,000 · civil clock
2026-09-30T01:48:28.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-09-30T02:54:51.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-09-30T02:54:51.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-09-30T02:54:51.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-09-30T02:54:51.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-09-30T02:54:51.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-09-30T02:54:51.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

64100
One recorded observation · trend unavailableUpdate 100 · civil clock
2026-09-30T08:17:47.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

64100
One recorded observation · trend unavailableUpdate 100 · civil clock
2026-09-30T08:17:47.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

64100
One recorded observation · trend unavailableUpdate 100 · civil clock
2026-09-30T08:17:47.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

64100
One recorded observation · trend unavailableUpdate 100 · civil clock
2026-09-30T08:17:47.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

64100
One recorded observation · trend unavailableUpdate 100 · civil clock
2026-09-30T08:17:47.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

64100
One recorded observation · trend unavailableUpdate 100 · civil clock
2026-09-30T08:17:47.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

642,000
One recorded observation · trend unavailableUpdate 2,000 · civil clock
2026-09-30T09:25:30.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

642,000
One recorded observation · trend unavailableUpdate 2,000 · civil clock
2026-09-30T09:25:30.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

642,000
One recorded observation · trend unavailableUpdate 2,000 · civil clock
2026-09-30T09:25:30.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

642,000
One recorded observation · trend unavailableUpdate 2,000 · civil clock
2026-09-30T09:25:30.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

642,000
One recorded observation · trend unavailableUpdate 2,000 · civil clock
2026-09-30T09:25:30.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

642,000
One recorded observation · trend unavailableUpdate 2,000 · civil clock
2026-09-30T09:25:30.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

6410,000
One recorded observation · trend unavailableUpdate 10,000 · civil clock
2026-09-30T14:23:54.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

6410,000
One recorded observation · trend unavailableUpdate 10,000 · civil clock
2026-09-30T14:23:54.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

6410,000
One recorded observation · trend unavailableUpdate 10,000 · civil clock
2026-09-30T14:23:54.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

6410,000
One recorded observation · trend unavailableUpdate 10,000 · civil clock
2026-09-30T14:23:54.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

6410,000
One recorded observation · trend unavailableUpdate 10,000 · civil clock
2026-09-30T14:23:54.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

6410,000
One recorded observation · trend unavailableUpdate 10,000 · civil clock
2026-09-30T14:23:54.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-10-01T01:28:20.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-10-01T01:28:20.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-10-01T01:28:20.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-10-01T01:28:20.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-10-01T01:28:20.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

6420,000
One recorded observation · trend unavailableUpdate 20,000 · civil clock
2026-10-01T01:28:20.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

6430,000
One recorded observation · trend unavailableUpdate 30,000 · civil clock
2026-10-04T14:05:44.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

6430,000
One recorded observation · trend unavailableUpdate 30,000 · civil clock
2026-10-04T14:05:44.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

6430,000
One recorded observation · trend unavailableUpdate 30,000 · civil clock
2026-10-04T14:05:44.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

6430,000
One recorded observation · trend unavailableUpdate 30,000 · civil clock
2026-10-04T14:05:44.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

6430,000
One recorded observation · trend unavailableUpdate 30,000 · civil clock
2026-10-04T14:05:44.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

6430,000
One recorded observation · trend unavailableUpdate 30,000 · civil clock
2026-10-04T14:05:44.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

6440,000
One recorded observation · trend unavailableUpdate 40,000 · civil clock
2026-10-05T10:26:54.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

6440,000
One recorded observation · trend unavailableUpdate 40,000 · civil clock
2026-10-05T10:26:54.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

6440,000
One recorded observation · trend unavailableUpdate 40,000 · civil clock
2026-10-05T10:26:54.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

6440,000
One recorded observation · trend unavailableUpdate 40,000 · civil clock
2026-10-05T10:26:54.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

6440,000
One recorded observation · trend unavailableUpdate 40,000 · civil clock
2026-10-05T10:26:54.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

6440,000
One recorded observation · trend unavailableUpdate 40,000 · civil clock
2026-10-05T10:26:54.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

6450,000
One recorded observation · trend unavailableUpdate 50,000 · civil clock
2026-10-05T12:31:47.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

6450,000
One recorded observation · trend unavailableUpdate 50,000 · civil clock
2026-10-05T12:31:47.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

6450,000
One recorded observation · trend unavailableUpdate 50,000 · civil clock
2026-10-05T12:31:47.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

6450,000
One recorded observation · trend unavailableUpdate 50,000 · civil clock
2026-10-05T12:31:47.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

6450,000
One recorded observation · trend unavailableUpdate 50,000 · civil clock
2026-10-05T12:31:47.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

9tokens

950,000
One recorded observation · trend unavailableUpdate 50,000 · civil clock
2026-10-05T12:31:47.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

01-chat generated tokens

64tokens

6460,000
One recorded observation · trend unavailableUpdate 60,000 · civil clock
2026-10-05T14:34:29.000Z

01-chat generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

02-follow-up generated tokens

64tokens

6460,000
One recorded observation · trend unavailableUpdate 60,000 · civil clock
2026-10-05T14:34:29.000Z

02-follow-up generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

03-factual generated tokens

64tokens

6460,000
One recorded observation · trend unavailableUpdate 60,000 · civil clock
2026-10-05T14:34:29.000Z

03-factual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

04-arithmetic generated tokens

64tokens

6460,000
One recorded observation · trend unavailableUpdate 60,000 · civil clock
2026-10-05T14:34:29.000Z

04-arithmetic generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

05-multilingual generated tokens

64tokens

6460,000
One recorded observation · trend unavailableUpdate 60,000 · civil clock
2026-10-05T14:34:29.000Z

05-multilingual generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations

06-reasoning generated tokens

64tokens

6460,000
One recorded observation · trend unavailableUpdate 60,000 · civil clock
2026-10-05T14:34:29.000Z

06-reasoning generation speed

Not recorded

No generation-only host or device timer was recorded for this prediction.
0 recorded observations
Run evidence & detailed records Identity, checkpoint banks, exact loss statistics and host phases

Run identity and coverage

Run
baguette-distinct-prefix-invocation-33-20260930
Source commit
cc4dcefdbc814799d2b87a3e0a9421de8203c8eb
ELF SHA-256
01245e64e74a8a4319e4ec5dd5ca039f06f16734bc0d3ed91410dfac23a8f19b
Compiler
Alpha bootstrap compiler
Pod and card
5e4ac4e6447d · RTX 3070 · driver 580.126.20
Dataset revision
0d6813a2966662c39f22f0b9af28a0c1c9f7a437
Tokenizer SHA-256
6bd28f1009d62677b4e550ab687efb89bec266535c783af4b4288bedf20f36c5
Updates
65,999 · invocation 32 · pass 0
Shards
500 verified of 500 uploaded
Cursor
synth_003.parquet · global window 64409/64410; input 298fd9d5412c
Last loss-file change (pod civil time)
2026-10-05T15:39:59.000Z

Data coverage

MeasurementValueClock and observationSource or absence
Eligible rows — Only the first source-row prefix was decoded; whole-corpus eligibility is not known.—Only the first source-row prefix was decoded; whole-corpus eligibility is not known.
Consumed rows — Source-row attribution for confirmed GPU updates is not yet published.—Source-row attribution for confirmed GPU updates is not yet published.
Rejected rows — The input manifest links selected windows; it does not contain a reconciled full-prefix rejection count.—The input manifest links selected windows; it does not contain a reconciled full-prefix rejection count.
Eligible tokens — Only the first source-row prefix was tokenized; whole-corpus eligible tokens are not known.—Only the first source-row prefix was tokenized; whole-corpus eligible tokens are not known.
Consumed tokens 263,819,264civil · 2026-10-05T18:39:33.645ZDistinct target-token positions from the previous verified prefix and confirmed updates in global windows 62410–64409; input 298fd9d5412c
Repeated tokens 6,512,640civil · 2026-10-05T18:39:33.645ZRepeat exposure accumulated before this distinct-window invocation; input 298fd9d5412c

Distinct SYNTH windows in this invocation

Shard
synth_003.parquet
Decoded source rows
7,168
Global window interval
62,410–64,409
Target token interval
[255,631,361, 263,823,361)
Input SHA-256
298fd9d5412cc4065d0f486b39c28e5ceb9b7dfbb65801dcb88ab86044aaf537

These invocation windows do not repeat each other. The full 500-shard corpus has not been consumed.

Learning

MeasurementValueClock and observationSource or absence
Training loss 3.5969civil · 2026-10-05T15:39:59.000Zalpha.learning.mean_loss@1; invocation update 1999; AOS a7288ff8be31; no event timestamp
Independent evaluation loss — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Training perplexity (derived) 36.4857civil · 2026-10-05T15:39:59.000Zexp(pre-update training cross-entropy in nats); AOS a7288ff8be31; distinct real SYNTH-prefix input, not independent evaluation
Learning rate — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Gradient norm — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Nonfinite values at verified checkpoint 0civil · 2026-10-05T15:22:24.474ZIndependent complete scan of four checkpoint banks at update 64000; manifest 1b28bfbe6755

Exact loss record summary

MeasurementValueClock and observationSource or absence
First pre-update loss 3.7679civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
Lowest pre-update loss 2.852civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
Highest pre-update loss 5.1372civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
Median pre-update loss 3.5234civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
10th percentile pre-update loss 3.2961civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
90th percentile pre-update loss 3.8197civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
Mean pre-update loss 3.5506civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
Population standard deviation 0.239civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
Change from first loss -0.1709civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
Lowest-loss update index 64,944civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens
Highest-loss update index 65,788civil · 2026-10-05T15:39:59.000ZDerived from 2000 exact alpha.learning.mean_loss@1 records; sealed AOS a7288ff8be31; distinct real SYNTH-prefix tokens

2,000 exact pre-update AOS records cover 8,192,000 real SYNTH-prefix target-token evaluations. A pre-update loss does not prove that update completed.

Model diagnostics

MeasurementValueClock and observationSource or absence
Training target top-1 accuracy — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Training target top-5 accuracy — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Held-out target top-1 accuracy — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Maximum target-row logit — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Logit RMS — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Attention entropy — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Maximum attention probability — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Residual stream RMS — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Attention output RMS — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Feed-forward output RMS — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Maximum absolute gradient — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Nonfinite gradients — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Gradient clip coefficient — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Master parameter L2 norm 477.0144 L2civil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; master bank
Parameter update L2 norm — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Update to weight norm ratio — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
First-moment L2 norm 0.2263 L2civil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; firstMoment bank
Second-moment L2 norm 0.0005 L2civil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; secondMoment bank
Nonfinite BF16 parameter values 0 valuescivil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; bf16 bank
Master parameter RMS 0.0266 F32 valuecivil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; master bank; L2 divided by sqrt(320956992 parameters)
Largest absolute master parameter 1.4729 F32 valuecivil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; master bank
BF16 parameter L2 norm 477.0125 L2civil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; bf16 bank
BF16 parameter RMS 0.0266 F32 valuecivil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; bf16 bank; L2 divided by sqrt(320956992 parameters)
Largest absolute BF16 parameter 1.4766 F32 valuecivil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; bf16 bank
First-moment RMS 0 F32 valuecivil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; firstMoment bank; L2 divided by sqrt(320956992 parameters)
Largest absolute first moment 0.0024 F32 valuecivil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; firstMoment bank
Second-moment RMS 0 F32 valuecivil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; secondMoment bank; L2 divided by sqrt(320956992 parameters)
Largest absolute second moment 0 F32 valuecivil · 2026-10-05T15:22:24.474ZIndependent complete checkpoint scan; update 64000; manifest 1b28bfbe6755; secondMoment bank
Last 100 pre-update losses: mean 3.534 nats/tokencivil · 2026-10-05T15:39:59.000ZDerived from exact last 100 pre-update loss records; sealed AOS a7288ff8be31; synthetic fixture
Previous 100 pre-update losses: mean 3.5176 nats/tokencivil · 2026-10-05T15:39:59.000ZDerived from exact previous 100 pre-update loss records; sealed AOS a7288ff8be31; synthetic fixture
Change between 100-loss windows 0.0164 nats/tokencivil · 2026-10-05T15:39:59.000ZDerived from exact last two nonoverlapping 100-record windows pre-update loss records; sealed AOS a7288ff8be31; synthetic fixture
Last 100 pre-update losses: minimum 3.112 nats/tokencivil · 2026-10-05T15:39:59.000ZDerived from exact last 100 pre-update loss records; sealed AOS a7288ff8be31; synthetic fixture
Last 100 pre-update losses: maximum 4.6133 nats/tokencivil · 2026-10-05T15:39:59.000ZDerived from exact last 100 pre-update loss records; sealed AOS a7288ff8be31; synthetic fixture
Last 100 pre-update losses: standard deviation 0.2294 nats/tokencivil · 2026-10-05T15:39:59.000ZDerived from exact last 100 pre-update loss records; sealed AOS a7288ff8be31; synthetic fixture
Largest adjacent pre-update loss rise 1.4892 nats/tokencivil · 2026-10-05T15:39:59.000ZDerived from exact adjacent pre-update loss records; sealed AOS a7288ff8be31; synthetic fixture
Largest adjacent pre-update loss fall -1.5776 nats/tokencivil · 2026-10-05T15:39:59.000ZDerived from exact adjacent pre-update loss records; sealed AOS a7288ff8be31; synthetic fixture
Tied embedding gradient L2 norm — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Attention projection gradient L2 norm — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Feed-forward gradient L2 norm — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Normalization gain gradient L2 norm — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Nonfinite activations — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Nonfinite logits — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Nonfinite optimizer updates — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Maximum causal attention score — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Minimum causal attention score — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Maximum attention log-normalizer — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Minimum attention log-normalizer — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Largest absolute residual activation — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Normalization input RMS — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Normalization output RMS — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Feed-forward gate saturation — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Mean target-token probability — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.
Next-token prediction entropy — No sealed native AOS record for this diagnostic in the current qualification invocation.—No sealed native AOS record for this diagnostic in the current qualification invocation.

Speed and resources

MeasurementValueClock and observationSource or absence
Effective training speed — A second advancing publisher poll is needed; this run has no per-update device timer.—A second advancing publisher poll is needed; this run has no per-update device timer.
Device execution speed — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Update wall time — A second advancing publisher poll is needed; this run has no per-update device timer.—A second advancing publisher poll is needed; this run has no per-update device timer.
Host wait — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Data fetch — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
GPU utilization 0%civil · 2026-10-05T18:39:28Znvidia-smi pod-wide sample on pod civil clock; 70.167.32.130
VRAM used (device-wide) 1,048,576 bytescivil · 2026-10-05T18:39:28Znvidia-smi pod-wide sample on pod civil clock; 70.167.32.130
VRAM peak — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
VRAM free (device-wide) 8,220,835,840 bytescivil · 2026-10-05T18:39:28Znvidia-smi pod-wide sample on pod civil clock; 70.167.32.130
Power 16.07 Wcivil · 2026-10-05T18:39:28Znvidia-smi pod-wide sample on pod civil clock; 70.167.32.130
Temperature 30 °Ccivil · 2026-10-05T18:39:28Znvidia-smi pod-wide sample on pod civil clock; 70.167.32.130
S3 throughput — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.

Time breakdown

MeasurementValueClock and observationSource or absence
Launches — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Timed launches — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Untimed launches — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Forward — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Backward — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Optimizer update — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Checkpoint — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.
Data — Not emitted by this Alpha qualification run.—Not emitted by this Alpha qualification run.

Alpha host phases

Elapsed from request start on the host's monotonic clock. Device execution time is separate.

PhaseSince startSince prior phaseMonotonic tick (ns)
request-host:begin0 ms—1551767194746564
request-host:request-checked3,130.469 ms3,130.469 ms1551770325215122
request-host:channel-ready3,172.18 ms41.712 ms1551770366926910
request-host:uploaded3,252.561 ms80.381 ms1551770447308017
request-host:learner-scalars3,254.641 ms2.079 ms1551770449387259
request-host:done1,052,617.65 ms1,049,363.009 ms1552819812396249

Checkpoint and reliability

Latest complete checkpoint: 08b6ef6056fe81c95fec3d62fd726d2cd195dbfb6d578c4146ec7911a0e82cfe. Reload observed by publisher: 2026-10-05T15:23:09.489Z. Loaded checkpoint at update 64,000: 08b6ef6056fe81c95fec3d62fd726d2cd195dbfb6d578c4146ec7911a0e82cfe. Failed attempts: 0. Telemetry gaps: 0. Stalled time: — Not emitted by this Alpha qualification run..

Checkpoint 64000 numerical banks

MeasurementValueClock and observationSource or absence
master minimum -0.1847civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
master maximum 1.4729civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
master nonfinite values 0civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
bf16 minimum -0.1846civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
bf16 maximum 1.4766civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
bf16 nonfinite values 0civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
firstMoment minimum -0.0024civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
firstMoment maximum 0.0021civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
firstMoment nonfinite values 0civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
secondMoment minimum 0civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
secondMoment maximum 0civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time
secondMoment nonfinite values 0civil · 2026-10-05T15:22:24.474ZIndependent full-bank checkpoint scan; manifest 1b28bfbe6755; verified update 64000; pod civil verification time

Discord milestones

  • 2026-09-30T22:25:47.461Z · dataset-500-verified · sent · checkpoint none
  • 2026-10-05T15:29:42.831Z · fresh-process-resume · sent · checkpoint 08b6ef6056fe81c95fec3d62fd726d2cd195dbfb6d578c4146ec7911a0e82cfe

Snapshot 2026-10-05T18:39:33.645Z · loss AOS stream a7288ff8be3193f55871d6e9b6b4d34c2ef9c503eb0cf84b45d6d068066db095 · loss seal feecf3b42bd3564f2811403e755757f24c7a5b26dbc48f577d48aaabe7d66b8a · host AOS stream 0582818f59620a7da661e543c7ee86766084846252007485cbb5bfb6b6242793 · host seal 27192209d4cda3255ddff90392d431af2314fc06a0136f3ce85d2b0a94499bcc · checkpoint manifest 1b28bfbe6755867863645f6c569b9b006f64f3c5b778d832abf4f8d473405dcb · inference receipts e527ef6349c3817727b3e162e86412e5e8cf934955f51a7222f7a6c8733c05b3 · failed prefix none · hotspot scan none.