$8.585Observed · provisional
0.682Public leaderboard
185of 5,391 teams
0.765Same ranked snapshot
-0.083Balanced accuracy
18:00Daily · America/New_York
01 / Capital ledger
USD · FAILED ATTEMPTS INCLUDED ONCE| Campaign / item | Observed $ | As of | Billing state | Source |
|---|---|---|---|---|
| Qwen one-pass model | 4.856 | 14:05 EST | Completed; provisional invoice rows | Receipt ↗ |
| Qwen failed hosts | 0.899 | 14:05 EST | Two failures included, not added again | Receipt ↗ |
| Qwen three-pass run | 2.292 | 14:05 EST | Accumulating; unbilled remainder unknown | Receipt ↗ |
| GLM CUDA pilot | 0.538 | 14:25 EST | Four attempts; rentals destroyed; provisional | Receipt ↗ |
QWEN LIMIT$8.047 / $24
GLM LIMIT$0.538 / $10
Active Qwen charges after 14:05 EST are not included. Provider settlement remains open. Limits are separate; not available account balances.
02 / Scored result tape
PUBLIC LEADERBOARD / NOT VALIDATION| Entry | Score | Date | Source |
|---|---|---|---|
| Previous public ensemble | 0.651 | 27 Oct | Receipt ↗ |
| Public reproduction | 0.681 | 10 Nov | Receipt ↗ |
| Matched public-only control | 0.682 | 10 Nov | Receipt ↗ |
Matched additions / vs 0.682 public control
| Member | Score | Δ |
|---|---|---|
| +10% old Qwen | 0.678 | -0.004 |
| +10% tokenizer | 0.675 | -0.007 |
Scored progression / zero-to-one scale
+0.031
Token + byte entry pending at last receipt. Public 0.683 control and new full-pool blend passed visible offline T4 execution; neither has our hidden score.
03 / Experiment blotter
NAMED BASELINES · SOURCE RECEIPTS · MEASURED OUTCOMES| # | Experiment | Evidence type | Result | Baseline → score | Interpretation / limits | Source |
|---|---|---|---|---|---|---|
| 01 | Full-pool Qwen | validation | +3.37 pp | 0.587169 → 0.620882 | Standalone: five prompt folds, one seed. Trainer and exposure changed together. | Receipt ↗ |
| 02 | Mistral diversity lead | validation | +1.78 pp | 0.622322 → 0.640127 | Older full-stack reference, cross-fitted blend weight; author-supplied predictions, not fully nested. Own full training staged. | Receipt ↗ |
| 03 | OLMo diversity lead | validation | +0.89 pp | 0.622322 → 0.631228 | Same older reference; author-supplied predictions and cross-fitted weight. Interval crosses zero; own weights not reproduced. | Receipt ↗ |
| 04 | Liquid frozen encoder | validation | −0.74 pp | 0.543060 → 0.535650 | 5% blend on old public reference; exact checkpoint and all 2,699 rows. Full-backbone training is a separate experiment. | Receipt ↗ |
| 05 | Pruned GLM | engineering | CUDA works | — | 45 layers, 62 exact shard hashes. Three passages at 0.743–0.794 seconds; 36.80 GiB peak reserved. First eight experts, not calibrated. No classifier accuracy yet. | Receipt ↗ |
| 06 | Qwen three-pass | progress | 9,000 / 44,851 | — | Latest committed queue snapshot: step 9,000. Newer recovery file received 15:10 EST; its step is not asserted here. No completed validation score. | Receipt ↗ |
| 07 | Liquid full-backbone | progress | Training started | — | 15:24 EST queue receipt: one held-out pilot plus sequential five-fold / three-seed continuation. No completed fold evaluation in inspected output. CPU; no incremental rental. | Receipt ↗ |
No experiments match. Clear search or select ALL.
04 / Next bets
RESEARCH QUEUE / NOT NEW SPEND AUTHORIZATION- 01Score the matched 0.683 control and new full-pool Qwen blend after runtime and quota checks.
- 02Train our own Mistral on the corrected 309,365-row pool; retain OLMo as a complementary lead.
- 03Prove GLM on dual T4, then select experts using competition prose.
- 04Finish Liquid supervised folds before choosing whether to scale that architecture.
05 / Accounting & evidence notes
READ THE TIMESTAMPSSelected current rental campaigns and research results — not an all-time account reconciliation.
Separate historical API usage: $6.9033 + $0.1963 reserve. Receipt ↗
Basis, exclusions & comparison limits
Separate historical API project; not included in the rental subtotal. Earlier pilot, local electricity, subscriptions, free Kaggle quota and other older rentals are not reconciled here.
- Percentage-point validation gains use different named baselines. Never add them or compare them with leaderboard gains.
- Validation shares explored on shared prompts; public-model training exposure not independently established.
- No all-time spend total is claimed. Historical API usage is separate; other older work is not reconciled.
- Project budget caps are separate limits, not money spent or a fungible account balance.