Dataset Viewer
Auto-converted to Parquet Duplicate
operations
int64
sentences
int64
lanes
dict
unsettled_payments
dict
questions
dict
means
dict
stuck
int64
gate
dict
1,500
549
{ "scored": 1255, "right": 1174, "uncertain": 5, "agreement": 0.9354581673306773, "confusions": { "trade->spam": 48, "trade->remittance": 4, "onboarding->spam": 3, "other->spam": 3, "trade->contract": 4, "other->onboarding": 7, "trade->onboarding": 5, "onboarding->remittance": ...
{ "remittance": 43, "trade": 5, "spam": 148, "uncertain": 8, "other": 8, "onboarding": 32 }
{ "Does this involve USDC?": { "auc": 0.9970390322835267, "bal": 0.9899281711263714, "pos": 515, "neg": 984 }, "Does this involve XLM?": { "auc": 0.9551326941032824, "bal": 0.9411728794081735, "pos": 680, "neg": 819 }, "Is this a smart contract call?": { "auc": 0.9929822938...
{ "auc": 0.9097266199358819, "bal": 0.7811576928479532, "count": 18 }
1
{ "answerable": 34, "refused": [ "Is this a farming reward?" ], "nonsense": 14, "turned_away": 14 }
1,500
549
{ "scored": 1255, "right": 1253, "uncertain": 0, "agreement": 0.998406374501992, "confusions": { "trade->spam": 2, "trade->remittance": null, "onboarding->spam": null, "other->spam": null, "trade->contract": null, "other->onboarding": null, "trade->onboarding": null, "onboardin...
{ "remittance": 43, "trade": 6, "spam": 182, "uncertain": 7, "other": 6, "onboarding": null }
{ "Does this involve USDC?": { "auc": 1, "bal": 1, "pos": 515, "neg": 985 }, "Does this involve XLM?": { "auc": 0.9978568866571018, "bal": 0.9963414634146341, "pos": 680, "neg": 820 }, "Is this a smart contract call?": { "auc": 1, "bal": 1, "pos": 361, "neg": 11...
{ "auc": 0.9884005805617551, "bal": 0.9330152957363946, "count": 18 }
0
{ "answerable": 34, "refused": [ "Is an account removing a trustline?", "Does an asset go round the order book and come back as itself?", "Is a claimable balance being created?", "Is this a farming reward?", "Is this a salary payment?" ], "nonsense": 14, "turned_away": 14 }

Stellar operation questions

The benchmark results behind goktugoguz/laya-multilingual-stellar.

  • eval-base.json, eval-blend-0.5.json: lanes, unsettled payments, viewer questions, stuck rows and gate on 1,500 fixture operations, for the base and the shipped model.
  • probes-*.json: the question gate under the radar's current probes (live) and probes rendered in the current template (rendered).
  • heldout-*.json: answers on 2,405 test operations whose sentences never appeared in training.
  • analysis.md: every variant measured (blend 0.5, blend 0.7, unblended), the gate residuals read probe by probe, and the held-out verdicts.

No sentences or memo text are published here. Memos are written by third parties, so the training rows stay private.

Downloads last month
67

Models trained or fine-tuned on goktugoguz/stellar-laya-bench