xc_validation_strategy_vllm_zhouyuanxi

zhouyuanxi/xc_validation_strategy_vllm_zhouyuanxi

Fork 0

...

Author	SHA1	Message	Date
zhouyuanxi	da5b9f8612	restore all 10 accounts to ACCOUNTS; gate the active submitters behind ACTIVE_ACCOUNT_LABELS The six accounts dropped in v1.1.0 are back in ACCOUNTS so their tokens are not lost. Which ones actually submit is now controlled by ACTIVE_ACCOUNT_LABELS, currently the four requested for this round; set it to [] to rotate over all ten. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-20 17:20:28 +08:00
zhouyuanxi	e088e90dda	submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-20 17:15:54 +08:00

Author

SHA1

Message

Date

zhouyuanxi

da5b9f8612

restore all 10 accounts to ACCOUNTS; gate the active submitters behind ACTIVE_ACCOUNT_LABELS

The six accounts dropped in v1.1.0 are back in ACCOUNTS so their tokens are not
lost. Which ones actually submit is now controlled by ACTIVE_ACCOUNT_LABELS,
currently the four requested for this round; set it to [] to rotate over all ten.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

2026-09-20 17:20:28 +08:00

zhouyuanxi

e088e90dda

submit 1366 models from the relaxed filter, round-robin across 4 accounts

Filter condition 2 relaxed from "verified on another card" to "verified OR
verifying on another card": a model already under verification elsewhere has
passed the platform's basic checks, it just has no result yet. Same candidate
pool yields 1366 instead of 12 (114x) - the two original conditions are近乎
mutually exclusive, since a model verified on another card has long since been
tried on every card.

Account handling changed from sequential fallback to round-robin over
zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly
(verified: 342/342/341/341) instead of piling onto the first account until it
hits 60007.

Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

2026-09-20 17:15:54 +08:00

1 changed files with 1424 additions and 18 deletions

1442

main.py

View File

File diff suppressed because it is too large Load Diff

Compare commits

2 Commits

v1.0.36 ... main

1442

main.py

View File

Compare commits

2 Commits v1.0.36 ... main

1442 main.py View File

2 Commits

v1.0.36 ... main

1442

main.py

View File