This 1,069,302,417-parameter ranker continues Experiment 1.9 from its best checkpoint, step 328 / 104,960 scheduled samples. The parent's monitoring Spearman was 0.08534306347402562, not a new Experiment 1.9.1 test result. The parent architecture and pinned data/monitoring split are unchanged. The requested run uses twelve Hunter MI300A nodes, four GPU workers each, BF16 compute and FP32 master weights/Adam. The per-update learning rate is capped at 0.00006. Global batch increases from 320 to 1536, with microbatch 2 and balanced gradient accumulation. Adam and sampler state are restored; RNG streams are reseeded for 48 workers. This is not bitwise continuation. Best/latest inference…
Independent publisher
Y
YL95
LLM applications
Models in Library1
Datasets in Library0
Models on Hugging Face31
Followers—
Compare the published evidence
This profile connects the 1 model and dataset record shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.