transformer-weights / scripts /paper_reference_loss.csv
angerami's picture
feat: add eval metrics pipeline (perplexity, paper reference, dashboard overlay)
17def50
Raw
History Blame Contribute Delete
339 Bytes
model,step,metric,value,source,notes
pythia-410m-deduped,10000,train_lm_loss,2.651,kalavai2024,From KALAVAI paper (arXiv 2405.xxxxx) Table 2; step 10K
pythia-1b-deduped,10000,train_lm_loss,2.474,kalavai2024,From KALAVAI paper Table 2; step 10K
pythia-6.9b-deduped,10000,train_lm_loss,2.320,kalavai2024,From KALAVAI paper Table 2; step 10K