Machine translations of the Dolci-Think-SFT-32B dataset, produced with gemma-4-31B-it. The samples selected for translation are those where contentquality == "excellent" according to the propella annotations. Each row is a translated conversation plus the result of a post-translation quality filter: - id — source record id. - messages — the translated conversation (list of {content, role}). - filterpass — true if the row passed every filter check (the recommended clean subset). - rejectreasons — empty when filterpass, otherwise any of: structure (broken /turn mismatch), truncated (a trace/answer hit the generation token cap or has unbalanced delimiters), wronglang (target-language character…
Publicly accessible
apache-2.0
Evaluation scores for the two OpenEuroLLM 32B cooldown (annealing) runs, every 2000 iterations from 100k to 120k, scored on the engbasemain and deubasemain suites of base-eval via Weights live in openeurollm/oellm-32b, one branch per checkpoint ( iter ). Both runs share globalbatchsize 4096 × seqlength 4096 = 16,777,216 tokens/step, so tokenstrained = 16777216 × step (120000 → 2.013T). The two runs are on an identical step grid, so they can be compared point by point. Same schema as openeurollm/prelude-base-eval-scores. - model, step, tokenstrained, modeldisplayname, modelpath - task, metric, score, scorestderr, numfewshot, language, nsamples - tasktype — benchmark (leaf task), taskgroup…
Publicly accessible
apache-2.0
