CoolFace
Datasetpublic

laion/sft-repro-thinking-step630-nemotron-terminal-step1888-openthoughts-tblite-2026-08-13

Nemotron Terminal SFT reproduction evaluation artifacts This repository contains the complete Harbor artifact tree for the 300-trial OpenThoughts-TBLite evaluation of laion/sft-repro-thinking-step630-nemotron-terminal-step1888. The checkpoint was trained from the Grug stage-2 thinking checkpoint on the Nemotron Terminal corpus for 1,888 steps. Result Measure Value Attempted / completed 300 / 300 Verifier-scoreable 259 (86.33%) Aggregate reward, all… See the full description on the dataset page: https://huggingface.co/datasets/laion/sft-repro-thinking-step630-nemotron-terminal-step1888-openthoughts-tblite-2026-08-13.

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face
0likes54downloads
Dataset Card

Nemotron Terminal SFT reproduction evaluation artifacts

This repository contains the complete Harbor artifact tree for the 300-trial OpenThoughts-TBLite evaluation of `laion/sft-repro-thinking-step630-nemotron-terminal-step1888`. The checkpoint was trained from the Grug stage-2 thinking checkpoint on the Nemotron Terminal corpus for 1,888 steps.

Result

MeasureValue
Attempted / completed300 / 300
Verifier-scoreable259 (86.33%)
Aggregate reward, all attempts0.1449016619
Mean reward, scoreable trials0.1678397628
Full-reward trials39
Nonzero-reward trials49
ContextManagementInfrastructureError39

The archive preserves all trial results, trajectories, verifier outputs, FineStore tables, aggregate files, and the launch record. Signed Iris endpoint tokens and credential-shaped task fixture contents were replaced with explicit redaction markers before publication. redaction-report.json records the shape and file location of every replacement without retaining the secret text.

The exact training launcher, export drivers, model/eval configs, launch records, and analysis source are available in the reproduction source gist.

Provenance

  • —Run: 20260812-200420-sft-repro-thinking-step630-nemotron-terminal-step1888-openthoughts-tblite-terminus2-control-a40b
  • —Dataset: open-thoughts/OpenThoughts-TBLite@main
  • —Harness: Terminus-2
  • —Attempts per task: 3
  • —Temperature: 0.6
  • —Context: 24,576 input + 8,192 output tokens
  • —Marin: 505d4d799bdf9777aedc03945d0ae0f050460013
  • —Harbor: 24e5e67ac93d21c1a2202d6906f6093e0a57b86c
  • —Evalchemy: 5ef8b1604a535d04798f3424daa3fe4facd9bf32