sci-eval-final / README.md

Commit History

2 days ago

  1. Organize four final Sci-LLM stages; select 12x24 step35 agentic RL and preserve intermediates

    Verified signature
    f88b1bd

3 days ago

  1. Refresh aligned agentic index with Lightning, Cascade, Granite and GPT-OSS20

    Verified signature
    09a222f

4 days ago

  1. Refresh current temperature-0.7 agentic results and configuration audit

    Verified signature
    6d8a8c2

6 days ago

  1. Update index for latest Qwen IT agentic evaluation

    Verified signature
    dfa9edd
  2. Move agentic RL step 50 and 60 evaluations into backup

    Verified signature
    8c7071e

7 days ago

  1. Restore step-180 Olympiad result

    Verified signature
    b2c5d86
  2. Replace step-180 FS and HLE with the completed H100 rerun⁠

    Verified signature
    b835568
  3. Replace canonical arXiv Math results with complete avg@20 evaluations for 13 checkpoints⁠

    Verified signature
    95b4a36

11 days ago

  1. Archive Nemotron Bolt results under backup; exclude from report comparisons

    Verified signature
    ad8fda3
  2. Publish finished Granite, Bolt, Cascade, archived Muse and SFT SciCoding rerun⁠

    Verified signature
    b2e6160
  3. Archive report-excluded Flash, GPT-OSS-120B and Luna baselines under backup

    Verified signature
    d42dc63
  4. Publish aligned H100 agentic reruns; replace stale results and audit configuration exceptions

    Verified signature
    72aaf7a

16 days ago

  1. Combine HLE Gold and Revision into one benchmark per model⁠

    Verified signature
    39f4085
  2. Organize HLE results and provenance under each model⁠

    Verified signature
    eefd875

20 days ago

  1. Publish seven-model HLE-Verified scores and all 11200 saved trajectories⁠

    Verified signature
    ed4f0c5
  2. Publish GPT-OSS-20B and OLMo-3-32B-Think evaluations and full trajectories⁠

    Verified signature
    cf2a9b0
  3. Publish GPT-OSS-120B High non-agentic results and full trajectories⁠

    Verified signature
    e56ddda
  4. Complete Qwen35-35B agentic evaluation with full trajectories and two explicit zero adjudications⁠

    Verified signature
    95b8779

21 days ago

  1. Publish Qwen35-35B non-agentic and Nemotron full evaluation with trajectories⁠

    Verified signature
    f9bbb35
  2. Publish completed Gemma-4-12B evaluation with recovery and adjudication provenance⁠

    Verified signature
    de46502

22 days ago

  1. Publish Flash-0731 High evaluation trajectories, scores and completeness provenance⁠

    Verified signature
    f7d0134