text stringlengths 0 123 |
|---|
# E1: p-analysis eval — math (svdcos, p10-p90) + tool (cosine, p10-p90) |
# |
# Run merge first: bash rebuttal/run_merge.sh rebuttal/exp/e1_p_merge.txt |
# Then run eval: bash rebuttal/run_eval.sh rebuttal/exp/e1_p_eval.txt |
# |
# Output base: rebuttal/exp/E1_p_analysis/ |
# ── Math 1.5B (svdcos models) ───────────────────────────────────────────────── |
# math merged_models/rebuttal/svdcos/full/p10/math_simag_svdcos_p10_1.5B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p20/math_simag_svdcos_p20_1.5B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p30/math_simag_svdcos_p30_1.5B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p40/math_simag_svdcos_p40_1.5B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p50/math_simag_svdcos_p50_1.5B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p60/math_simag_svdcos_p60_1.5B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p70/math_simag_svdcos_p70_1.5B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p80/math_simag_svdcos_p80_1.5B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p90/math_simag_svdcos_p90_1.5B rebuttal/exp/E1_p_analysis |
# ── Math 4B (svdcos models) ─────────────────────────────────────────────────── |
# math merged_models/rebuttal/svdcos/full/p10/math_simag_svdcos_p10_4B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p20/math_simag_svdcos_p20_4B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p30/math_simag_svdcos_p30_4B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p40/math_simag_svdcos_p40_4B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p50/math_simag_svdcos_p50_4B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p60/math_simag_svdcos_p60_4B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p70/math_simag_svdcos_p70_4B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p80/math_simag_svdcos_p80_4B rebuttal/exp/E1_p_analysis |
# math merged_models/rebuttal/svdcos/full/p90/math_simag_svdcos_p90_4B rebuttal/exp/E1_p_analysis |
# ── Tool 1.5B (cosine, run merge first) ─────────────────────────────────────── |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p10_1.5B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p20_1.5B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p30_1.5B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p40_1.5B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p50_1.5B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p60_1.5B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p70_1.5B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p80_1.5B rebuttal/exp/E1_p_analysis |
tool merged_models/rebuttal/p_analysis/tool_merge_simag_p90_1.5B rebuttal/exp/E1_p_analysis |
# ── Tool 3B (cosine, run merge first) ───────────────────────────────────────── |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p10_3B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p20_3B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p30_3B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p40_3B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p50_3B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p60_3B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p70_3B rebuttal/exp/E1_p_analysis |
# tool merged_models/rebuttal/p_analysis/tool_merge_simag_p80_3B rebuttal/exp/E1_p_analysis |
tool merged_models/rebuttal/p_analysis/tool_merge_simag_p90_3B rebuttal/exp/E1_p_analysis |
# E2: sigma-analysis eval |
# |
# Non-redundant models only (same redundancy for math and tool): |
# sigma0.5 (all sizes) — lower threshold, more LN modules preserved |
# sigma2_4B / sigma2_3B — only non-trivially different from sigma1 for larger sizes |
# |
# Excluded (redundant): |
# math: sigma2_1.5B ≡ sigma3_1.5B ≡ baseline; sigma3_4B ≡ sigma2_4B |
# tool: same pattern applies |
# |
# Output base: rebuttal/exp/sigma_analysis/ |
# ── Math ────────────────────────────────────────────────────────────────────── |
# math merged_models/rebuttal/sigma_sweep/math_merge_simag_sigma0.5_1.5B rebuttal/exp/sigma_analysis |
# math merged_models/rebuttal/sigma_sweep/math_merge_simag_sigma0.5_4B rebuttal/exp/sigma_analysis |
# math merged_models/rebuttal/sigma_sweep/math_merge_simag_sigma2_4B rebuttal/exp/sigma_analysis |
# ── Tool (run merge first) ──────────────────────────────────────────────────── |
# tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma0.5_1.5B rebuttal/exp/sigma_analysis |
# tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma1_1.5B rebuttal/exp/sigma_analysis |
# tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma2_1.5B rebuttal/exp/sigma_analysis |
# tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma3_1.5B rebuttal/exp/sigma_analysis |
# tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma0.5_3B rebuttal/exp/sigma_analysis |
# tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma1_3B rebuttal/exp/sigma_analysis |
tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma2_3B rebuttal/exp/sigma_analysis |
tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma3_3B rebuttal/exp/sigma_analysis |
tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma0.5_1.5B rebuttal/exp/E2_sigma_analysis/tool |
# tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma1_1.5B rebuttal/exp/E2_sigma_analysis/tool |
tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma2_1.5B rebuttal/exp/E2_sigma_analysis/tool |
tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma3_1.5B rebuttal/exp/E2_sigma_analysis/tool |
tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma0.5_3B rebuttal/exp/E2_sigma_analysis/tool |
# tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma1_3B rebuttal/exp/E2_sigma_analysis/tool |
tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma2_3B rebuttal/exp/E2_sigma_analysis/tool |
tool merged_models/rebuttal/sigma_sweep/tool_merge_simag_sigma3_3B rebuttal/exp/E2_sigma_analysis/tool |
# E3: component ablation eval — bb_only and ln_only |
# |
# Math results already exist at: rebuttal/exp/E3_component_ablation/math/ |
# Tool models merged but not yet evaluated. |
# |
# Output base: rebuttal/exp/E3_component_ablation/ |
# ── Math (DONE — comment out if re-running is not needed) ───────────────────── |
# math merged_models/rebuttal/e1e3/bb_only/p70/math_merge_simag_bbonly_p70_1.5B rebuttal/exp/E3_component_ablation |
# math merged_models/rebuttal/e1e3/ln_only/p70/math_merge_simag_lnonly_p70_1.5B rebuttal/exp/E3_component_ablation |
# math merged_models/rebuttal/e1e3/bb_only/p80/math_merge_simag_bbonly_p80_4B rebuttal/exp/E3_component_ablation |
# math merged_models/rebuttal/e1e3/ln_only/p80/math_merge_simag_lnonly_p80_4B rebuttal/exp/E3_component_ablation |
# ── Tool ────────────────────────────────────────────────────────────────────── |
tool merged_models/rebuttal/e1e3/bb_only/p50/tool_merge_simag_bbonly_p50_1.5B rebuttal/exp/E3_component_ablation |
tool merged_models/rebuttal/e1e3/ln_only/p50/tool_merge_simag_lnonly_p50_1.5B rebuttal/exp/E3_component_ablation |
End of preview. Expand in Data Studio
No dataset card yet
- Downloads last month
- -