ikp
IKP: Incompressive Knowledge Probes
파일 탐색기
- effective-params.yml
- __init__.py
- __main__.py
- evaluate.py
- judge.py
- main.py
- presets.py
- probes.py
- progress.py
- query.py
- research.py
- all_models.json
- experiment.json
- models.json
- claude-3-haiku.json
- claude-3.5-haiku.json
- claude-haiku-4.5-think.json
- claude-haiku-4.5.json
- claude-opus-4-think.json
- claude-opus-4.5-think.json
- claude-opus-4.5.json
- claude-opus-4.6.json
- claude-opus-4.json
- claude-sonnet-4-think.json
- claude-sonnet-4.6.json
- claude-sonnet-4.json
- command-a.json
- command-r-plus.json
- command-r7b.json
- deepseek-r1-distill-llama-70b-think.json
- deepseek-r1-distill-qwen-32b-think.json
- deepseek-r1-think.json
- deepseek-v3.2-think.json
- deepseek-v3.2.json
- deepseek-v3.json
- ernie-4.5-300b-a47b.json
- evaluation_summary.json
- gemini-2.0-flash.json
- gemini-2.5-flash-lite-think.json
- gemini-2.5-flash-lite.json
- gemini-2.5-flash-think.json
- gemini-2.5-flash.json
- gemini-2.5-pro-think.json
- gemini-3-flash-think.json
- gemini-3-flash.json
- gemini-3.1-pro.json
- gemma-3-12b.json
- gemma-3-27b.json
- gemma-3-4b.json
- gemma-4-26b-a4b.json
- gemma-4-31b.json
- glm-4-32b.json
- glm-4.5-air-think.json
- glm-4.5-think.json
- glm-4.6-think.json
- glm-4.7-flash-think.json
- glm-4.7-think.json
- glm-5-think.json
- glm-5-turbo-think.json
- glm-5.1-think.json
- gpt-3.5-turbo.json
- gpt-4.1-mini.json
- gpt-4.1-nano.json
- gpt-4.1.json
- gpt-4.json
- gpt-4o-mini.json
- gpt-4o.json
- gpt-5-mini-think.json
- gpt-5-nano-think.json
- gpt-5-nano.json
- gpt-5-think.json
- gpt-5.4-mini.json
- gpt-5.4.json
- gpt-oss-120b-think.json
- gpt-oss-20b-think.json
- grok-3-mini-think.json
- grok-3.json
- grok-4.1-fast-think.json
- grok-4.20-think.json
- grok-4.20.json
- grok-4.json
- hermes-3-405b.json
- hunyuan-a13b-think.json
- hunyuan-a13b.json
- jamba-large.json
- kimi-k2.5-think.json
- kimi-k2.json
- llama-3-70b.json
- llama-3-8b.json
- llama-3.1-70b.json
- llama-3.1-8b.json
- llama-3.2-1b.json
- llama-3.2-3b.json
- llama-3.3-70b.json
- llama-4-maverick.json
- llama-4-scout.json
- mimo-v2-flash-think.json
- mimo-v2-flash.json
- mimo-v2-pro-think.json
- minimax-m1-think.json
- minimax-m2.7-think.json
- ministral-8b.json
- mistral-large.json
- mistral-medium-3.1.json
- mistral-small-24b.json
- nemotron-70b.json
- nemotron-super-49b-think.json
- nemotron-ultra-253b.json
- nova-micro.json
- olmo-3.1-32b.json
- phi-4.json
- qwen-2.5-72b.json
- qwen-2.5-7b.json
- qwen3-14b-think.json
- qwen3-235b-a22b-think.json
- qwen3-30b-a3b-think.json
- qwen3-32b-think.json
- qwen3-8b-think.json
- qwen3-next-80b-a3b.json
- qwen3.5-122b-a10b-think.json
- qwen3.5-27b-think.json
- qwen3.5-35b-a3b-think.json
- qwen3.5-397b-a17b-think.json
- qwen3.5-9b-think.json
- qwq-32b-think.json
- benchmark_scores.csv
- joined_per_model.csv
- raw_anthropic.md
- raw_deepseek_qwen_kimi_glm.md
- raw_google_meta.md
- raw_openai.md
- raw_others.md
- regression_summary.csv
- SUMMARY.md
- time_coefficients.csv
- all_calibration_results.json
- calibration_fit.json
- calibration_fit_before_v4.json
- analysis_8cell_websearch.md
- analysis_densing_law.md
- analysis_final_findings.md
- analysis_fingerprint.md
- ANALYSIS_REPORT.md
- analysis_researcher_significance.md
- analysis_scaling.md
- analysis_thinking_arch.md
- analysis_tiers.md
- analysis_transcripts.md
- analysis_wikidata_probes.md
- wikidata_quality_findings.md
- author_citation_sample.json
- author_tier_distribution.json
- calibration_sample.json
- chinese_corpus_grounded.json
- chinese_probes.json
- commoncrawl_sample.json
- corpus_grounded_all.json
- corpus_grounded_metadata.json
- corpus_grounded_probes.json
- dblp_author_profiles.json
- dblp_author_profiles_v2.json
- final_probe_set.json
- final_probe_set_v2.json
- final_probe_set_v3.json
- final_probe_set_v4.json
- final_probe_set_v5.json
- final_probe_set_v6.csv
- final_probe_set_v6.json
- final_probe_set_v7.json
- llm_probes_batch2.json
- llm_probes_batch2_calibrated.json
- llm_probes_batch3.json
- llm_probes_batch3_calibrated.json
- llm_probes_calibrated.json
- llm_probes_for_calibration.json
- llm_probes_hard.json
- llm_probes_manual.json
- llm_probes_medium.json
- llm_probes_very_hard.json
- new_researchers_fill.json
- proceedings_authors.json
- proceedings_authors_v2.json
- researcher_corpus_grounded.json
- researcher_dblp_grounded.json
- researcher_field_probes.json
- researcher_field_probes_v2.json
- researcher_field_probes_v3.json
- researcher_field_probes_v4.json
- researcher_probes_100_per_tier.json
- researcher_probes_balanced.json
- researcher_probes_fill.json
- researcher_probes_final.json
- researcher_probes_mid_tier.json
- researcher_probes_v5.json
- researcher_probes_v6.json
- sigcomm2024_author_citations.json
- T1_candidates.json
- T1_final.json
- T2_candidates.json
- T2_final.json
- T3_candidates.json
- T3_corpus_grounded.json
- T3_final.json
- T4_candidates.json
- T4_corpus_grounded.json
- T4_final.json
- T5_candidates.json
- T5_corpus_grounded.json
- T5_final.json
- T6_candidates.json
- T6_corpus_grounded.json
- T6_final.json
- T7_candidates.json
- T7_corpus_grounded.json
- T7_final.json
- web_grounded_methodology.json
- wiki_en_articles_v2.json
- wiki_en_grounded.json
- wiki_en_grounded_v2.json
- wiki_zh_articles_v2.json
- wiki_zh_grounded.json
- wiki_zh_grounded_v2.json
- wikidata_diverse_probes.json
- wikidata_diverse_v2.json
- wikidata_probes.json
- clean_mask.json
- final_probe_set_clean.json
- final_probe_set_v8.json
- final_probe_set_v9.json
- researcher_collisions.json
- researcher_collisions_v2.json
- researcher_gold_enriched.json
- researcher_probes.json
- split_manifest_v2.json
- t7_candidates.json
- t7_replacements.json
- wikidata_ambiguity.json
- adversarial_ikp.json
- analysis.json
- analysis_before_v4.json
- calibration_refit_v2.json
- claude-3-haiku.json
- claude-3.5-haiku.json
- claude-3.7-sonnet-think.json
- claude-3.7-sonnet.json
- claude-fable-5.json
- claude-haiku-4.5-think.json
- claude-haiku-4.5.json
- claude-opus-4-think.json
- claude-opus-4.1-think.json
- claude-opus-4.1.json
- claude-opus-4.5-think.json
- claude-opus-4.5.json
- claude-opus-4.6-think.json
- claude-opus-4.6.json
- claude-opus-4.7-think.json
- claude-opus-4.7.json
- claude-opus-4.8.json
- claude-opus-4.json
- claude-sonnet-4-think.json
- claude-sonnet-4.5-think.json
- claude-sonnet-4.5.json
- claude-sonnet-4.6-think.json
- claude-sonnet-4.6.json
- claude-sonnet-4.json
- claude-sonnet-5.json
- command-a.json
- command-r-plus.json
- command-r7b.json
- deepseek-r1-distill-llama-70b-think.json
- deepseek-r1-distill-qwen-32b-think.json
- deepseek-r1-think.json
- deepseek-v3.1-nex-n1.json
- deepseek-v3.1.json
- deepseek-v3.2-think.json
- deepseek-v3.2.json
- deepseek-v3.json
- deepseek-v4-flash-think.json
- deepseek-v4-flash.json
- deepseek-v4-flash_strict_rejudge.json
- deepseek-v4-pro-think.json
- deepseek-v4-pro.json
- effective_params.json
- ernie-4.5-300b-a47b.json
- evaluation_summary.json
- evaluation_summary_pre_v2.json
- final_assembly.json
- gemini-2.0-flash.json
- gemini-2.5-flash-lite-think.json
- gemini-2.5-flash-lite.json
- gemini-2.5-flash-think.json
- gemini-2.5-flash.json
- gemini-2.5-pro-think.json
- gemini-2.5-pro.json
- gemini-3-flash-think.json
- gemini-3-flash.json
- gemini-3.1-flash-lite.json
- gemini-3.1-pro.json
- gemma-2-27b.json
- gemma-2-2b.json
- gemma-3-12b.json
- gemma-3-1b.json
- gemma-3-270m.json
- gemma-3-27b.json
- gemma-3-4b.json
- gemma-3n-e4b.json
- gemma-4-26b-a4b.json
- gemma-4-31b.json
- glm-4-32b.json
- glm-4.5-air-think.json
- glm-4.5-think.json
- glm-4.6-think.json
- glm-4.7-flash-think.json
- glm-4.7-think.json
- glm-5-think.json
- glm-5-turbo-think.json
- glm-5-turbo.json
- glm-5.1-think.json
- glm-5.1.json
- glm-5.2.json
- glm-5.json
- gpt-3.5-turbo.json
- gpt-4-turbo.json
- gpt-4.1-mini.json
- gpt-4.1-nano.json
- gpt-4.1.json
- gpt-4.json
- gpt-4o-mini.json
- gpt-4o.json
- gpt-5-mini-think.json
- gpt-5-mini.json
- gpt-5-nano-think.json
- gpt-5-nano.json
- gpt-5-pro.json
- gpt-5-think.json
- gpt-5.1.json
- gpt-5.2-pro.json
- gpt-5.2.json
- gpt-5.3.json
- gpt-5.4-mini.json
- gpt-5.4-nano.json
- gpt-5.4-pro.json
- gpt-5.4.json
- gpt-5.5-pro.json
- gpt-5.5-think.json
- gpt-5.5.json
- gpt-5.json
- gpt-oss-120b-think.json
- gpt-oss-20b-think.json
- granite-3.3-2b.json
- grok-3-mini-think.json
- grok-3.json
- grok-4.20-think.json
- grok-4.20.json
- grok-4.json
- hermes-3-405b.json
- hermes-4-405b.json
- hunyuan-a13b-think.json
- hunyuan-a13b.json
- hy3-preview.json
- ikp_v2_validation.json
- intellect-3-think.json
- jamba-large.json
- kimi-k2.5-think.json
- kimi-k2.6-think.json
- kimi-k2.7-code.json
- kimi-k2.json
- ling-2.6-flash.json
- llama-3-70b.json
- llama-3-8b.json
- llama-3.1-70b.json
- llama-3.1-8b.json
- llama-3.2-1b.json
- llama-3.2-3b.json
- llama-3.3-70b.json
- llama-4-maverick.json
- llama-4-scout.json
- mimo-v2-flash-think.json
- mimo-v2-flash.json
- mimo-v2-pro-think.json
- mimo-v2-pro.json
- mimo-v2.5-pro-think.json
- mimo-v2.5-pro.json
- mimo-v2.5-think.json
- mimo-v2.5.json
- minimax-m1-think.json
- minimax-m2.7-think.json
- minimax-m3.json
- ministral-3b.json
- ministral-8b.json
- mistral-7b.json
- mistral-large.json
- mistral-medium-3.1.json
- mistral-nemo-12b.json
- mistral-small-24b.json
- mixtral-8x22b.json
- mixtral-8x7b.json
- nemotron-3-nano-30b.json
- nemotron-3-super-120b.json
- nemotron-3-ultra.json
- nemotron-70b.json
- nemotron-nano-9b-v2.json
- nemotron-super-49b-think.json
- nova-micro.json
- nova-premier.json
- nova-pro.json
- o1.json
- o3-mini.json
- o3.json
- o4-mini-think.json
- olmo-3.1-32b.json
- penalty_sensitivity.csv
- penalty_sensitivity_v2.csv
- phi-3-mini.json
- phi-4.json
- proprietary_sensitivity.csv
- qwen-2.5-0.5b.json
- qwen-2.5-72b.json
- qwen-2.5-7b.json
- qwen-max.json
- qwen-plus.json
- qwen-turbo.json
- qwen3-0.6b.json
- qwen3-14b-think.json
- qwen3-235b-a22b-think.json
- qwen3-30b-a3b-think.json
- qwen3-32b-think.json
- qwen3-8b-think.json
- qwen3-max.json
- qwen3-next-80b-a3b.json
- qwen3.5-0.8b.json
- qwen3.5-122b-a10b-think.json
- qwen3.5-27b-think.json
- qwen3.5-35b-a3b-think.json
- qwen3.5-397b-a17b-think.json
- qwen3.5-9b-think.json
- qwen3.5-flash-think.json
- qwen3.5-plus-think.json
- qwen3.6-plus-think.json
- qwen3.7-max.json
- qwen3.7-plus.json
- qwq-32b-think.json
- rescore_penalty_sweep.csv
- seed-1.6-flash-think.json
- seed-1.6-think.json
- seed-2.0-lite-think.json
- seed-2.0-mini-think.json
- smollm2-1.7b.json
- smollm2-135m.json
- smollm2-360m.json
- step-3.5-flash-think.json
- step-3.7-flash.json
- claude-3-haiku.json
- claude-3.5-haiku.json
- claude-3.7-sonnet-think.json
- claude-3.7-sonnet.json
- claude-haiku-4.5-think.json
- claude-haiku-4.5.json
- claude-opus-4-think.json
- claude-opus-4.1-think.json
- claude-opus-4.1.json
- claude-opus-4.5-think.json
- claude-opus-4.5.json
- claude-opus-4.6-think.json
- claude-opus-4.6.json
- claude-opus-4.7-think.json
- claude-opus-4.7.json
- claude-opus-4.json
- claude-sonnet-4-think.json
- claude-sonnet-4.5-think.json
- claude-sonnet-4.5.json
- claude-sonnet-4.6-think.json
- claude-sonnet-4.6.json
- claude-sonnet-4.json
- command-a.json
- command-r-plus.json
- command-r7b.json
- deepseek-r1-distill-llama-70b-think.json
- deepseek-r1-distill-qwen-32b-think.json
- deepseek-r1-think.json
- deepseek-v3.1-nex-n1.json
- deepseek-v3.1.json
- deepseek-v3.2-think.json
- deepseek-v3.2.json
- deepseek-v3.json
- deepseek-v4-flash-think.json
- deepseek-v4-flash.json
- deepseek-v4-pro-think.json
- deepseek-v4-pro.json
- ernie-4.5-300b-a47b.json
- gemini-2.0-flash.json
- gemini-2.5-flash-lite-think.json
- gemini-2.5-flash-lite.json
- gemini-2.5-flash-think.json
- gemini-2.5-flash.json
- gemini-2.5-pro-think.json
- gemini-2.5-pro.json
- gemini-3-flash-think.json
- gemini-3-flash.json
- gemini-3.1-flash-lite.json
- gemini-3.1-pro.json
- gemma-2-27b.json
- gemma-2-2b.json
- gemma-3-12b.json
- gemma-3-1b.json
- gemma-3-270m.json
- gemma-3-27b.json
- gemma-3-4b.json
- gemma-3n-e4b.json
- gemma-4-26b-a4b.json
- gemma-4-31b.json
- glm-4-32b.json
- glm-4.5-air-think.json
- glm-4.5-think.json
- glm-4.6-think.json
- glm-4.7-flash-think.json
- glm-4.7-think.json
- glm-5-think.json
- glm-5-turbo-think.json
- glm-5.1-think.json
- gpt-3.5-turbo.json
- gpt-4-turbo.json
- gpt-4.1-mini.json
- gpt-4.1-nano.json
- gpt-4.1.json
- gpt-4.json
- gpt-4o-mini.json
- gpt-4o.json
- gpt-5-mini-think.json
- gpt-5-mini.json
- gpt-5-nano-think.json
- gpt-5-nano.json
- gpt-5-pro.json
- gpt-5-think.json
- gpt-5.1.json
- gpt-5.2-pro.json
- gpt-5.2.json
- gpt-5.3.json
- gpt-5.4-mini.json
- gpt-5.4-nano.json
- gpt-5.4-pro.json
- gpt-5.4.json
- gpt-5.5-pro.json
- gpt-5.5-think.json
- gpt-5.5.json
- gpt-5.json
- gpt-oss-120b-think.json
- gpt-oss-20b-think.json
- granite-3.3-2b.json
- grok-3-mini-think.json
- grok-3.json
- grok-4.20-think.json
- grok-4.20.json
- grok-4.json
- hermes-3-405b.json
- hermes-4-405b.json
- hunyuan-a13b-think.json
- hunyuan-a13b.json
- hy3-preview.json
- intellect-3-think.json
- jamba-large.json
- kimi-k2.5-think.json
- kimi-k2.6-think.json
- kimi-k2.json
- ling-2.6-flash.json
- llama-3-70b.json
- llama-3-8b.json
- llama-3.1-70b.json
- llama-3.1-8b.json
- llama-3.2-1b.json
- llama-3.2-3b.json
- llama-3.3-70b.json
- llama-4-maverick.json
- llama-4-scout.json
- mimo-v2-flash-think.json
- mimo-v2-flash.json
- mimo-v2-pro-think.json
- mimo-v2-pro.json
- mimo-v2.5-pro-think.json
- mimo-v2.5-pro.json
- mimo-v2.5-think.json
- mimo-v2.5.json
- minimax-m1-think.json
- minimax-m2.7-think.json
- ministral-3b.json
- ministral-8b.json
- mistral-7b.json
- mistral-large.json
- mistral-medium-3.1.json
- mistral-nemo-12b.json
- mistral-small-24b.json
- mixtral-8x22b.json
- mixtral-8x7b.json
- nemotron-3-nano-30b.json
- nemotron-3-super-120b.json
- nemotron-70b.json
- nemotron-nano-9b-v2.json
- nemotron-super-49b-think.json
- nova-micro.json
- nova-premier.json
- nova-pro.json
- o1.json
- o3-mini.json
- o3.json
- o4-mini-think.json
- olmo-3.1-32b.json
- phi-3-mini.json
- phi-4.json
- qwen-2.5-0.5b.json
- qwen-2.5-72b.json
- qwen-2.5-7b.json
- qwen-max.json
- qwen-plus.json
- qwen-turbo.json
- qwen3-0.6b.json
- qwen3-14b-think.json
- qwen3-235b-a22b-think.json
- qwen3-30b-a3b-think.json
- qwen3-32b-think.json
- qwen3-8b-think.json
- qwen3-max.json
- qwen3-next-80b-a3b.json
- qwen3.5-122b-a10b-think.json
- qwen3.5-27b-think.json
- qwen3.5-35b-a3b-think.json
- qwen3.5-397b-a17b-think.json
- qwen3.5-9b-think.json
- qwen3.5-flash-think.json
- qwen3.5-plus-think.json
- qwen3.6-plus-think.json
- qwq-32b-think.json
- researcher_summary.json
- seed-1.6-flash-think.json
- seed-1.6-think.json
- seed-2.0-lite-think.json
- seed-2.0-mini-think.json
- smollm2-1.7b.json
- smollm2-135m.json
- smollm2-360m.json
- step-3.5-flash-think.json
- analysis_densing_law.md
- densing_analysis_data.csv
- ml_researcher_recognition.json
- ml_researchers_sample.json
- pipeline_store.json
- README.md
- researcher_citations.json
- researcher_recognition_rates.json
- thinking_mode_report.json
- wikidata_probe_recognition.json
- proposal-incompressible-proxy-estimation.md
- adversarial_ikp.pdf
- adversarial_ikp.png
- benchmark_comparison.pdf
- benchmark_comparison.png
- fig1_calibration.pdf
- fig1_calibration.png
- fig2_tier_heatmap.pdf
- fig2_tier_heatmap.png
- fig3_thinking_effect.pdf
- fig3_thinking_effect.png
- fig4_moe_params.pdf
- fig4_moe_params.png
- fig5_researcher_citations.pdf
- fig5_researcher_citations.png
- fig6_fingerprint_heatmap.pdf
- fig6_fingerprint_heatmap.png
- fig7_loo_validation.pdf
- fig7_loo_validation.png
- fig8_densing_law.pdf
- fig8_densing_law.png
- fig9_family_lineage.pdf
- fig9_family_lineage.png
- fig_a1_tier_boxplots.pdf
- fig_a1_tier_boxplots.png
- fig_a2_vendor_hallucination.pdf
- fig_a2_vendor_hallucination.png
- fig_a3_generation_trajectories.pdf
- fig_a3_generation_trajectories.png
- fig_a4_gpt5_family.pdf
- fig_a4_gpt5_family.png
- fig_lambda_sensitivity.pdf
- fig_lambda_sensitivity.png
- generate_appendix_figures.py
- generate_figures.py
- generate_lambda_figure.py
- ikp_v2_intervals.pdf
- ikp_v2_intervals.png
- frontier_estimates.tex
- full_accuracy.tex
- full_hallucination.tex
- lambda_floor_ablation.tex
- lambda_sensitivity.tex
- moe_dense_fits.tex
- moe_dense_sens.tex
- appendix.tex
- main.pdf
- main.tex
- references.bib
- research-plan.md
- __init__.py
- calibrate.py
- generate_llm_probes.py
- generate_mid_researchers.py
- generate_researchers.py
- generate_wikidata.py
- ingest.py
- judge.py
- landmarks.py
- run.py
- search_count.py
- store.py
- fig1_calibration_all.pdf
- fig1_calibration_all.png
- fig1_calibration_curve.pdf
- fig1_calibration_curve.png
- fig2_tier_sigmoids.pdf
- fig2_tier_sigmoids.png
- fig3_loocv.pdf
- fig3_loocv.png
- fig5_heatmap_all.pdf
- fig5_heatmap_all.png
- fig5_tier_heatmap.pdf
- fig5_tier_heatmap.png
- fig6_tier_ablation.pdf
- fig6_tier_ablation.png
- fig7_probe_count_ablation.pdf
- fig7_probe_count_ablation.png
- fig_densing_law.pdf
- fig_densing_law.png
- fig_densing_law.pdf
- fig_densing_law.png
- fp_all_families.tex
- fp_controls.tex
- fp_cross_vendor.tex
- lambda_floor_ablation.json
- lambda_floor_ablation.txt
- table1_calibration.tex
- ablation_results.json
- comprehensive_fingerprint_report.txt
- comprehensive_fingerprint_results.json
- 05_analyze_and_plot.py
- 06_live_analysis.py
- 07_ablations.py
- calibrate_batch2.py
- calibrate_batch3.py
- calibrate_landmarks.py
- calibrate_probes_parallel.py
- calibrate_tiers.py
- fill_researcher_probes.py
- fill_researchers_v2.py
- fix_calibration.py
- ikp_query.py
- retry_model.py
- run_batch2.py
- run_calibration.py
- run_new_models.py
- test_thinking_modes.py
- 01_generate_probes.py
- 01b_generate_t6_t7.py
- 02_run_calibration.py
- 03_fit_calibration.py
- 04_run_targets.py
- 08_chinese_probes.py
- 09_researcher_probes.py
- 10_web_grounded_probes.py
- 11_corpus_grounded_t5t7.py
- 12_fingerprint_probes.py
- 13_distillation_detection.py
- 14_comprehensive_fingerprinting.py
- 15_densing_law_analysis.py
- 16_benchmark_comparison.py
- 17_adversarial_robustness.py
- 18_v2_validation.py
- 19_effective_params.py
- add_release_dates.py
- analyze_results.py
- apply_manual_enrichment.py
- apply_t7_substitutions.py
- assemble_final_dataset.py
- build_clean_probe_set.py
- calibration_refit_v2.py
- clean_build_mask.py
- clean_flag_researchers.py
- clean_flag_wikidata.py
- enrich_researcher_gold.py
- final_assembly.py
- find_t7_remaining.py
- find_t7_replacements.py
- frontier_table.py
- full_results_tables.py
- ikp_budget.py
- ikp_estimate.py
- ikp_estimate_v2.py
- inspect_smoke_test.py
- lambda_floor_ablation.py
- lambda_sensitivity.py
- loo_cv_analysis.py
- make_probe_split.py
- moe_dense_analysis.py
- patch_429_failures.py
- penalty_sensitivity.py
- phase2_calibrate_t7.py
- proprietary_sensitivity.py
- README.md
- refine_collisions.py
- refit_with_evidence.py
- rerun_researcher_probes.py
- rescore_clean_lambda0.py
- rescore_with_penalty.py
- rewrite_researcher_probes.py
- run_all_models.py
- run_evaluation.py
- show_progress.py
- strict_rejudge_v4flash.py
- __init__.py
- api_client.py
- calibration.py
- evidence_judge.py
- llm_judge.py
- probe_generator.py
- probe_runner.py
- scorer.py
- test_effective_params.py
- claude-3-haiku__claude-3.5-haiku.json
- claude-3.5-haiku__claude-haiku-4.5.json
- claude-3.5-haiku__ernie-4.5-300b-a47b.json
- claude-3.5-haiku__hy3-preview.json
- claude-3.5-haiku__kimi-k2.json
- claude-3.5-haiku__mistral-large.json
- claude-3.7-sonnet__claude-sonnet-4.json
- claude-haiku-4.5-think__deepseek-v3.2.json
- claude-haiku-4.5__deepseek-v3.2.json
- claude-haiku-4.5__ernie-4.5-300b-a47b.json
- claude-haiku-4.5__gpt-4.1.json
- claude-haiku-4.5__llama-3-70b.json
- claude-haiku-4.5__llama-4-maverick.json
- claude-haiku-4.5__mistral-large.json
- claude-opus-4.1__claude-opus-4.5.json
- claude-opus-4.1__gpt-4.1-nano.json
- claude-opus-4.1__qwen3.5-35b-a3b-think.json
- claude-opus-4.5__claude-opus-4.6.json
- claude-opus-4.6__claude-opus-4.7.json
- claude-opus-4.6__deepseek-v3.1.json
- claude-opus-4.7-think__deepseek-v4-flash.json
- claude-opus-4.7-think__llama-3-70b.json
- claude-opus-4.7-think__mistral-large.json
- claude-opus-4.7-think__qwen3.7-max.json
- claude-opus-4.7__deepseek-v4-flash.json
- claude-opus-4.7__ernie-4.5-300b-a47b.json
- claude-opus-4.7__gemini-2.5-flash-think.json
- claude-opus-4.7__gemini-2.5-flash.json
- claude-opus-4.7__gemini-2.5-pro-think.json
- claude-opus-4.7__glm-5.json
- claude-opus-4.7__gpt-4.1.json
- claude-opus-4.7__gpt-4o.json
- claude-opus-4.7__gpt-5-think.json
- claude-opus-4.7__gpt-5.5-think.json
- claude-opus-4.7__gpt-5.json
- claude-opus-4.7__kimi-k2.6-think.json
- claude-opus-4.7__mixtral-8x22b.json
- claude-opus-4.7__o3.json
- claude-opus-4.7__qwen3-max.json
- claude-opus-4.7__qwen3.7-max.json
- claude-opus-4.8__deepseek-v4-flash.json
- claude-opus-4.8__gemini-2.0-flash.json
- claude-opus-4.8__gemini-3.1-flash-lite.json
- claude-opus-4.8__glm-4.5-air-think.json
- claude-opus-4.8__gpt-4o-mini.json
- claude-opus-4.8__grok-3-mini-think.json
- claude-opus-4.8__hy3-preview.json
- claude-opus-4.8__kimi-k2.json
- claude-opus-4.8__nova-pro.json
- claude-opus-4__claude-opus-4.1.json
- claude-sonnet-4-think__glm-5-think.json
- claude-sonnet-4-think__olmo-3.1-32b.json
- claude-sonnet-4.5-think__gemini-3.1-flash-lite.json
- claude-sonnet-4.5-think__gpt-4.json
- claude-sonnet-4.5-think__gpt-4o.json
- claude-sonnet-4.5-think__llama-3.1-70b.json
- claude-sonnet-4.5-think__qwen-max.json
- claude-sonnet-4.5__claude-sonnet-4.6.json
- claude-sonnet-4.5__gemini-3.1-flash-lite.json
- claude-sonnet-4.5__glm-5-think.json
- claude-sonnet-4.5__gpt-3.5-turbo.json
- claude-sonnet-4.5__gpt-4.1.json
- claude-sonnet-4.5__gpt-4.json
- claude-sonnet-4.5__gpt-4o.json
- claude-sonnet-4.5__kimi-k2.json
- claude-sonnet-4.5__llama-4-maverick.json
- claude-sonnet-4.5__mistral-small-24b.json
- claude-sonnet-4.6__deepseek-v3.1.json
- claude-sonnet-4__claude-sonnet-4.5.json
- command-a__command-r-plus.json
- command-a__deepseek-v4-flash.json
- command-a__gpt-4.json
- command-a__gpt-5-mini-think.json
- command-r-plus__command-r7b.json
- command-r-plus__glm-4.7-think.json
- command-r-plus__gpt-5.json
- command-r-plus__grok-4.json
- command-r-plus__o3.json
- command-r-plus__seed-1.6-think.json
- deepseek-r1-distill-llama-70b-think__hermes-4-405b.json
- deepseek-r1-distill-llama-70b-think__llama-3-70b.json
- deepseek-r1-distill-llama-70b-think__llama-3.1-70b.json
- deepseek-r1-distill-qwen-32b-think__hermes-4-405b.json
- deepseek-v3.1__deepseek-v3.2.json
- deepseek-v3.1__qwen-plus.json
- deepseek-v3__deepseek-v3.1.json
- deepseek-v4-flash__gemini-3-flash-think.json
- deepseek-v4-flash__llama-3.1-70b.json
- deepseek-v4-flash__mistral-large.json
- deepseek-v4-flash__qwen-max.json
- deepseek-v4-pro__gemini-2.0-flash.json
- deepseek-v4-pro__gemini-3.1-flash-lite.json
- deepseek-v4-pro__gpt-3.5-turbo.json
- deepseek-v4-pro__gpt-4.1.json
- deepseek-v4-pro__gpt-4o-mini.json
- deepseek-v4-pro__nemotron-3-ultra.json
- deepseek-v4-pro__o4-mini-think.json
- deepseek-v4-pro__qwen3-next-80b-a3b.json
- ernie-4.5-300b-a47b__glm-5.2.json
- ernie-4.5-300b-a47b__gpt-4o.json
- ernie-4.5-300b-a47b__grok-3.json
- ernie-4.5-300b-a47b__llama-3-70b.json
- ernie-4.5-300b-a47b__llama-4-maverick.json
- ernie-4.5-300b-a47b__mistral-large.json
- ernie-4.5-300b-a47b__mistral-small-24b.json
- ernie-4.5-300b-a47b__o4-mini-think.json
- ernie-4.5-300b-a47b__qwen-2.5-72b.json
- ernie-4.5-300b-a47b__qwen-max.json
- gemini-2.0-flash__gemini-2.5-flash.json
- gemini-2.0-flash__gpt-4o.json
- gemini-2.0-flash__llama-3-70b.json
- gemini-2.0-flash__llama-3.1-70b.json
- gemini-2.0-flash__qwen-max.json
- gemini-2.5-flash-lite__gemini-3.1-flash-lite.json
- gemini-2.5-flash__gemini-3-flash.json
- gemini-2.5-flash__glm-5.1.json
- gemini-2.5-flash__gpt-5-pro.json
- gemini-2.5-flash__gpt-5-think.json
- gemini-2.5-flash__gpt-5.5.json
- gemini-2.5-flash__gpt-5.json
- gemini-2.5-flash__o3.json
- gemini-2.5-pro-think__gpt-5.json
- gemini-2.5-pro-think__grok-4.json
- gemini-2.5-pro-think__kimi-k2.6-think.json
- gemini-2.5-pro__gemini-3.1-pro.json
- gemini-2.5-pro__gpt-5.json
- gemini-2.5-pro__grok-4.json
- gemini-3-flash-think__gpt-4o.json
- gemini-3-flash-think__grok-3-mini-think.json
- gemini-3-flash__gpt-4o.json
- gemini-3-flash__grok-3-mini-think.json
- gemini-3.1-flash-lite__gpt-4o.json
- gemini-3.1-flash-lite__llama-3.1-70b.json
- gemma-2-27b__gemma-3-12b.json
- gemma-2-27b__gemma-3-27b.json
- gemma-2-27b__gpt-5.json
- gemma-2-2b__gemma-3-1b.json
- gemma-3-12b__gemma-3-4b.json
- gemma-3-1b__gemma-3-4b.json
- gemma-3-270m__llama-3.3-70b.json
- gemma-3-27b__gemma-4-31b.json
- gemma-4-31b__hermes-4-405b.json
- glm-4.5-air-think__gpt-4o.json
- glm-4.5-air-think__gpt-5-mini-think.json
- glm-4.5-air-think__llama-3.1-70b.json
- glm-4.5-think__glm-4.6-think.json
- glm-4.6-think__glm-4.7-think.json
- glm-4.7-think__glm-5-think.json
- glm-5-think__glm-5.1-think.json
- glm-5-turbo-think__gpt-5.json
- glm-5-turbo-think__grok-4.json
- glm-5.1-think__gpt-5-think.json
- glm-5.1-think__gpt-5.json
- glm-5.1__gpt-5.json
- gpt-3.5-turbo__gpt-4.json
- gpt-3.5-turbo__hermes-4-405b.json
- gpt-3.5-turbo__llama-3.1-70b.json
- gpt-3.5-turbo__qwen-max.json
- gpt-4-turbo__gpt-4o.json
- gpt-4-turbo__llama-3.1-70b.json
- gpt-4-turbo__llama-3.3-70b.json
- gpt-4-turbo__qwen-max.json
- gpt-4.1-mini__gpt-4o-mini.json
- gpt-4.1-mini__gpt-5-mini.json
- gpt-4.1-nano__gpt-5-nano.json
- gpt-4.1-nano__hermes-4-405b.json
- gpt-4.1-nano__llama-3-70b.json
- gpt-4.1-nano__llama-3.1-70b.json
- gpt-4.1__gpt-4o.json
- gpt-4.1__gpt-5.json
- gpt-4.1__grok-3-mini-think.json
- gpt-4.1__llama-3.1-70b.json
- gpt-4.1__llama-3.3-70b.json
- gpt-4__gpt-4-turbo.json
- gpt-4__llama-3.1-70b.json
- gpt-4__llama-3.3-70b.json
- gpt-4__qwen-max.json
- gpt-4o-mini__hermes-4-405b.json
- gpt-4o-mini__llama-3.1-70b.json
- gpt-4o-mini__qwen-max.json
- gpt-4o__grok-3-mini-think.json
- gpt-4o__grok-3.json
- gpt-4o__kimi-k2.json
- gpt-4o__llama-3-70b.json
- gpt-4o__llama-3.3-70b.json
- gpt-4o__llama-4-maverick.json
- gpt-4o__mistral-small-24b.json
- gpt-4o__o1.json
- gpt-4o__qwen-2.5-72b.json
- gpt-4o__qwen3-max.json
- gpt-5-mini-think__llama-4-maverick.json
- gpt-5-mini-think__qwen3.5-122b-a10b-think.json
- gpt-5-mini-think__qwen3.5-27b-think.json
- gpt-5-mini-think__seed-1.6-think.json
- gpt-5-mini__gpt-5.4-mini.json
- gpt-5-nano__gpt-5.4-nano.json
- gpt-5-pro__gpt-5-think.json
- gpt-5-pro__grok-4.json
- gpt-5-pro__kimi-k2.6-think.json
- gpt-5-think__grok-4.json
- gpt-5-think__kimi-k2.6-think.json
- gpt-5-think__mistral-small-24b.json
- gpt-5.1__gpt-5.2.json
- gpt-5.1__mistral-medium-3.1.json
- gpt-5.2-pro__gpt-5.3.json
- gpt-5.2__gpt-5.2-pro.json
- gpt-5.2__gpt-5.3.json
- gpt-5.3__gpt-5.4.json
- gpt-5.3__grok-4.json
- gpt-5.4__gpt-5.4-pro.json
- gpt-5.5-pro__kimi-k2.6-think.json
- gpt-5.5-think__kimi-k2.6-think.json
- gpt-5.5__grok-4.json
- gpt-5__gpt-5-pro.json
- gpt-5__gpt-5.1.json
- gpt-5__grok-4.json
- gpt-5__kimi-k2.5-think.json
- gpt-5__kimi-k2.6-think.json
- grok-3-mini-think__mimo-v2-pro.json
- grok-3__grok-4.json
- grok-3__llama-3-70b.json
- grok-3__llama-3.1-70b.json
- grok-3__mistral-large.json
- grok-3__o1.json
- grok-3__qwen-max.json
- grok-3__qwen3-max.json
- grok-4__grok-4.20.json
- grok-4__kimi-k2.5-think.json
- grok-4__kimi-k2.6-think.json
- grok-4__o3.json
- hermes-3-405b__hermes-4-405b.json
- hermes-4-405b__nemotron-3-super-120b.json
- hermes-4-405b__qwen3-next-80b-a3b.json
- hermes-4-405b__qwen3.5-35b-a3b-think.json
- hy3-preview__mistral-large.json
- kimi-k2.5-think__kimi-k2.6-think.json
- kimi-k2.6-think__o3.json
- kimi-k2__kimi-k2.5-think.json
- kimi-k2__llama-3-70b.json
- kimi-k2__llama-3.1-70b.json
- kimi-k2__mistral-large.json
- ling-2.6-flash__qwen3.5-122b-a10b-think.json
- ling-2.6-flash__qwen3.5-plus-think.json
- ling-2.6-flash__seed-1.6-think.json
- llama-3-70b__llama-3.1-70b.json
- llama-3-70b__mistral-large.json
- llama-3-70b__mistral-small-24b.json
- llama-3-70b__qwen-max.json
- llama-3-8b__llama-3.1-8b.json
- llama-3.1-70b__llama-3.3-70b.json
- llama-3.1-70b__nemotron-3-ultra.json
- llama-3.1-70b__nova-pro.json
- llama-3.1-70b__o3-mini.json
- llama-3.1-70b__qwen3-235b-a22b-think.json
- llama-3.1-70b__qwen3-max.json
- llama-3.1-70b__qwen3.5-35b-a3b-think.json
- llama-3.1-8b__qwen3.5-27b-think.json
- llama-4-maverick__llama-4-scout.json
- llama-4-maverick__mistral-large.json
- llama-4-maverick__qwen-max.json
- ministral-3b__ministral-8b.json
- mistral-7b__mistral-nemo-12b.json
- mistral-large__mistral-medium-3.1.json
- mistral-large__qwen-max.json
- mistral-nemo-12b__mistral-small-24b.json
- mistral-small-24b__qwen-max.json
- mixtral-8x22b__mixtral-8x7b.json
- nova-micro__nova-pro.json
- nova-premier__nova-pro.json
- nova-pro__qwen-max.json
- o1__o3-mini.json
- o3__o3-mini.json
- o3__o4-mini-think.json
- phi-3-mini__phi-4.json
- qwen-2.5-72b__qwq-32b-think.json
- qwen-2.5-7b__qwen3-8b-think.json
- qwen-max__qwen3-max.json
- qwen-plus__qwen3.5-plus-think.json
- qwen3-235b-a22b-think__qwen3.5-397b-a17b-think.json
- qwen3.5-plus-think__qwen3.6-plus-think.json
- claude-3-haiku.json
- claude-3.5-haiku.json
- claude-3.7-sonnet-think.json
- claude-3.7-sonnet.json
- claude-fable-5.json
- claude-haiku-4.5-think.json
- claude-haiku-4.5.json
- claude-opus-4-think.json
- claude-opus-4.1-think.json
- claude-opus-4.1.json
- claude-opus-4.5-think.json
- claude-opus-4.5.json
- claude-opus-4.6-think.json
- claude-opus-4.6.json
- claude-opus-4.7-think.json
- claude-opus-4.7.json
- claude-opus-4.8.json
- claude-opus-4.json
- claude-sonnet-4-think.json
- claude-sonnet-4.5-think.json
- claude-sonnet-4.5.json
- claude-sonnet-4.6-think.json
- claude-sonnet-4.6.json
- claude-sonnet-4.json
- claude-sonnet-5.json
- command-a.json
- command-r-plus.json
- command-r7b.json
- deepseek-r1-distill-llama-70b-think.json
- deepseek-r1-distill-qwen-32b-think.json
- deepseek-r1-think.json
- deepseek-v3.1-nex-n1.json
- deepseek-v3.1.json
- deepseek-v3.2-think.json
- deepseek-v3.2.json
- deepseek-v3.json
- deepseek-v4-flash-think.json
- deepseek-v4-flash.json
- deepseek-v4-pro-think.json
- deepseek-v4-pro.json
- ernie-4.5-300b-a47b.json
- gemini-2.0-flash.json
- gemini-2.5-flash-lite-think.json
- gemini-2.5-flash-lite.json
- gemini-2.5-flash-think.json
- gemini-2.5-flash.json
- gemini-2.5-pro-think.json
- gemini-2.5-pro.json
- gemini-3-flash-think.json
- gemini-3-flash.json
- gemini-3.1-flash-lite.json
- gemini-3.1-pro.json
- gemma-2-27b.json
- gemma-2-2b.json
- gemma-3-12b.json
- gemma-3-1b.json
- gemma-3-270m.json
- gemma-3-27b.json
- gemma-3-4b.json
- gemma-3n-e4b.json
- gemma-4-26b-a4b.json
- gemma-4-31b.json
- glm-4-32b.json
- glm-4.5-air-think.json
- glm-4.5-think.json
- glm-4.6-think.json
- glm-4.7-flash-think.json
- glm-4.7-think.json
- glm-5-think.json
- glm-5-turbo-think.json
- glm-5-turbo.json
- glm-5.1-think.json
- glm-5.1.json
- glm-5.2.json
- glm-5.json
- gpt-3.5-turbo.json
- gpt-4-turbo.json
- gpt-4.1-mini.json
- gpt-4.1-nano.json
- gpt-4.1.json
- gpt-4.json
- gpt-4o-mini.json
- gpt-4o.json
- gpt-5-mini-think.json
- gpt-5-mini.json
- gpt-5-nano-think.json
- gpt-5-nano.json
- gpt-5-pro.json
- gpt-5-think.json
- gpt-5.1.json
- gpt-5.2-pro.json
- gpt-5.2.json
- gpt-5.3.json
- gpt-5.4-mini.json
- gpt-5.4-nano.json
- gpt-5.4-pro.json
- gpt-5.4.json
- gpt-5.5-pro.json
- gpt-5.5-think.json
- gpt-5.5.json
- gpt-5.json
- gpt-oss-120b-think.json
- gpt-oss-20b-think.json
- granite-3.3-2b.json
- grok-3-mini-think.json
- grok-3.json
- grok-4.20-think.json
- grok-4.20.json
- grok-4.json
- hermes-3-405b.json
- hermes-4-405b.json
- hunyuan-a13b-think.json
- hunyuan-a13b.json
- hy3-preview.json
- intellect-3-think.json
- jamba-large.json
- kimi-k2.5-think.json
- kimi-k2.6-think.json
- kimi-k2.7-code.json
- kimi-k2.json
- ling-2.6-flash.json
- llama-3-70b.json
- llama-3-8b.json
- llama-3.1-70b.json
- llama-3.1-8b.json
- llama-3.2-1b.json
- llama-3.2-3b.json
- llama-3.3-70b.json
- llama-4-maverick.json
- llama-4-scout.json
- mimo-v2-flash-think.json
- mimo-v2-flash.json
- mimo-v2-pro-think.json
- mimo-v2-pro.json
- mimo-v2.5-pro-think.json
- mimo-v2.5-pro.json
- mimo-v2.5-think.json
- mimo-v2.5.json
- minimax-m1-think.json
- minimax-m2.7-think.json
- minimax-m3.json
- ministral-3b.json
- ministral-8b.json
- mistral-7b.json
- mistral-large.json
- mistral-medium-3.1.json
- mistral-nemo-12b.json
- mistral-small-24b.json
- mixtral-8x22b.json
- mixtral-8x7b.json
- nemotron-3-nano-30b.json
- nemotron-3-super-120b.json
- nemotron-3-ultra.json
- nemotron-70b.json
- nemotron-nano-9b-v2.json
- nemotron-super-49b-think.json
- nova-micro.json
- nova-premier.json
- nova-pro.json
- o1.json
- o3-mini.json
- o3.json
- o4-mini-think.json
- olmo-3.1-32b.json
- phi-3-mini.json
- phi-4.json
- qwen-2.5-0.5b.json
- qwen-2.5-72b.json
- qwen-2.5-7b.json
- qwen-max.json
- qwen-plus.json
- qwen-turbo.json
- qwen3-0.6b.json
- qwen3-14b-think.json
- qwen3-235b-a22b-think.json
- qwen3-30b-a3b-think.json
- qwen3-32b-think.json
- qwen3-8b-think.json
- qwen3-max.json
- qwen3-next-80b-a3b.json
- qwen3.5-0.8b.json
- qwen3.5-122b-a10b-think.json
- qwen3.5-27b-think.json
- qwen3.5-35b-a3b-think.json
- qwen3.5-397b-a17b-think.json
- qwen3.5-9b-think.json
- qwen3.5-flash-think.json
- qwen3.5-plus-think.json
- qwen3.6-plus-think.json
- qwen3.7-max.json
- qwen3.7-plus.json
- qwq-32b-think.json
- seed-1.6-flash-think.json
- seed-1.6-think.json
- seed-2.0-lite-think.json
- seed-2.0-mini-think.json
- smollm2-1.7b.json
- smollm2-135m.json
- smollm2-360m.json
- step-3.5-flash-think.json
- step-3.7-flash.json
- T1.json
- T2.json
- T3.json
- T4.json
- T5.json
- T6.json
- T7.json
- benchmarks.json
- calibration.json
- densing.json
- fingerprint.json
- generations.json
- hallucination.json
- index.json
- models.json
- pipeline.json
- probes.json
- recognition.json
- sensitivity.json
- thinking_pairs.json
- prepare_data.py
- Loading.tsx
- Tag.tsx
- TierBar.tsx
- Calibration.tsx
- Densing.tsx
- Fingerprint.tsx
- FingerprintPair.tsx
- Generations.tsx
- Hallucination.tsx
- Home.tsx
- ModelDetail.tsx
- Models.tsx
- Pipeline.tsx
- ProbeDetail.tsx
- Probes.tsx
- Recognition.tsx
- Thinking.tsx
- TierHeatmap.tsx
- App.tsx
- data.ts
- index.css
- main.tsx
- types.ts
- util.ts
- .gitignore
- index.html
- package-lock.json
- package.json
- postcss.config.js
- README.md
- tailwind.config.js
- tsconfig.json
- vite.config.ts
- .gitignore
- ADVERSARIAL_IKP.md
- IKP_V2.md
- Makefile
- PARAM_POLICY.md
- README.md
- REPRODUCTION.md
- requirements.txt
- TOOLKIT.md
# CDN으로 사용하기
jsDelivrjsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.
링크
예시
명령어 용어집
이 문서에서 사용된 명령어를 모아봤습니다. 낯선 명령어가 있다면 펼쳐서 확인해보세요.
pip install
설명 보기 ▼
pip install
Install Python packages.
pip install {{package1 package2 ...}}
Install one or more packages:
pip install {{package1 package2 ...}} {{[-U|--upgrade]}}
Upgrade all specified packages to the latest version, installing any that are not already present:
pip install {{package}}=={{version}}
Install a specific version of a package:
python
설명 보기 ▼
python
Python 언어 인터프리터.
더 많은 정보: <https://docs.python.org/using/cmdline.html>.
python
REPL(대화형 셸) 시작:
python {{path/to/file.py}}
특정 Python 파일 실행:
python -i {{path/to/file.py}}
특정 Python 파일 실행 후 REPL 시작:
// repository documentation
Was this content helpful?
(0 ratings)
