Skip to page content
SAVRN
Search Plan a project

Dataset · Question answering

TPBench

by Papersubmission 4papersubmission/TPBench

TPBench measures three retrieval targets after dialogue compression: This repository includes data, construction inputs, compression and reader code, scorers, saved answers, and paper results.

Rows51,732
Configurations3
Size572.5 MB
Licenseother
AccessPublicly accessible
Monthly Downloads640

Dataset Card

TPBench measures three retrieval targets after dialogue compression: This repository includes data, construction inputs, compression and reader code, scorers, saved answers, and paper results. RESULTSMAP.md connects the paper comparisons to their evidence files. Use Python 3.10 or newer. No model access is needed for these checks. The checks validate file hashes and metadata, regenerate all 42 main probe cells, rescore saved answers, and test FMTS context construction. Outputs from tests are written to a temporary directory, not over the released results. Each row fixes the example identifier, method, budget, reference answer, and reader prompts. Identifiers remain stable across…

Excerpt from the card by Papersubmission, licensed other.

Structure

current_value 22,932 rows

SplitRowsSize
probes22,93248.6 MB
dialogue_idstringdatasetstringslotstringslot_humanstringold_valuestringnew_valuestringmethodstringcompressed_text_charsint64metavalueprompt_systemstringprobe_typestringgoldstringprompt_userstringcompressed_text_chars_usedint64compressed_text_usedstringtemplate_sha16string

initial_goal 19,200 rows

SplitRowsSize
probes19,20033.0 MB
dialogue_idstringdatasetstringslotstringslot_humanstringold_valuestringnew_valuestringmethodstringcompressed_text_charsint64metavalueprompt_systemstringprobe_typestringgoldstringprompt_userstringcompressed_text_chars_usedint64compressed_text_usedstringtemplate_sha16string

joint_recovery 9,600 rows

SplitRowsSize
probes9,60021.0 MB
dialogue_idstringdatasetstringslotstringslot_humanstringold_valuestringnew_valuestringstory_label_phrasestringmethodstringmetavalueprompt_systemstringprobe_typestringgoldstringgold_story_labelstringgold_valuestringprompt_userstringcompressed_text_charsint64compressed_text_usedstringtemplate_sha16string

Details

Repository
4papersubmission/TPBench
Publisher
Papersubmission
Task category
Question answering
Tags
dialogue, long-context, evaluation
Size category
Not stated by the source
Languages
en, zh
Revision
4cf58bb41539094bd394c9abd84dea95c0396340
Last updated
2026-09-28

Files

476 files, 572.5 MB in total.

Data413 files · 572.1 MB
Documentation8 files · 30.7 KB
Other54 files · 316.5 KB
Repository1 file · 3.2 KB
Every file
FileTypeSizeSHA-256
LICENSES/AnswerEquivalence-Apache-2.0.txtData11.4 KB—
LICENSES/Apache-2.0.txtData11.4 KB—
LICENSES/CC-BY-4.0.txtData547 B—
LICENSES/LongMemEval.txtData1.0 KB—
LICENSES/MultiWOZ_2.2.txtData582 B—
LICENSES/RiSAWOZ.txtData1.3 KB—
LICENSES/SGD.txtData765 B—
batches/fmts_refpool/gold_tqa_indirect_seed42_refpool_r010.jsonlData508.4 KB—
batches/fmts_refpool/gold_tqa_indirect_seed42_refpool_r030.jsonlData508.4 KB—
batches/fmts_refpool/gold_tqa_multiwoz_scope_refpool_r010.jsonlData697.3 KB—
batches/fmts_refpool/gold_tqa_multiwoz_scope_refpool_r030.jsonlData697.3 KB—
batches/fmts_refpool/gold_tqa_sgd_scope_refpool_r010.jsonlData721.7 KB—
batches/fmts_refpool/gold_tqa_sgd_scope_refpool_r030.jsonlData721.7 KB—
batches/fmts_refpool/responses_tqa_indirect_seed42_refpool_r010.jsonlData56.6 KB—
batches/fmts_refpool/responses_tqa_indirect_seed42_refpool_r030.jsonlData58.3 KB—
batches/fmts_refpool/responses_tqa_multiwoz_scope_refpool_r010.jsonlData100.6 KB—
batches/fmts_refpool/responses_tqa_multiwoz_scope_refpool_r030.jsonlData101.4 KB—
batches/fmts_refpool/responses_tqa_sgd_scope_refpool_r010.jsonlData90.8 KB—
batches/fmts_refpool/responses_tqa_sgd_scope_refpool_r030.jsonlData92.4 KB—
croissant.jsonData45.8 KB—
data/construction/multiwoz_compressed_r10.jsonlData2.0 MB—
data/construction/multiwoz_compressed_r30.jsonlData3.3 MB—
data/construction/multiwoz_compressed_r50.jsonlData3.3 MB—
data/construction/multiwoz_compressed_r70.jsonlData4.2 MB—
data/construction/sgd_compressed_r10.jsonlData2.8 MB—
data/construction/sgd_compressed_r30.jsonlData4.5 MB—
data/construction/sgd_compressed_r50.jsonlData4.4 MB—
data/construction/sgd_compressed_r70.jsonlData5.7 MB—
data/fmts/fmts_indirect_seed42.jsonlData327.0 KB—
data/fmts/fmts_multiwoz_scope.jsonlData489.6 KB—
data/fmts/fmts_sgd_scope.jsonlData539.4 KB—
data/probes/probes_lme_ku_all72_r30_s42-46.jsonlData50.5 MB4cff2db83727
data/probes/probes_lme_ku_s42_r10_n72.jsonlData20.9 MBa2e17b0b2c31
data/probes/probes_lme_ku_s42_r30_n72.jsonlData32.6 MB61ac40f7c431
data/probes/probes_multiwoz_s42_r10_p1_n200.jsonlData3.1 MB—
data/probes/probes_multiwoz_s42_r10_p3_n200.jsonlData2.4 MB—
data/probes/probes_multiwoz_s42_r30_n200_late_intent.jsonlData4.4 MB—
data/probes/probes_multiwoz_s42_r30_p1_n200.jsonlData3.9 MB—
data/probes/probes_multiwoz_s42_r30_p3_n200.jsonlData3.5 MB—
data/probes/probes_multiwoz_s42_r50_p3_n200.jsonlData2.3 MB—
data/probes/probes_multiwoz_s42_r70_p3_n200.jsonlData2.7 MB—
data/probes/probes_multiwoz_s43_r10_p1_n200.jsonlData3.1 MB—
data/probes/probes_multiwoz_s43_r10_p3_n200.jsonlData2.5 MB—
data/probes/probes_multiwoz_s43_r30_n200_late_intent.jsonlData4.4 MB—
data/probes/probes_multiwoz_s43_r30_p1_n200.jsonlData3.9 MB—
data/probes/probes_multiwoz_s43_r30_p3_n200.jsonlData3.2 MB—
data/probes/probes_multiwoz_s43_r50_p3_n200.jsonlData2.4 MB—
data/probes/probes_multiwoz_s43_r70_p3_n200.jsonlData2.9 MB—
data/probes/probes_multiwoz_s44_r10_p1_n200.jsonlData3.1 MB—
data/probes/probes_multiwoz_s44_r10_p3_n200.jsonlData2.5 MB—
data/probes/probes_multiwoz_s44_r30_n200_late_intent.jsonlData4.3 MB—
data/probes/probes_multiwoz_s44_r30_p1_n200.jsonlData3.8 MB—
data/probes/probes_multiwoz_s44_r30_p3_n200.jsonlData3.1 MB—
data/probes/probes_multiwoz_s44_r50_p3_n200.jsonlData2.3 MB—
data/probes/probes_multiwoz_s44_r70_p3_n200.jsonlData2.8 MB—
data/probes/probes_risawoz_all207_r30_s42-46.jsonlData7.8 MB—
data/probes/probes_risawoz_s42_r10_n200.jsonlData3.7 MB—
data/probes/probes_risawoz_s42_r30_n200.jsonlData4.5 MB—
data/probes/probes_sgd_s42_r10_p1_n200.jsonlData3.1 MB—
data/probes/probes_sgd_s42_r10_p3_n200.jsonlData1.9 MB—
data/probes/probes_sgd_s42_r30_n200_late_intent.jsonlData4.1 MB—
data/probes/probes_sgd_s42_r30_p1_n200.jsonlData4.0 MB—
data/probes/probes_sgd_s42_r30_p3_n200.jsonlData2.8 MB—
data/probes/probes_sgd_s42_r50_p3_n200.jsonlData1.9 MB—
data/probes/probes_sgd_s42_r70_p3_n200.jsonlData2.2 MB—
data/probes/probes_sgd_s43_r10_p1_n200.jsonlData3.1 MB—
data/probes/probes_sgd_s43_r10_p3_n200.jsonlData1.8 MB—
data/probes/probes_sgd_s43_r30_n200_late_intent.jsonlData4.2 MB—
data/probes/probes_sgd_s43_r30_p1_n200.jsonlData3.9 MB—
data/probes/probes_sgd_s43_r30_p3_n200.jsonlData2.3 MB—
data/probes/probes_sgd_s43_r50_p3_n200.jsonlData1.8 MB—
data/probes/probes_sgd_s43_r70_p3_n200.jsonlData2.1 MB—
data/probes/probes_sgd_s44_r10_p1_n200.jsonlData3.1 MB—
data/probes/probes_sgd_s44_r10_p3_n200.jsonlData2.0 MB—
data/probes/probes_sgd_s44_r30_n200_late_intent.jsonlData4.2 MB—
data/probes/probes_sgd_s44_r30_p1_n200.jsonlData3.9 MB—
data/probes/probes_sgd_s44_r30_p3_n200.jsonlData2.5 MB—
data/probes/probes_sgd_s44_r50_p3_n200.jsonlData1.9 MB—
data/probes/probes_sgd_s44_r70_p3_n200.jsonlData2.3 MB—
data/semantic_evaluation/ae_development_inputs.jsonlData212.6 KB—
data/semantic_evaluation/ae_development_key.jsonlData75.2 KB—
data/semantic_evaluation/ae_source_manifest.jsonData4.3 KB—
data/semantic_evaluation/ae_test_inputs.jsonlData582.6 KB—
data/semantic_evaluation/ae_test_key.jsonlData225.7 KB—
data/semantic_evaluation/core_goal_prompt.txtData2.4 KB—
data/semantic_evaluation/judge_prompt.txtData2.1 KB—
data/semantic_evaluation/p1_inputs.jsonlData650.1 KB—
data/semantic_evaluation/p1_key.jsonlData2.8 MB—
data/semantic_evaluation/p1_source_manifest.jsonData1.1 KB—
data/semantic_evaluation/rubric_controls_inputs.jsonlData3.1 KB—
data/sources/longmemeval_knowledge_updates.jsonData40.1 MB3b376e0c0d1f
data/sources/risawoz_test.jsonData11.4 MB39aad40f5357
dataset_card/croissant.jsonData45.8 KB—
manifest.jsonData60.1 KB—
requirements-fmts.txtData171 B—
requirements.txtData1.0 KB—
results/aggregates_flat.jsonlData153.1 KB—
results/bench/efficiency_multiwoz_r10.jsonData1.2 KB—
results/bench/efficiency_multiwoz_r30.jsonData1.8 KB—
results/bench/efficiency_sgd_r10.jsonData1.2 KB—
results/bench/efficiency_sgd_r30.jsonData2.1 KB—
results/diagnostics/chunkkv_chunklen/mw_r30_p3_chunkkv_chunklen_r03_aggregate.jsonData880 B—
results/diagnostics/chunkkv_chunklen/mw_r30_p3_chunkkv_chunklen_r03_scored.jsonlData491.0 KB—
results/diagnostics/mistral_128/mw_r10_s42_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r10_s42_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r10_s42_p3_reader_all.jsonlData1.2 MB—
results/diagnostics/mistral_128/mw_r10_s42_p3_scored.jsonlData970.6 KB—
results/diagnostics/mistral_128/mw_r10_s43_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r10_s43_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r10_s43_p3_reader_all.jsonlData1.3 MB—
results/diagnostics/mistral_128/mw_r10_s43_p3_scored.jsonlData1.0 MB—
results/diagnostics/mistral_128/mw_r10_s44_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r10_s44_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r10_s44_p3_reader_all.jsonlData1.3 MB—
results/diagnostics/mistral_128/mw_r10_s44_p3_scored.jsonlData1.0 MB—
results/diagnostics/mistral_128/mw_r30_s42_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r30_s42_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r30_s42_p3_reader_all.jsonlData1.7 MB—
results/diagnostics/mistral_128/mw_r30_s42_p3_scored.jsonlData1.4 MB—
results/diagnostics/mistral_128/mw_r30_s43_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r30_s43_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r30_s43_p3_reader_all.jsonlData1.3 MB—
results/diagnostics/mistral_128/mw_r30_s43_p3_scored.jsonlData1.0 MB—
results/diagnostics/mistral_128/mw_r30_s44_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r30_s44_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/mw_r30_s44_p3_reader_all.jsonlData1.3 MB—
results/diagnostics/mistral_128/mw_r30_s44_p3_scored.jsonlData1.0 MB—
results/diagnostics/mistral_128/seed_summary_mistral128.jsonData127.0 KB—
results/diagnostics/mistral_128/sgd_r10_s42_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r10_s42_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r10_s42_p3_reader_all.jsonlData935.4 KB—
results/diagnostics/mistral_128/sgd_r10_s42_p3_scored.jsonlData763.0 KB—
results/diagnostics/mistral_128/sgd_r10_s43_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r10_s43_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r10_s43_p3_reader_all.jsonlData914.0 KB—
results/diagnostics/mistral_128/sgd_r10_s43_p3_scored.jsonlData744.1 KB—
results/diagnostics/mistral_128/sgd_r10_s44_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r10_s44_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r10_s44_p3_reader_all.jsonlData965.8 KB—
results/diagnostics/mistral_128/sgd_r10_s44_p3_scored.jsonlData784.1 KB—
results/diagnostics/mistral_128/sgd_r30_s42_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r30_s42_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r30_s42_p3_reader_all.jsonlData1.3 MB—
results/diagnostics/mistral_128/sgd_r30_s42_p3_scored.jsonlData1.1 MB—
results/diagnostics/mistral_128/sgd_r30_s43_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r30_s43_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r30_s43_p3_reader_all.jsonlData918.0 KB—
results/diagnostics/mistral_128/sgd_r30_s43_p3_scored.jsonlData741.3 KB—
results/diagnostics/mistral_128/sgd_r30_s44_p1_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r30_s44_p3_aggregate.jsonData2.1 KB—
results/diagnostics/mistral_128/sgd_r30_s44_p3_reader_all.jsonlData971.0 KB—
results/diagnostics/mistral_128/sgd_r30_s44_p3_scored.jsonlData781.3 KB—
results/diagnostics/update_evidence/fmts_support_ablation_input.jsonlData712.9 KB—
results/diagnostics/update_evidence/fmts_support_ablation_reader_llama.jsonlData133.2 KB—
results/diagnostics/update_evidence/fmts_support_ablation_scored_llama.jsonlData134.9 KB—
results/diagnostics/update_evidence/fmts_support_ablation_summary_llama.jsonData1.6 KB—
results/diagnostics/update_evidence/multiwoz_filtered_gold.jsonlData1.2 MB—
results/diagnostics/update_evidence/multiwoz_goal_labels.jsonlData749.0 KB—
results/diagnostics/update_evidence/origin_index_multiwoz.jsonlData327.7 KB—
results/diagnostics/update_evidence/origin_index_sgd.jsonlData403.8 KB—
results/diagnostics/update_evidence/origin_summary.jsonData2.0 KB—
results/diagnostics/update_evidence/p2_aggregate_llama.jsonData2.8 KB—
results/diagnostics/update_evidence/p2_all_reader_llama.jsonlData1.7 MB—
results/diagnostics/update_evidence/p2_reader_input.jsonlData6.6 MB—
results/diagnostics/update_evidence/p2_reader_input_multiwoz.jsonlData3.6 MB—
results/diagnostics/update_evidence/p2_reader_input_sgd.jsonlData3.0 MB—
results/diagnostics/update_evidence/p2_scored_llama.jsonlData1.7 MB—
results/diagnostics/update_evidence/p2_summary_llama.jsonData8.2 KB—
results/diagnostics/update_evidence/p2_variants_multiwoz.jsonlData2.9 MB—
results/diagnostics/update_evidence/p2_variants_sgd.jsonlData2.4 MB—
results/diagnostics/update_evidence/p3_crossseed_llama.jsonData54.7 KB—
results/diagnostics/update_evidence/p3_crossseed_mistral.jsonData55.1 KB—
results/diagnostics/update_evidence/p3_multiwoz_s42_aggregate_chunkkv.jsonData405 B—
results/diagnostics/update_evidence/p3_multiwoz_s42_aggregate_llama.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_multiwoz_s42_aggregate_mistral.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_multiwoz_s42_reader_chunkkv.jsonlData201.2 KB—
results/diagnostics/update_evidence/p3_multiwoz_s42_reader_llama.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s42_reader_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s42_scored_chunkkv.jsonlData214.3 KB—
results/diagnostics/update_evidence/p3_multiwoz_s42_scored_llama.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s42_scored_mistral.jsonlData1.6 MB—
results/diagnostics/update_evidence/p3_multiwoz_s43_aggregate_llama.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_multiwoz_s43_aggregate_mistral.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_multiwoz_s43_reader_llama.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s43_reader_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s43_scored_llama.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s43_scored_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s44_aggregate_llama.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_multiwoz_s44_aggregate_mistral.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_multiwoz_s44_reader_llama.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s44_reader_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s44_scored_llama.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_multiwoz_s44_scored_mistral.jsonlData1.6 MB—
results/diagnostics/update_evidence/p3_pool_multiwoz.jsonData11.3 KB—
results/diagnostics/update_evidence/p3_pool_sgd.jsonData14.1 KB—
results/diagnostics/update_evidence/p3_sgd_s42_aggregate_chunkkv.jsonData412 B—
results/diagnostics/update_evidence/p3_sgd_s42_aggregate_llama.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_sgd_s42_aggregate_mistral.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_sgd_s42_reader_chunkkv.jsonlData197.8 KB—
results/diagnostics/update_evidence/p3_sgd_s42_reader_llama.jsonlData1.4 MB—
results/diagnostics/update_evidence/p3_sgd_s42_reader_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_sgd_s42_scored_chunkkv.jsonlData205.9 KB—
results/diagnostics/update_evidence/p3_sgd_s42_scored_llama.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_sgd_s42_scored_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_sgd_s43_aggregate_llama.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_sgd_s43_aggregate_mistral.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_sgd_s43_reader_llama.jsonlData1.4 MB—
results/diagnostics/update_evidence/p3_sgd_s43_reader_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_sgd_s43_scored_llama.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_sgd_s43_scored_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_sgd_s44_aggregate_llama.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_sgd_s44_aggregate_mistral.jsonData3.2 KB—
results/diagnostics/update_evidence/p3_sgd_s44_reader_llama.jsonlData1.4 MB—
results/diagnostics/update_evidence/p3_sgd_s44_reader_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_sgd_s44_scored_llama.jsonlData1.5 MB—
results/diagnostics/update_evidence/p3_sgd_s44_scored_mistral.jsonlData1.5 MB—
results/diagnostics/update_evidence/per_item_multiwoz.jsonlData108.3 KB—
results/diagnostics/update_evidence/per_item_sgd.jsonlData84.2 KB—
results/diagnostics/update_evidence/probes_multiwoz_s42_r30_n200_late_intent.jsonlData4.4 MB—
results/diagnostics/update_evidence/probes_multiwoz_s42_r30_n200_late_intent_chunkkv.jsonlData980.6 KB—
results/diagnostics/update_evidence/probes_multiwoz_s43_r30_n200_late_intent.jsonlData4.4 MB—
results/diagnostics/update_evidence/probes_multiwoz_s44_r30_n200_late_intent.jsonlData4.3 MB—
results/diagnostics/update_evidence/probes_sgd_s42_r30_n200_late_intent.jsonlData4.1 MB—
results/diagnostics/update_evidence/probes_sgd_s42_r30_n200_late_intent_chunkkv.jsonlData927.4 KB—
results/diagnostics/update_evidence/probes_sgd_s43_r30_n200_late_intent.jsonlData4.2 MB—
results/diagnostics/update_evidence/probes_sgd_s44_r30_n200_late_intent.jsonlData4.2 MB—
results/diagnostics/update_evidence/rank_inconsistency_final_slot.jsonData10.4 KB—
results/diagnostics/update_evidence/sgd_filtered_gold.jsonlData1.5 MB—
results/diagnostics/update_evidence/sgd_goal_labels.jsonlData924.6 KB—
results/fmts_execution/answers.jsonlData2.0 MB—
results/fmts_execution/protocol.jsonData1.1 KB—
results/fmts_execution/tasks.jsonlData4.0 MB—
results/fmts_refpool/indirect_seed42_refpool_r010.jsonData1.7 MB—
results/fmts_refpool/indirect_seed42_refpool_r030.jsonData2.2 MB—
results/fmts_refpool/multiwoz_scope_refpool_r010.jsonData2.8 MB—
results/fmts_refpool/multiwoz_scope_refpool_r030.jsonData3.5 MB—
results/fmts_refpool/scored_tqa_indirect_seed42_refpool_r010.jsonData98.0 KB—
results/fmts_refpool/scored_tqa_indirect_seed42_refpool_r030.jsonData99.4 KB—
results/fmts_refpool/scored_tqa_multiwoz_scope_refpool_r010.jsonData182.4 KB—
results/fmts_refpool/scored_tqa_multiwoz_scope_refpool_r030.jsonData183.0 KB—
results/fmts_refpool/scored_tqa_sgd_scope_refpool_r010.jsonData172.6 KB—
results/fmts_refpool/scored_tqa_sgd_scope_refpool_r030.jsonData173.9 KB—
results/fmts_refpool/sgd_scope_refpool_r010.jsonData2.9 MB—
results/fmts_refpool/sgd_scope_refpool_r030.jsonData3.7 MB—
results/kv_methods/multiwoz_r30_s42_p1_pyramidkv_aggregate.jsonData269 B—
results/kv_methods/multiwoz_r30_s42_p1_snapkv_aggregate.jsonData266 B—
results/kv_methods/multiwoz_r30_s42_p1_streamingllm_aggregate.jsonData276 B—
results/kv_methods/multiwoz_r30_s42_p3_pyramidkv_aggregate.jsonData309 B—
results/kv_methods/multiwoz_r30_s42_p3_pyramidkv_scored.jsonlData166.3 KB—
results/kv_methods/multiwoz_r30_s42_p3_snapkv_aggregate.jsonData307 B—
results/kv_methods/multiwoz_r30_s42_p3_snapkv_scored.jsonlData165.8 KB—
results/kv_methods/multiwoz_r30_s42_p3_streamingllm_aggregate.jsonData314 B—
results/kv_methods/multiwoz_r30_s42_p3_streamingllm_scored.jsonlData167.0 KB—
results/kv_methods/sgd_r30_s42_p1_pyramidkv_aggregate.jsonData272 B—
results/kv_methods/sgd_r30_s42_p1_snapkv_aggregate.jsonData264 B—
results/kv_methods/sgd_r30_s42_p1_streamingllm_aggregate.jsonData271 B—
results/kv_methods/sgd_r30_s42_p3_pyramidkv_aggregate.jsonData303 B—
results/kv_methods/sgd_r30_s42_p3_pyramidkv_scored.jsonlData131.0 KB—
results/kv_methods/sgd_r30_s42_p3_snapkv_aggregate.jsonData300 B—
results/kv_methods/sgd_r30_s42_p3_snapkv_scored.jsonlData130.5 KB—
results/kv_methods/sgd_r30_s42_p3_streamingllm_aggregate.jsonData302 B—
results/kv_methods/sgd_r30_s42_p3_streamingllm_scored.jsonlData132.0 KB—
results/lme_ku/lme_ku_s42_r10_n72_sdpa_aggregate.jsonData1.4 KB—
results/lme_ku/lme_ku_s42_r10_n72_sdpa_reader_shard0.jsonlData244.1 KB—
results/lme_ku/lme_ku_s42_r10_n72_sdpa_scored.jsonlData195.8 KB—
results/lme_ku/lme_ku_s42_r30_n72_sdpa_aggregate.jsonData1.4 KB—
results/lme_ku/lme_ku_s42_r30_n72_sdpa_reader_shard0.jsonlData245.4 KB—
results/lme_ku/lme_ku_s42_r30_n72_sdpa_scored.jsonlData195.2 KB—
results/lme_ku/lme_ku_s43_r10_n72_sdpa_aggregate.jsonData1.4 KB—
results/lme_ku/lme_ku_s43_r10_n72_sdpa_reader_shard0.jsonlData244.1 KB—
results/lme_ku/lme_ku_s43_r10_n72_sdpa_scored.jsonlData195.8 KB—
results/lme_ku/lme_ku_s43_r30_n72_sdpa_aggregate.jsonData1.4 KB—
results/lme_ku/lme_ku_s43_r30_n72_sdpa_reader_shard0.jsonlData245.4 KB—
results/lme_ku/lme_ku_s43_r30_n72_sdpa_scored.jsonlData195.2 KB—
results/main/chunkkv_mw_r30_p1_aggregate.jsonData182 B—
results/main/chunkkv_mw_r30_p3_aggregate.jsonData309 B—
results/main/chunkkv_mw_r30_p3_scored.jsonlData164.4 KB—
results/main/chunkkv_sgd_r30_p1_aggregate.jsonData182 B—
results/main/chunkkv_sgd_r30_p3_aggregate.jsonData304 B—
results/main/chunkkv_sgd_r30_p3_scored.jsonlData130.3 KB—
results/main/mw_r10_s42_aggregate.jsonData1.4 KB—
results/main/mw_r10_s42_p1_tokenf1_audit_aggregate.jsonData1.8 KB—
results/main/mw_r10_s42_p3_aggregate.jsonData2.2 KB—
results/main/mw_r10_s42_p3_scored.jsonlData965.7 KB—
results/main/mw_r10_s43_aggregate.jsonData1.4 KB—
results/main/mw_r10_s43_p3_aggregate.jsonData2.2 KB—
results/main/mw_r10_s43_p3_scored.jsonlData1.0 MB—
results/main/mw_r10_s44_aggregate.jsonData1.4 KB—
results/main/mw_r10_s44_p3_aggregate.jsonData2.2 KB—
results/main/mw_r10_s44_p3_scored.jsonlData1.0 MB—
results/main/mw_r30_s42_aggregate.jsonData1.4 KB—
results/main/mw_r30_s42_p1_tokenf1_audit_aggregate.jsonData1.8 KB—
results/main/mw_r30_s42_p3_aggregate.jsonData2.2 KB—
results/main/mw_r30_s42_p3_scored.jsonlData1.2 MB—
results/main/mw_r30_s43_aggregate.jsonData1.4 KB—
results/main/mw_r30_s43_p3_aggregate.jsonData2.2 KB—
results/main/mw_r30_s43_p3_scored.jsonlData1.0 MB—
results/main/mw_r30_s44_aggregate.jsonData1.4 KB—
results/main/mw_r30_s44_p3_aggregate.jsonData2.2 KB—
results/main/mw_r30_s44_p3_scored.jsonlData1.0 MB—
results/main/mw_r50_s42_aggregate.jsonData1.3 KB—
results/main/mw_r50_s42_p3_aggregate.jsonData1.4 KB—
results/main/mw_r50_s42_p3_scored.jsonlData523.5 KB—
results/main/mw_r50_s43_aggregate.jsonData858 B—
results/main/mw_r50_s43_p3_aggregate.jsonData1.4 KB—
results/main/mw_r50_s43_p3_scored.jsonlData545.6 KB—
results/main/mw_r50_s44_aggregate.jsonData1.3 KB—
results/main/mw_r50_s44_p3_aggregate.jsonData1.4 KB—
results/main/mw_r50_s44_p3_scored.jsonlData547.6 KB—
results/main/mw_r70_s42_aggregate.jsonData1.3 KB—
results/main/mw_r70_s42_p3_aggregate.jsonData1.4 KB—
results/main/mw_r70_s42_p3_scored.jsonlData523.8 KB—
results/main/mw_r70_s43_aggregate.jsonData854 B—
results/main/mw_r70_s43_p3_aggregate.jsonData1.4 KB—
results/main/mw_r70_s43_p3_scored.jsonlData545.7 KB—
results/main/mw_r70_s44_aggregate.jsonData1.3 KB—
results/main/mw_r70_s44_p3_aggregate.jsonData1.4 KB—
results/main/mw_r70_s44_p3_scored.jsonlData547.9 KB—
results/main/p1_token_f1_audit.jsonData5.7 KB—
results/main/seed_summary.jsonData101.0 KB—
results/main/seed_summary_wider.jsonData71.3 KB—
results/main/sgd_r10_s42_aggregate.jsonData1.4 KB—
results/main/sgd_r10_s42_p1_tokenf1_audit_aggregate.jsonData1.8 KB—
results/main/sgd_r10_s42_p3_aggregate.jsonData2.2 KB—
results/main/sgd_r10_s42_p3_scored.jsonlData759.7 KB—
results/main/sgd_r10_s43_aggregate.jsonData1.4 KB—
results/main/sgd_r10_s43_p3_aggregate.jsonData2.1 KB—
results/main/sgd_r10_s43_p3_scored.jsonlData738.1 KB—
results/main/sgd_r10_s44_aggregate.jsonData1.4 KB—
results/main/sgd_r10_s44_p3_aggregate.jsonData2.2 KB—
results/main/sgd_r10_s44_p3_scored.jsonlData778.5 KB—
results/main/sgd_r30_s42_aggregate.jsonData1.4 KB—
results/main/sgd_r30_s42_p1_tokenf1_audit_aggregate.jsonData1.8 KB—
results/main/sgd_r30_s42_p3_aggregate.jsonData2.2 KB—
results/main/sgd_r30_s42_p3_scored.jsonlData946.9 KB—
results/main/sgd_r30_s43_aggregate.jsonData1.4 KB—
results/main/sgd_r30_s43_p3_aggregate.jsonData2.1 KB—
results/main/sgd_r30_s43_p3_scored.jsonlData739.0 KB—
results/main/sgd_r30_s44_aggregate.jsonData1.4 KB—
results/main/sgd_r30_s44_p3_aggregate.jsonData2.2 KB—
results/main/sgd_r30_s44_p3_scored.jsonlData779.2 KB—
results/main/sgd_r50_s42_aggregate.jsonData1.3 KB—
results/main/sgd_r50_s42_p3_aggregate.jsonData1.4 KB—
results/main/sgd_r50_s42_p3_scored.jsonlData411.8 KB—
results/main/sgd_r50_s43_aggregate.jsonData848 B—
results/main/sgd_r50_s43_p3_aggregate.jsonData1.3 KB—
results/main/sgd_r50_s43_p3_scored.jsonlData399.9 KB—
results/main/sgd_r50_s44_aggregate.jsonData1.3 KB—
results/main/sgd_r50_s44_p3_aggregate.jsonData1.4 KB—
results/main/sgd_r50_s44_p3_scored.jsonlData421.7 KB—
results/main/sgd_r70_s42_aggregate.jsonData1.3 KB—
results/main/sgd_r70_s42_p3_aggregate.jsonData1.4 KB—
results/main/sgd_r70_s42_p3_scored.jsonlData410.9 KB—
results/main/sgd_r70_s43_aggregate.jsonData843 B—
results/main/sgd_r70_s43_p3_aggregate.jsonData1.3 KB—
results/main/sgd_r70_s43_p3_scored.jsonlData399.4 KB—
results/main/sgd_r70_s44_aggregate.jsonData1.3 KB—
results/main/sgd_r70_s44_p3_aggregate.jsonData1.4 KB—
results/main/sgd_r70_s44_p3_scored.jsonlData421.3 KB—
results/p1_saved/mw_r10_s42_scored.jsonlData483.0 KB—
results/p1_saved/mw_r10_s43_scored.jsonlData479.5 KB—
results/p1_saved/mw_r10_s44_scored.jsonlData484.5 KB—
results/p1_saved/mw_r30_s42_scored.jsonlData482.1 KB—
results/p1_saved/mw_r30_s43_scored.jsonlData479.8 KB—
results/p1_saved/mw_r30_s44_scored.jsonlData484.2 KB—
results/p1_saved/provenance.jsonData1.9 KB—
results/p1_saved/sgd_r10_s42_scored.jsonlData428.6 KB—
results/p1_saved/sgd_r10_s43_scored.jsonlData435.1 KB—
results/p1_saved/sgd_r10_s44_scored.jsonlData430.3 KB—
results/p1_saved/sgd_r30_s42_scored.jsonlData429.0 KB—
results/p1_saved/sgd_r30_s43_scored.jsonlData435.7 KB—
results/p1_saved/sgd_r30_s44_scored.jsonlData430.8 KB—
results/p3_segment_sensitivity/changed_answers.jsonlData628.5 KB—
results/p3_segment_sensitivity/per_answer.jsonlData12.0 MBe79a208279f3
results/p3_segment_sensitivity/summary.jsonData137.8 KB—
results/risawoz/risawoz_s42_r10_n200_aggregate.jsonData1.7 KB—
results/risawoz/risawoz_s42_r30_n200_aggregate.jsonData1.7 KB—
results/seed_robustness/aggregates/lme_ku_p2.jsonData2.5 KB—
results/seed_robustness/aggregates/risawoz_p1.jsonData2.8 KB—
results/seed_robustness/aggregates/risawoz_p2.jsonData2.4 KB—
results/seed_robustness/experiment.jsonData2.7 KB—
results/seed_robustness/manifest.jsonData2.9 KB—
results/seed_robustness/reader/lme_ku_all72_r30_llama.jsonlData402.6 KB—
results/seed_robustness/reader/risawoz_all207_r30_llama.jsonlData1.9 MB—
results/seed_robustness/scored/lme_ku_p2.jsonlData351.6 KB—
results/seed_robustness/scored/risawoz_p1.jsonlData981.2 KB—
results/seed_robustness/scored/risawoz_p2.jsonlData1.0 MB—
results/seed_robustness/stats/summary.jsonData19.7 KB—
results/semantic_evaluation/ae_development/execution_summary.jsonData358 B—
results/semantic_evaluation/ae_development/judgments.jsonlData53.0 KB—
results/semantic_evaluation/ae_development/model_record.jsonData719 B—
results/semantic_evaluation/ae_development/raw_responses.jsonlData1.4 MB—
results/semantic_evaluation/ae_development_metrics.jsonData6.8 KB—
results/semantic_evaluation/ae_test/execution_summary.jsonData391 B—
results/semantic_evaluation/ae_test/judgments.jsonlData159.7 KB—
results/semantic_evaluation/ae_test/model_record.jsonData719 B—
results/semantic_evaluation/ae_test/raw_responses.jsonlData4.1 MB—
results/semantic_evaluation/ae_test_metrics.jsonData6.8 KB—
results/semantic_evaluation/p1/execution_summary.jsonData377 B—
results/semantic_evaluation/p1/judgments.jsonlData965.2 KB—
results/semantic_evaluation/p1/model_record.jsonData719 B—
results/semantic_evaluation/p1/raw_responses.jsonlData21.4 MB754459097f00
results/semantic_evaluation/p1_action_domain/execution_summary.jsonData463 B—
results/semantic_evaluation/p1_action_domain/judgments.jsonlData731.0 KB—
results/semantic_evaluation/p1_action_domain/model_record.jsonData719 B—
results/semantic_evaluation/p1_action_domain/raw_responses.jsonlData20.6 MBe19b07f1de98
results/semantic_evaluation/p1_report/rank_and_disagreement_analysis.jsonData8.0 KB—
results/semantic_evaluation/p1_summary.jsonData34.0 KB—
results/semantic_evaluation/rubric_controls/conformance.jsonData2.0 KB—
results/semantic_evaluation/rubric_controls/execution_summary.jsonData346 B—
results/semantic_evaluation/rubric_controls/judgments.jsonlData3.4 KB—
results/semantic_evaluation/rubric_controls/model_record.jsonData719 B—
results/semantic_evaluation/rubric_controls/raw_responses.jsonlData113.0 KB—
LICENSEDocumentation1.8 KB—
README.mdDocumentation8.4 KB—
RESULTS_MAP.mdDocumentation4.0 KB—
SEMANTIC_EVALUATION.mdDocumentation7.3 KB—
results/p3_segment_sensitivity/REPORT.mdDocumentation3.0 KB—
results/seed_robustness/stats/summary.mdDocumentation1.4 KB—
results/semantic_evaluation/p1_report/REPORT.mdDocumentation2.2 KB—
scripts/fmts_refpool/README.mdDocumentation2.6 KB—
run_all.shOther550 B—
scripts/audit/anchor_filter_audit.pyOther5.1 KB—
scripts/audit/rank_inconsistency_final_slot.pyOther4.7 KB—
scripts/audit/recompute_p1_audit.pyOther2.9 KB—
scripts/audit/validate_fmts_support_ablation.pyOther10.8 KB—
scripts/audit/validate_saved_readers.pyOther4.7 KB—
scripts/build_aggregates_flat.pyOther10.6 KB—
scripts/builders/build_deletions.pyOther2.2 KB—
scripts/builders/build_lme_probes.pyOther7.7 KB—
scripts/builders/build_lme_transitions.pyOther9.9 KB—
scripts/builders/build_probes.pyOther4.6 KB—
scripts/builders/build_risawoz_probes.pyOther13.8 KB—
scripts/builders/construction.pyOther6.7 KB—
scripts/check_artifact.pyOther1.8 KB—
scripts/fmts_refpool/core/boundary_detector.pyOther6.2 KB—
scripts/fmts_refpool/core/compression_methods.pyOther6.3 KB—
scripts/fmts_refpool/core/evaluation.pyOther6.3 KB—
scripts/fmts_refpool/core/turn_types.pyOther154 B—
scripts/fmts_refpool/run_answers.pyOther3.2 KB—
scripts/fmts_refpool/run_compression.pyOther10.9 KB—
scripts/normalize.pyOther5.8 KB—
scripts/reader/chunkkv_vendored/__init__.pyOther1.0 KB—
scripts/reader/chunkkv_vendored/base_press.pyOther4.1 KB—
scripts/reader/chunkkv_vendored/chunkkv_press.pyOther4.8 KB—
scripts/reader/chunkkv_vendored/scorer_press.pyOther1.8 KB—
scripts/reader/chunkkv_vendored/snapkv_press.pyOther3.3 KB—
scripts/reader/chunkkv_vendored/utils.pyOther3.3 KB—
scripts/reader/reader_worker_chunkkv.pyOther15.1 KB—
scripts/reader/reader_worker_kvpress_extra.pyOther16.5 KB—
scripts/reader/reader_worker_local.pyOther8.4 KB—
scripts/scorer/scorer_lme_ku.pyOther10.0 KB—
scripts/scorer/scorer_p1.pyOther11.1 KB—
scripts/scorer/scorer_p1_late.pyOther12.9 KB—
scripts/scorer/scorer_p1_zh.pyOther6.1 KB—
scripts/scorer/scorer_p3.pyOther8.2 KB—
scripts/seed_robustness/bootstrap_stats.pyOther10.6 KB—
scripts/seed_robustness/build_seed_robustness.pyOther7.0 KB—
scripts/seed_robustness/make_manifest.pyOther2.8 KB—
scripts/seed_robustness/validate_seed_robustness.pyOther4.3 KB—
scripts/semantic_evaluation/aggregate_ae_judge.pyOther4.3 KB—
scripts/semantic_evaluation/aggregate_p1_semantic.pyOther5.9 KB—
scripts/semantic_evaluation/audit_p3_segments.pyOther11.0 KB—
scripts/semantic_evaluation/check_core_rubric.pyOther2.5 KB—
scripts/semantic_evaluation/prepare_ae_validation.pyOther6.4 KB—
scripts/semantic_evaluation/prepare_p1_semantic.pyOther3.0 KB—
scripts/semantic_evaluation/recover_request_ids.pyOther2.1 KB—
scripts/semantic_evaluation/report_p1_semantic.pyOther5.4 KB—
scripts/semantic_evaluation/run_core_goal_judge.pyOther670 B—
scripts/semantic_evaluation/run_semantic_judge.pyOther8.4 KB—
scripts/semantic_evaluation/validate_evaluation_outputs.pyOther3.2 KB—
scripts/tests/test_construction.pyOther1.6 KB—
scripts/tests/test_corpus_construction.pyOther1.3 KB—
scripts/tests/test_fmts.pyOther1.7 KB—
scripts/validate_corpus_scorers.pyOther2.9 KB—
.gitattributesRepository3.2 KB—

License and Download

License
other
Access
No access gate
Download from Papersubmission

Released by Papersubmission through its official repository on Hugging Face.