Hugging Face's logo

koichi12
/
llm-scripts

Model card Files
xet
Community
koichi12 commited on
Commit
4801538
·
verified ·
1 Parent(s): f3d9fcb

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. scripts/yans/lm-evaluation-harness/tests/testdata/anli_r1-v0-res.json +1 -0
  2. scripts/yans/lm-evaluation-harness/tests/testdata/arithmetic_1dc-v0-res.json +1 -0
  3. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_animate_subject_trans-v0-res.json +1 -0
  4. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_determiner_noun_agreement_2-v0-res.json +1 -0
  5. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_determiner_noun_agreement_irregular_1-v0-res.json +1 -0
  6. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_determiner_noun_agreement_irregular_2-v0-loglikelihood +1 -0
  7. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_determiner_noun_agreement_irregular_2-v0-res.json +1 -0
  8. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_drop_argument-v0-res.json +1 -0
  9. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_existential_there_quantifiers_2-v0-res.json +1 -0
  10. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_left_branch_island_simple_question-v0-loglikelihood +1 -0
  11. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_npi_present_2-v0-loglikelihood +1 -0
  12. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_principle_A_case_1-v0-loglikelihood +1 -0
  13. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_principle_A_case_2-v0-loglikelihood +1 -0
  14. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_principle_A_domain_1-v0-res.json +1 -0
  15. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_principle_A_domain_3-v0-res.json +1 -0
  16. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_superlative_quantifiers_1-v0-res.json +1 -0
  17. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_tough_vs_raising_1-v0-res.json +1 -0
  18. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_wh_questions_subject_gap_long_distance-v0-loglikelihood +1 -0
  19. scripts/yans/lm-evaluation-harness/tests/testdata/blimp_wh_vs_that_with_gap-v0-res.json +1 -0
  20. scripts/yans/lm-evaluation-harness/tests/testdata/cb-v0-loglikelihood +1 -0
  21. scripts/yans/lm-evaluation-harness/tests/testdata/cola-v0-res.json +1 -0
  22. scripts/yans/lm-evaluation-harness/tests/testdata/coqa-v1-greedy_until +1 -0
  23. scripts/yans/lm-evaluation-harness/tests/testdata/crows_pairs_english_age-v0-loglikelihood +1 -0
  24. scripts/yans/lm-evaluation-harness/tests/testdata/crows_pairs_english_physical_appearance-v0-res.json +1 -0
  25. scripts/yans/lm-evaluation-harness/tests/testdata/crows_pairs_english_sexual_orientation-v0-loglikelihood +1 -0
  26. scripts/yans/lm-evaluation-harness/tests/testdata/crows_pairs_french_sexual_orientation-v0-loglikelihood +1 -0
  27. scripts/yans/lm-evaluation-harness/tests/testdata/cycle_letters-v0-res.json +1 -0
  28. scripts/yans/lm-evaluation-harness/tests/testdata/ethics_cm-v0-loglikelihood +1 -0
  29. scripts/yans/lm-evaluation-harness/tests/testdata/gguf_test_44e268d15decc4d2d0f99e57e1476269826cd3b54262f7a0981f75ddd45b25d0.pkl +3 -0
  30. scripts/yans/lm-evaluation-harness/tests/testdata/gpt3_test_8025023377febbd8c5f2b9f26705c394ff375d0cad7c89c10fd9b8e1eb66ff1c.pkl +3 -0
  31. scripts/yans/lm-evaluation-harness/tests/testdata/headqa-v0-res.json +1 -0
  32. scripts/yans/lm-evaluation-harness/tests/testdata/headqa_es-v0-loglikelihood +1 -0
  33. scripts/yans/lm-evaluation-harness/tests/testdata/headqa_es-v0-res.json +1 -0
  34. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-abstract_algebra-v0-res.json +1 -0
  35. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-electrical_engineering-v0-loglikelihood +1 -0
  36. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_biology-v0-loglikelihood +1 -0
  37. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_biology-v0-res.json +1 -0
  38. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_computer_science-v0-res.json +1 -0
  39. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_european_history-v0-res.json +1 -0
  40. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_macroeconomics-v0-loglikelihood +1 -0
  41. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-moral_disputes-v0-loglikelihood +1 -0
  42. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-nutrition-v0-loglikelihood +1 -0
  43. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-prehistory-v0-loglikelihood +1 -0
  44. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-professional_law-v0-res.json +1 -0
  45. scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-professional_medicine-v0-res.json +1 -0
  46. scripts/yans/lm-evaluation-harness/tests/testdata/lambada_mt_de-v0-res.json +1 -0
  47. scripts/yans/lm-evaluation-harness/tests/testdata/lambada_mt_fr-v0-res.json +1 -0
  48. scripts/yans/lm-evaluation-harness/tests/testdata/lambada_standard_cloze-v0-loglikelihood +1 -0
  49. scripts/yans/lm-evaluation-harness/tests/testdata/logiqa-v0-loglikelihood +1 -0
  50. scripts/yans/lm-evaluation-harness/tests/testdata/math_algebra-v0-greedy_until +1 -0
scripts/yans/lm-evaluation-harness/tests/testdata/anli_r1-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"anli_r1": {"acc": 0.334, "acc_stderr": 0.014922019523732967}}, "versions": {"anli_r1": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/arithmetic_1dc-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"arithmetic_1dc": {"acc": 0.0, "acc_stderr": 0.0}}, "versions": {"arithmetic_1dc": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_animate_subject_trans-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_animate_subject_trans": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_animate_subject_trans": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_determiner_noun_agreement_2-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_determiner_noun_agreement_2": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_determiner_noun_agreement_2": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_determiner_noun_agreement_irregular_1-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_determiner_noun_agreement_irregular_1": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_determiner_noun_agreement_irregular_1": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_determiner_noun_agreement_irregular_2-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ ddb24ddfaebe076b3aa7107937d71bf5f4503a78283bc889e39200368603681e
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_determiner_noun_agreement_irregular_2-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_determiner_noun_agreement_irregular_2": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_determiner_noun_agreement_irregular_2": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_drop_argument-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_drop_argument": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_drop_argument": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_existential_there_quantifiers_2-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_existential_there_quantifiers_2": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_existential_there_quantifiers_2": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_left_branch_island_simple_question-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ 6cb36bbdae7754f8832f50872c3dd511ce12547e00fa0771deb747be3355eb85
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_npi_present_2-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ fdb688ac6259bb65d234ef0a36e9a9ee449f9608f633b12e1943b462aead8e17
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_principle_A_case_1-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ 49d2b8ce6667a6166fdc2a2e5dbe7ff07d9b8415e9f33482aef15956b3ebc24a
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_principle_A_case_2-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ cd68adb65c891d672e22bf53c054b2083ab08bc1da43951732b409c942d14bc7
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_principle_A_domain_1-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_principle_A_domain_1": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_principle_A_domain_1": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_principle_A_domain_3-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_principle_A_domain_3": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_principle_A_domain_3": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_superlative_quantifiers_1-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_superlative_quantifiers_1": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_superlative_quantifiers_1": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_tough_vs_raising_1-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_tough_vs_raising_1": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_tough_vs_raising_1": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_wh_questions_subject_gap_long_distance-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ 37483dfda688b62ad27161c9fc1e1e7710c5a6e6a7cd3474df119bcafd30e97f
scripts/yans/lm-evaluation-harness/tests/testdata/blimp_wh_vs_that_with_gap-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"blimp_wh_vs_that_with_gap": {"acc": 0.485, "acc_stderr": 0.0158121796418149}}, "versions": {"blimp_wh_vs_that_with_gap": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/cb-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ ec3b1bbb9561e39c43c6f77a23b4060b15c606141c5346e3d0791b3e92aaa5d0
scripts/yans/lm-evaluation-harness/tests/testdata/cola-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"cola": {"mcc": -0.04538802810223175, "mcc_stderr": 0.023100371589225246}}, "versions": {"cola": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/coqa-v1-greedy_until ADDED
@@ -0,0 +1 @@
 
 
1
+ 57581470b921435d40da97872bb1cfda6ecf963ccc4b0240a3b04e3fea8c8e3a
scripts/yans/lm-evaluation-harness/tests/testdata/crows_pairs_english_age-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ de74d2ac7f926f2f486c045d84aae8f71711102f9d77b31f758fd148810d13d3
scripts/yans/lm-evaluation-harness/tests/testdata/crows_pairs_english_physical_appearance-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"crows_pairs_english_physical_appearance": {"likelihood_difference": 0.3221673223187262, "likelihood_difference_stderr": 0.026978346460100555, "pct_stereotype": 0.4027777777777778, "pct_stereotype_stderr": 0.05820650942569533}}, "versions": {"crows_pairs_english_physical_appearance": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/crows_pairs_english_sexual_orientation-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ e754a309296b157677dfba6e6feef983d1ce38dd0169ae726265621a7b573163
scripts/yans/lm-evaluation-harness/tests/testdata/crows_pairs_french_sexual_orientation-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ 2ce823fdb93d325aa8fb40db5d335b093b4b69792763532d940a752440ee3a76
scripts/yans/lm-evaluation-harness/tests/testdata/cycle_letters-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"cycle_letters": {"acc": 0.0, "acc_stderr": 0.0}}, "versions": {"cycle_letters": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/ethics_cm-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ 92d136ebb2bd86cd036e61699ad9a1417dbb48651f0a3afa5045cf57cef5a3f6
scripts/yans/lm-evaluation-harness/tests/testdata/gguf_test_44e268d15decc4d2d0f99e57e1476269826cd3b54262f7a0981f75ddd45b25d0.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d672564bf8af93738531c26d02efda9ed844d1fdcdd619a06e37fce7ef6d8485
3
+ size 153
scripts/yans/lm-evaluation-harness/tests/testdata/gpt3_test_8025023377febbd8c5f2b9f26705c394ff375d0cad7c89c10fd9b8e1eb66ff1c.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:590805560ee790d530c075ad76633eb2e9749440083e0bab63489ff920fdfd33
3
+ size 70917
scripts/yans/lm-evaluation-harness/tests/testdata/headqa-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"headqa": {"acc": 0.23559445660102116, "acc_norm": 0.25018234865062, "acc_norm_stderr": 0.008272783230806014, "acc_stderr": 0.008105688874297972}}, "versions": {"headqa": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/headqa_es-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ 767ca34d9714edd9fb030ddbcc35a64e5180d1e247b0cb557fbb22fdf971ad1f
scripts/yans/lm-evaluation-harness/tests/testdata/headqa_es-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"headqa_es": {"acc": 0.23559445660102116, "acc_norm": 0.25018234865062, "acc_norm_stderr": 0.008272783230806014, "acc_stderr": 0.008105688874297972}}, "versions": {"headqa_es": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-abstract_algebra-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"hendrycksTest-abstract_algebra": {"acc": 0.32, "acc_norm": 0.34, "acc_norm_stderr": 0.04760952285695235, "acc_stderr": 0.04688261722621504}}, "versions": {"hendrycksTest-abstract_algebra": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-electrical_engineering-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ b9b5d8b8bb02696302ec6bc2a99bf987a5504d3bae0e529d2c8f263538c97518
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_biology-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ d4dc051f37a49dc75c218741e87bc826fd44f31ee1309b55e0f33bd191c1bc78
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_biology-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"hendrycksTest-high_school_biology": {"acc": 0.23870967741935484, "acc_norm": 0.2709677419354839, "acc_norm_stderr": 0.025284416114900152, "acc_stderr": 0.024251071262208834}}, "versions": {"hendrycksTest-high_school_biology": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_computer_science-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"hendrycksTest-high_school_computer_science": {"acc": 0.2, "acc_norm": 0.22, "acc_norm_stderr": 0.04163331998932269, "acc_stderr": 0.04020151261036845}}, "versions": {"hendrycksTest-high_school_computer_science": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_european_history-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"hendrycksTest-high_school_european_history": {"acc": 0.23636363636363636, "acc_norm": 0.24242424242424243, "acc_norm_stderr": 0.03346409881055953, "acc_stderr": 0.033175059300091805}}, "versions": {"hendrycksTest-high_school_european_history": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-high_school_macroeconomics-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ ce4faae2fb6628caa48f6fc74cbc848880db49e6ff51079392778a2322bcefef
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-moral_disputes-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ d6ef028022c02b69d1516973e08bebaa14d8debcf2589a2bb124823178202d20
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-nutrition-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ 19e49d218f55ed5ec4bd1a6cd3f3388c6f620b81484e7abe8b298e5481c3044d
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-prehistory-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ 6983c560a562749f4f702249a3a6ae51fa495acc0643a980bf2cf52c6c5d4b95
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-professional_law-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"hendrycksTest-professional_law": {"acc": 0.2561929595827901, "acc_norm": 0.2470664928292047, "acc_norm_stderr": 0.011015752255279352, "acc_stderr": 0.011149173153110582}}, "versions": {"hendrycksTest-professional_law": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/hendrycksTest-professional_medicine-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"hendrycksTest-professional_medicine": {"acc": 0.23161764705882354, "acc_norm": 0.2536764705882353, "acc_norm_stderr": 0.02643132987078953, "acc_stderr": 0.025626533803777562}}, "versions": {"hendrycksTest-professional_medicine": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/lambada_mt_de-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"lambada_mt_de": {"acc": 0.0, "acc_stderr": 0.0, "ppl": 1.6479047769869253, "ppl_stderr": 0.006497321146240192}}, "versions": {"lambada_mt_de": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/lambada_mt_fr-v0-res.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"results": {"lambada_mt_fr": {"acc": 0.0, "acc_stderr": 0.0, "ppl": 1.6479047769869253, "ppl_stderr": 0.006497321146240192}}, "versions": {"lambada_mt_fr": 0}}
scripts/yans/lm-evaluation-harness/tests/testdata/lambada_standard_cloze-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ b604f00bc9f2a77ef41f8cfdb5a8509b3ae9266893b9e90abc665f5399ecba4e
scripts/yans/lm-evaluation-harness/tests/testdata/logiqa-v0-loglikelihood ADDED
@@ -0,0 +1 @@
 
 
1
+ 12495c50454ba5e1ce0753bd18c09aaca516bebd27648d815e37b15229dbf198
scripts/yans/lm-evaluation-harness/tests/testdata/math_algebra-v0-greedy_until ADDED
@@ -0,0 +1 @@
 
 
1
+ f19182ce697a2c095d9e5b56ee6659dc38c93994b69ca75d7c3d3f5fd87572b4