Upload llm-eval-Qwen2.5-0.5B-Instruct-ifeval.json

llm-eval-Qwen2.5-0.5B-Instruct-ifeval.json

@@ -0,0 +1,207 @@
+{
+    "arc_challenge": {
+        "alias": "arc_challenge",
+        "acc,none": 0.30631399317406144,
+        "acc_stderr,none": 0.013470584417276513,
+        "acc_norm,none": 0.33447098976109213,
+        "acc_norm_stderr,none": 0.013787460322441382
+    },
+    "gpqa_diamond_cot_n_shot": {
+        "alias": "gpqa_diamond_cot_n_shot",
+        "exact_match,strict-match": 0.0,
+        "exact_match_stderr,strict-match": 0.0,
+        "exact_match,flexible-extract": 0.15151515151515152,
+        "exact_match_stderr,flexible-extract": 0.0255456504266036
+    },
+    "gpqa_diamond_cot_zeroshot": {
+        "alias": "gpqa_diamond_cot_zeroshot",
+        "exact_match,strict-match": 0.0,
+        "exact_match_stderr,strict-match": 0.0,
+        "exact_match,flexible-extract": 0.12626262626262627,
+        "exact_match_stderr,flexible-extract": 0.02366435940288024
+    },
+    "gpqa_diamond_generative_n_shot": {
+        "alias": "gpqa_diamond_generative_n_shot",
+        "exact_match,strict-match": 0.005050505050505051,
+        "exact_match_stderr,strict-match": 0.0050505050505050535,
+        "exact_match,flexible-extract": 0.18686868686868688,
+        "exact_match_stderr,flexible-extract": 0.02777253333421898
+    },
+    "gpqa_diamond_n_shot": {
+        "alias": "gpqa_diamond_n_shot",
+        "acc,none": 0.23232323232323232,
+        "acc_stderr,none": 0.030088629490217483,
+        "acc_norm,none": 0.23232323232323232,
+        "acc_norm_stderr,none": 0.030088629490217483
+    },
+    "gpqa_diamond_zeroshot": {
+        "alias": "gpqa_diamond_zeroshot",
+        "acc,none": 0.2474747474747475,
+        "acc_stderr,none": 0.03074630074212451,
+        "acc_norm,none": 0.2474747474747475,
+        "acc_norm_stderr,none": 0.03074630074212451
+    },
+    "gpqa_extended_cot_n_shot": {
+        "alias": "gpqa_extended_cot_n_shot",
+        "exact_match,strict-match": 0.0,
+        "exact_match_stderr,strict-match": 0.0,
+        "exact_match,flexible-extract": 0.15934065934065933,
+        "exact_match_stderr,flexible-extract": 0.015677437397173054
+    },
+    "gpqa_extended_cot_zeroshot": {
+        "alias": "gpqa_extended_cot_zeroshot",
+        "exact_match,strict-match": 0.0018315018315018315,
+        "exact_match_stderr,strict-match": 0.0018315018315018447,
+        "exact_match,flexible-extract": 0.1336996336996337,
+        "exact_match_stderr,flexible-extract": 0.014578106095655245
+    },
+    "gpqa_extended_generative_n_shot": {
+        "alias": "gpqa_extended_generative_n_shot",
+        "exact_match,strict-match": 0.0,
+        "exact_match_stderr,strict-match": 0.0,
+        "exact_match,flexible-extract": 0.20146520146520147,
+        "exact_match_stderr,flexible-extract": 0.017181010109243344
+    },
+    "gpqa_extended_n_shot": {
+        "alias": "gpqa_extended_n_shot",
+        "acc,none": 0.2600732600732601,
+        "acc_stderr,none": 0.018790743352015988,
+        "acc_norm,none": 0.2600732600732601,
+        "acc_norm_stderr,none": 0.018790743352015988
+    },
+    "gpqa_extended_zeroshot": {
+        "alias": "gpqa_extended_zeroshot",
+        "acc,none": 0.26373626373626374,
+        "acc_stderr,none": 0.018875713580372485,
+        "acc_norm,none": 0.26373626373626374,
+        "acc_norm_stderr,none": 0.018875713580372485
+    },
+    "gpqa_main_cot_n_shot": {
+        "alias": "gpqa_main_cot_n_shot",
+        "exact_match,strict-match": 0.0,
+        "exact_match_stderr,strict-match": 0.0,
+        "exact_match,flexible-extract": 0.13616071428571427,
+        "exact_match_stderr,flexible-extract": 0.016221410863569787
+    },
+    "gpqa_main_cot_zeroshot": {
+        "alias": "gpqa_main_cot_zeroshot",
+        "exact_match,strict-match": 0.0,
+        "exact_match_stderr,strict-match": 0.0,
+        "exact_match,flexible-extract": 0.15625,
+        "exact_match_stderr,flexible-extract": 0.017173671221421365
+    },
+    "gpqa_main_generative_n_shot": {
+        "alias": "gpqa_main_generative_n_shot",
+        "exact_match,strict-match": 0.0,
+        "exact_match_stderr,strict-match": 0.0,
+        "exact_match,flexible-extract": 0.19642857142857142,
+        "exact_match_stderr,flexible-extract": 0.018791472419524228
+    },
+    "gpqa_main_n_shot": {
+        "alias": "gpqa_main_n_shot",
+        "acc,none": 0.2700892857142857,
+        "acc_stderr,none": 0.021000749078822385,
+        "acc_norm,none": 0.2700892857142857,
+        "acc_norm_stderr,none": 0.021000749078822385
+    },
+    "gpqa_main_zeroshot": {
+        "alias": "gpqa_main_zeroshot",
+        "acc,none": 0.30580357142857145,
+        "acc_stderr,none": 0.021792582688756983,
+        "acc_norm,none": 0.30580357142857145,
+        "acc_norm_stderr,none": 0.021792582688756983
+    },
+    "hellaswag": {
+        "alias": "hellaswag",
+        "acc,none": 0.4047002589125672,
+        "acc_stderr,none": 0.004898308167211847,
+        "acc_norm,none": 0.5237004580760805,
+        "acc_norm_stderr,none": 0.004984172621822888
+    },
+    "ifeval": {
+        "alias": "ifeval",
+        "prompt_level_strict_acc,none": 0.20147874306839186,
+        "prompt_level_strict_acc_stderr,none": 0.017260802262371488,
+        "inst_level_strict_acc,none": 0.3405275779376499,
+        "inst_level_strict_acc_stderr,none": "N/A",
+        "prompt_level_loose_acc,none": 0.2365988909426987,
+        "prompt_level_loose_acc_stderr,none": 0.018288827582625643,
+        "inst_level_loose_acc,none": 0.3752997601918465,
+        "inst_level_loose_acc_stderr,none": "N/A"
+    },
+    "mmlu_pro": {
+        "exact_match,custom-extract": 0.14037566489361702,
+        "exact_match_stderr,custom-extract": 0.003117505733576593,
+        "alias": "mmlu_pro"
+    },
+    "mmlu_pro_biology": {
+        "alias": " - biology",
+        "exact_match,custom-extract": 0.25662482566248257,
+        "exact_match_stderr,custom-extract": 0.016322882305354162
+    },
+    "mmlu_pro_business": {
+        "alias": " - business",
+        "exact_match,custom-extract": 0.1267427122940431,
+        "exact_match_stderr,custom-extract": 0.011851395705593072
+    },
+    "mmlu_pro_chemistry": {
+        "alias": " - chemistry",
+        "exact_match,custom-extract": 0.06537102473498234,
+        "exact_match_stderr,custom-extract": 0.00734989211563516
+    },
+    "mmlu_pro_computer_science": {
+        "alias": " - computer_science",
+        "exact_match,custom-extract": 0.14878048780487804,
+        "exact_match_stderr,custom-extract": 0.017596736073033845
+    },
+    "mmlu_pro_economics": {
+        "alias": " - economics",
+        "exact_match,custom-extract": 0.245260663507109,
+        "exact_match_stderr,custom-extract": 0.01481830928170158
+    },
+    "mmlu_pro_engineering": {
+        "alias": " - engineering",
+        "exact_match,custom-extract": 0.058823529411764705,
+        "exact_match_stderr,custom-extract": 0.007562639370979075
+    },
+    "mmlu_pro_health": {
+        "alias": " - health",
+        "exact_match,custom-extract": 0.13447432762836187,
+        "exact_match_stderr,custom-extract": 0.011935720476846831
+    },
+    "mmlu_pro_history": {
+        "alias": " - history",
+        "exact_match,custom-extract": 0.14698162729658792,
+        "exact_match_stderr,custom-extract": 0.018164310621441037
+    },
+    "mmlu_pro_law": {
+        "alias": " - law",
+        "exact_match,custom-extract": 0.13533151680290645,
+        "exact_match_stderr,custom-extract": 0.010314019468785332
+    },
+    "mmlu_pro_math": {
+        "alias": " - math",
+        "exact_match,custom-extract": 0.12731310140636565,
+        "exact_match_stderr,custom-extract": 0.009071913335559057
+    },
+    "mmlu_pro_other": {
+        "alias": " - other",
+        "exact_match,custom-extract": 0.15151515151515152,
+        "exact_match_stderr,custom-extract": 0.011801826546500667
+    },
+    "mmlu_pro_philosophy": {
+        "alias": " - philosophy",
+        "exact_match,custom-extract": 0.12625250501002003,
+        "exact_match_stderr,custom-extract": 0.014883268009546964
+    },
+    "mmlu_pro_physics": {
+        "alias": " - physics",
+        "exact_match,custom-extract": 0.0869899923017706,
+        "exact_match_stderr,custom-extract": 0.007822310824931376
+    },
+    "mmlu_pro_psychology": {
+        "alias": " - psychology",
+        "exact_match,custom-extract": 0.2543859649122807,
+        "exact_match_stderr,custom-extract": 0.0154267502913602
+    }
+}