Upload eval_results/luckeciano/Qwen-2.5-7B-RL-AC-BigLRv3-Fast-4-v3-AdamEps6/36ec96284216458fc05f39ffe9cd8816f474938f/eval_llm/results_2025-04-18T20-21-47.947931.json with huggingface_hub
Browse files
eval_results/luckeciano/Qwen-2.5-7B-RL-AC-BigLRv3-Fast-4-v3-AdamEps6/36ec96284216458fc05f39ffe9cd8816f474938f/eval_llm/results_2025-04-18T20-21-47.947931.json
ADDED
|
@@ -0,0 +1,212 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config_general": {
|
| 3 |
+
"lighteval_sha": "?",
|
| 4 |
+
"num_fewshot_seeds": 1,
|
| 5 |
+
"override_batch_size": -1,
|
| 6 |
+
"max_samples": null,
|
| 7 |
+
"job_id": 0,
|
| 8 |
+
"start_time": 30312.768263809,
|
| 9 |
+
"end_time": 30697.814470214,
|
| 10 |
+
"total_evaluation_time_secondes": "385.0462064050007",
|
| 11 |
+
"model_name": "luckeciano/Qwen-2.5-7B-RL-AC-BigLRv3-Fast-4-v3-AdamEps6",
|
| 12 |
+
"model_sha": "",
|
| 13 |
+
"model_dtype": null,
|
| 14 |
+
"model_size": null
|
| 15 |
+
},
|
| 16 |
+
"results": {
|
| 17 |
+
"custom|aime24|0": {
|
| 18 |
+
"extractive_match": 0.13333333333333333,
|
| 19 |
+
"extractive_match_stderr": 0.06312427686319992
|
| 20 |
+
},
|
| 21 |
+
"custom|gpqa:diamond|0": {
|
| 22 |
+
"extractive_match": 0.3282828282828283,
|
| 23 |
+
"extractive_match_stderr": 0.03345678422756777
|
| 24 |
+
},
|
| 25 |
+
"custom|math_500|0": {
|
| 26 |
+
"extractive_match": 0.638,
|
| 27 |
+
"extractive_match_stderr": 0.0215136625275824
|
| 28 |
+
},
|
| 29 |
+
"all": {
|
| 30 |
+
"extractive_match": 0.3665387205387205,
|
| 31 |
+
"extractive_match_stderr": 0.039364907872783364
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"versions": {
|
| 35 |
+
"custom|aime24|0": 1,
|
| 36 |
+
"custom|gpqa:diamond|0": 1,
|
| 37 |
+
"custom|math_500|0": 1
|
| 38 |
+
},
|
| 39 |
+
"config_tasks": {
|
| 40 |
+
"custom|aime24": {
|
| 41 |
+
"name": "aime24",
|
| 42 |
+
"prompt_function": "aime_prompt_fn",
|
| 43 |
+
"hf_repo": "HuggingFaceH4/aime_2024",
|
| 44 |
+
"hf_subset": "default",
|
| 45 |
+
"metric": [
|
| 46 |
+
{
|
| 47 |
+
"metric_name": "extractive_match",
|
| 48 |
+
"higher_is_better": true,
|
| 49 |
+
"category": "3",
|
| 50 |
+
"use_case": "1",
|
| 51 |
+
"sample_level_fn": "sample_level_fn",
|
| 52 |
+
"corpus_level_fn": "mean"
|
| 53 |
+
}
|
| 54 |
+
],
|
| 55 |
+
"hf_revision": null,
|
| 56 |
+
"hf_filter": null,
|
| 57 |
+
"hf_avail_splits": [
|
| 58 |
+
"train"
|
| 59 |
+
],
|
| 60 |
+
"trust_dataset": false,
|
| 61 |
+
"evaluation_splits": [
|
| 62 |
+
"train"
|
| 63 |
+
],
|
| 64 |
+
"few_shots_split": null,
|
| 65 |
+
"few_shots_select": null,
|
| 66 |
+
"generation_size": 32768,
|
| 67 |
+
"generation_grammar": null,
|
| 68 |
+
"stop_sequence": [],
|
| 69 |
+
"num_samples": null,
|
| 70 |
+
"suite": [
|
| 71 |
+
"custom"
|
| 72 |
+
],
|
| 73 |
+
"original_num_docs": 30,
|
| 74 |
+
"effective_num_docs": 30,
|
| 75 |
+
"must_remove_duplicate_docs": false,
|
| 76 |
+
"version": 1
|
| 77 |
+
},
|
| 78 |
+
"custom|gpqa:diamond": {
|
| 79 |
+
"name": "gpqa:diamond",
|
| 80 |
+
"prompt_function": "gpqa_prompt_fn",
|
| 81 |
+
"hf_repo": "Idavidrein/gpqa",
|
| 82 |
+
"hf_subset": "gpqa_diamond",
|
| 83 |
+
"metric": [
|
| 84 |
+
{
|
| 85 |
+
"metric_name": "extractive_match",
|
| 86 |
+
"higher_is_better": true,
|
| 87 |
+
"category": "3",
|
| 88 |
+
"use_case": "1",
|
| 89 |
+
"sample_level_fn": "sample_level_fn",
|
| 90 |
+
"corpus_level_fn": "mean"
|
| 91 |
+
}
|
| 92 |
+
],
|
| 93 |
+
"hf_revision": null,
|
| 94 |
+
"hf_filter": null,
|
| 95 |
+
"hf_avail_splits": [
|
| 96 |
+
"train"
|
| 97 |
+
],
|
| 98 |
+
"trust_dataset": true,
|
| 99 |
+
"evaluation_splits": [
|
| 100 |
+
"train"
|
| 101 |
+
],
|
| 102 |
+
"few_shots_split": null,
|
| 103 |
+
"few_shots_select": null,
|
| 104 |
+
"generation_size": 32768,
|
| 105 |
+
"generation_grammar": null,
|
| 106 |
+
"stop_sequence": [],
|
| 107 |
+
"num_samples": null,
|
| 108 |
+
"suite": [
|
| 109 |
+
"custom"
|
| 110 |
+
],
|
| 111 |
+
"original_num_docs": 198,
|
| 112 |
+
"effective_num_docs": 198,
|
| 113 |
+
"must_remove_duplicate_docs": false,
|
| 114 |
+
"version": 1
|
| 115 |
+
},
|
| 116 |
+
"custom|math_500": {
|
| 117 |
+
"name": "math_500",
|
| 118 |
+
"prompt_function": "math_prompt_fn",
|
| 119 |
+
"hf_repo": "HuggingFaceH4/MATH-500",
|
| 120 |
+
"hf_subset": "default",
|
| 121 |
+
"metric": [
|
| 122 |
+
{
|
| 123 |
+
"metric_name": "extractive_match",
|
| 124 |
+
"higher_is_better": true,
|
| 125 |
+
"category": "3",
|
| 126 |
+
"use_case": "1",
|
| 127 |
+
"sample_level_fn": "sample_level_fn",
|
| 128 |
+
"corpus_level_fn": "mean"
|
| 129 |
+
}
|
| 130 |
+
],
|
| 131 |
+
"hf_revision": null,
|
| 132 |
+
"hf_filter": null,
|
| 133 |
+
"hf_avail_splits": [
|
| 134 |
+
"test"
|
| 135 |
+
],
|
| 136 |
+
"trust_dataset": false,
|
| 137 |
+
"evaluation_splits": [
|
| 138 |
+
"test"
|
| 139 |
+
],
|
| 140 |
+
"few_shots_split": null,
|
| 141 |
+
"few_shots_select": null,
|
| 142 |
+
"generation_size": 32768,
|
| 143 |
+
"generation_grammar": null,
|
| 144 |
+
"stop_sequence": [],
|
| 145 |
+
"num_samples": null,
|
| 146 |
+
"suite": [
|
| 147 |
+
"custom"
|
| 148 |
+
],
|
| 149 |
+
"original_num_docs": 500,
|
| 150 |
+
"effective_num_docs": 500,
|
| 151 |
+
"must_remove_duplicate_docs": false,
|
| 152 |
+
"version": 1
|
| 153 |
+
}
|
| 154 |
+
},
|
| 155 |
+
"summary_tasks": {
|
| 156 |
+
"custom|aime24|0": {
|
| 157 |
+
"hashes": {
|
| 158 |
+
"hash_examples": "ddec8fc79d0a014b",
|
| 159 |
+
"hash_full_prompts": "fd2c067893d742e7",
|
| 160 |
+
"hash_input_tokens": "f6f823f35a4ceab6",
|
| 161 |
+
"hash_cont_tokens": "a0b02ca7d4a87327"
|
| 162 |
+
},
|
| 163 |
+
"truncated": 0,
|
| 164 |
+
"non_truncated": 30,
|
| 165 |
+
"padded": 0,
|
| 166 |
+
"non_padded": 30,
|
| 167 |
+
"effective_few_shots": 0.0,
|
| 168 |
+
"num_truncated_few_shots": 0
|
| 169 |
+
},
|
| 170 |
+
"custom|gpqa:diamond|0": {
|
| 171 |
+
"hashes": {
|
| 172 |
+
"hash_examples": "32a1ec97de8ab2eb",
|
| 173 |
+
"hash_full_prompts": "7ba629ab3e2cf72a",
|
| 174 |
+
"hash_input_tokens": "2690c3202b78e12c",
|
| 175 |
+
"hash_cont_tokens": "0d84ca13b98f57f8"
|
| 176 |
+
},
|
| 177 |
+
"truncated": 0,
|
| 178 |
+
"non_truncated": 198,
|
| 179 |
+
"padded": 0,
|
| 180 |
+
"non_padded": 198,
|
| 181 |
+
"effective_few_shots": 0.0,
|
| 182 |
+
"num_truncated_few_shots": 0
|
| 183 |
+
},
|
| 184 |
+
"custom|math_500|0": {
|
| 185 |
+
"hashes": {
|
| 186 |
+
"hash_examples": "adf0cc8311011db2",
|
| 187 |
+
"hash_full_prompts": "361f515b5e9f6586",
|
| 188 |
+
"hash_input_tokens": "a8fc04351e3cac24",
|
| 189 |
+
"hash_cont_tokens": "2745cb38e92388f9"
|
| 190 |
+
},
|
| 191 |
+
"truncated": 0,
|
| 192 |
+
"non_truncated": 500,
|
| 193 |
+
"padded": 0,
|
| 194 |
+
"non_padded": 500,
|
| 195 |
+
"effective_few_shots": 0.0,
|
| 196 |
+
"num_truncated_few_shots": 0
|
| 197 |
+
}
|
| 198 |
+
},
|
| 199 |
+
"summary_general": {
|
| 200 |
+
"hashes": {
|
| 201 |
+
"hash_examples": "bac52ac84e7a3754",
|
| 202 |
+
"hash_full_prompts": "c89b0364529a4332",
|
| 203 |
+
"hash_input_tokens": "5eb056e7deb4342b",
|
| 204 |
+
"hash_cont_tokens": "0b0db5e0bdaeff19"
|
| 205 |
+
},
|
| 206 |
+
"truncated": 0,
|
| 207 |
+
"non_truncated": 728,
|
| 208 |
+
"padded": 0,
|
| 209 |
+
"non_padded": 728,
|
| 210 |
+
"num_truncated_few_shots": 0
|
| 211 |
+
}
|
| 212 |
+
}
|