Upload eval_results/edbeeching/mixtral-8x7b-instruct-v0.1_merged/v0.2/arc/results_2024-04-14T13-11-24.629818.json with huggingface_hub
Browse files
eval_results/edbeeching/mixtral-8x7b-instruct-v0.1_merged/v0.2/arc/results_2024-04-14T13-11-24.629818.json
ADDED
|
@@ -0,0 +1,98 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config_general": {
|
| 3 |
+
"lighteval_sha": "?",
|
| 4 |
+
"num_fewshot_seeds": 1,
|
| 5 |
+
"override_batch_size": 1,
|
| 6 |
+
"max_samples": null,
|
| 7 |
+
"job_id": "",
|
| 8 |
+
"start_time": 3249169.019665027,
|
| 9 |
+
"end_time": 3252105.538397305,
|
| 10 |
+
"total_evaluation_time_secondes": "2936.518732278142",
|
| 11 |
+
"model_name": "edbeeching/mixtral-8x7b-instruct-v0.1_merged",
|
| 12 |
+
"model_sha": "02e5e37da870d6382b35fd16e9545bbab51dd218",
|
| 13 |
+
"model_dtype": "torch.bfloat16",
|
| 14 |
+
"model_size": "87.49 GB",
|
| 15 |
+
"config": null
|
| 16 |
+
},
|
| 17 |
+
"results": {
|
| 18 |
+
"leaderboard|arc:challenge|25": {
|
| 19 |
+
"acc": 0.5068259385665529,
|
| 20 |
+
"acc_stderr": 0.014610029151379813,
|
| 21 |
+
"acc_norm": 0.5503412969283277,
|
| 22 |
+
"acc_norm_stderr": 0.014537144444284748
|
| 23 |
+
},
|
| 24 |
+
"all": {
|
| 25 |
+
"acc": 0.5068259385665529,
|
| 26 |
+
"acc_stderr": 0.014610029151379813,
|
| 27 |
+
"acc_norm": 0.5503412969283277,
|
| 28 |
+
"acc_norm_stderr": 0.014537144444284748
|
| 29 |
+
}
|
| 30 |
+
},
|
| 31 |
+
"versions": {
|
| 32 |
+
"leaderboard|arc:challenge|25": 0
|
| 33 |
+
},
|
| 34 |
+
"config_tasks": {
|
| 35 |
+
"leaderboard|arc:challenge": {
|
| 36 |
+
"name": "arc:challenge",
|
| 37 |
+
"prompt_function": "arc",
|
| 38 |
+
"hf_repo": "ai2_arc",
|
| 39 |
+
"hf_subset": "ARC-Challenge",
|
| 40 |
+
"metric": [
|
| 41 |
+
"loglikelihood_acc",
|
| 42 |
+
"loglikelihood_acc_norm_nospace"
|
| 43 |
+
],
|
| 44 |
+
"hf_avail_splits": [
|
| 45 |
+
"train",
|
| 46 |
+
"test"
|
| 47 |
+
],
|
| 48 |
+
"evaluation_splits": [
|
| 49 |
+
"test"
|
| 50 |
+
],
|
| 51 |
+
"few_shots_split": null,
|
| 52 |
+
"few_shots_select": "random_sampling_from_train",
|
| 53 |
+
"generation_size": 1,
|
| 54 |
+
"stop_sequence": [
|
| 55 |
+
"\n"
|
| 56 |
+
],
|
| 57 |
+
"output_regex": null,
|
| 58 |
+
"frozen": false,
|
| 59 |
+
"suite": [
|
| 60 |
+
"leaderboard",
|
| 61 |
+
"arc"
|
| 62 |
+
],
|
| 63 |
+
"original_num_docs": 1172,
|
| 64 |
+
"effective_num_docs": 1172,
|
| 65 |
+
"trust_dataset": true,
|
| 66 |
+
"must_remove_duplicate_docs": null
|
| 67 |
+
}
|
| 68 |
+
},
|
| 69 |
+
"summary_tasks": {
|
| 70 |
+
"leaderboard|arc:challenge|25": {
|
| 71 |
+
"hashes": {
|
| 72 |
+
"hash_examples": "17b0cae357c0259e",
|
| 73 |
+
"hash_full_prompts": "d44b17600302e2b8",
|
| 74 |
+
"hash_input_tokens": "0149ae15d3f432fa",
|
| 75 |
+
"hash_cont_tokens": "289aa98c400841d8"
|
| 76 |
+
},
|
| 77 |
+
"truncated": 0,
|
| 78 |
+
"non_truncated": 1172,
|
| 79 |
+
"padded": 4660,
|
| 80 |
+
"non_padded": 27,
|
| 81 |
+
"effective_few_shots": 25.0,
|
| 82 |
+
"num_truncated_few_shots": 0
|
| 83 |
+
}
|
| 84 |
+
},
|
| 85 |
+
"summary_general": {
|
| 86 |
+
"hashes": {
|
| 87 |
+
"hash_examples": "aaa6929c6d3771fb",
|
| 88 |
+
"hash_full_prompts": "b29ea0f4ca87b218",
|
| 89 |
+
"hash_input_tokens": "6064dbfd3e9d9140",
|
| 90 |
+
"hash_cont_tokens": "a2928b768b852bf8"
|
| 91 |
+
},
|
| 92 |
+
"truncated": 0,
|
| 93 |
+
"non_truncated": 1172,
|
| 94 |
+
"padded": 4660,
|
| 95 |
+
"non_padded": 27,
|
| 96 |
+
"num_truncated_few_shots": 0
|
| 97 |
+
}
|
| 98 |
+
}
|