Upload eval_results/AI-MO/Eurus-7b-sft/aimo_v01.00/aimo_kaggle_cot/results_2024-04-29T13-03-30.759017.json with huggingface_hub
Browse files
eval_results/AI-MO/Eurus-7b-sft/aimo_v01.00/aimo_kaggle_cot/results_2024-04-29T13-03-30.759017.json
ADDED
|
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config_general": {
|
| 3 |
+
"lighteval_sha": "?",
|
| 4 |
+
"num_fewshot_seeds": 1,
|
| 5 |
+
"override_batch_size": 4,
|
| 6 |
+
"max_samples": null,
|
| 7 |
+
"job_id": "",
|
| 8 |
+
"start_time": 4218057.305847882,
|
| 9 |
+
"end_time": 4218311.447086045,
|
| 10 |
+
"total_evaluation_time_secondes": "254.14123816229403",
|
| 11 |
+
"model_name": "AI-MO/Eurus-7b-sft",
|
| 12 |
+
"model_sha": "c85aa9511e4c7a9f5f061d5d8e2016536482f006",
|
| 13 |
+
"model_dtype": "torch.bfloat16",
|
| 14 |
+
"model_size": "13.99 GB",
|
| 15 |
+
"config": null
|
| 16 |
+
},
|
| 17 |
+
"results": {
|
| 18 |
+
"custom|aimo_kaggle_cot:v0|0": {
|
| 19 |
+
"qem": 0.02,
|
| 20 |
+
"qem_stderr": 0.01999999999999998
|
| 21 |
+
},
|
| 22 |
+
"all": {
|
| 23 |
+
"qem": 0.02,
|
| 24 |
+
"qem_stderr": 0.01999999999999998
|
| 25 |
+
}
|
| 26 |
+
},
|
| 27 |
+
"versions": {
|
| 28 |
+
"custom|aimo_kaggle_cot:v0|0": 0
|
| 29 |
+
},
|
| 30 |
+
"config_tasks": {
|
| 31 |
+
"custom|aimo_kaggle_cot:v0": {
|
| 32 |
+
"name": "aimo_kaggle_cot:v0",
|
| 33 |
+
"prompt_function": "kaggle_prompt_fn",
|
| 34 |
+
"hf_repo": "AI-MO/kaggle-validation-set",
|
| 35 |
+
"hf_subset": "v0",
|
| 36 |
+
"metric": [
|
| 37 |
+
"quasi_exact_match_math"
|
| 38 |
+
],
|
| 39 |
+
"hf_avail_splits": [
|
| 40 |
+
"train"
|
| 41 |
+
],
|
| 42 |
+
"evaluation_splits": [
|
| 43 |
+
"train"
|
| 44 |
+
],
|
| 45 |
+
"few_shots_split": null,
|
| 46 |
+
"few_shots_select": null,
|
| 47 |
+
"generation_size": 2048,
|
| 48 |
+
"stop_sequence": null,
|
| 49 |
+
"output_regex": null,
|
| 50 |
+
"frozen": false,
|
| 51 |
+
"suite": [
|
| 52 |
+
"custom"
|
| 53 |
+
],
|
| 54 |
+
"original_num_docs": 50,
|
| 55 |
+
"effective_num_docs": 50,
|
| 56 |
+
"trust_dataset": null,
|
| 57 |
+
"must_remove_duplicate_docs": null
|
| 58 |
+
}
|
| 59 |
+
},
|
| 60 |
+
"summary_tasks": {
|
| 61 |
+
"custom|aimo_kaggle_cot:v0|0": {
|
| 62 |
+
"hashes": {
|
| 63 |
+
"hash_examples": "1bced43e11f6aee7",
|
| 64 |
+
"hash_full_prompts": "c4259c8a11b3c571",
|
| 65 |
+
"hash_input_tokens": "612b5f61d266e489",
|
| 66 |
+
"hash_cont_tokens": "58767869f663f9bc"
|
| 67 |
+
},
|
| 68 |
+
"truncated": 47,
|
| 69 |
+
"non_truncated": 3,
|
| 70 |
+
"padded": 33,
|
| 71 |
+
"non_padded": 17,
|
| 72 |
+
"effective_few_shots": 0.0,
|
| 73 |
+
"num_truncated_few_shots": 0
|
| 74 |
+
}
|
| 75 |
+
},
|
| 76 |
+
"summary_general": {
|
| 77 |
+
"hashes": {
|
| 78 |
+
"hash_examples": "0f7a1315b35f1242",
|
| 79 |
+
"hash_full_prompts": "2fcf79596cbf555f",
|
| 80 |
+
"hash_input_tokens": "b4608c7b72e1a1e0",
|
| 81 |
+
"hash_cont_tokens": "b944e0307fc34c56"
|
| 82 |
+
},
|
| 83 |
+
"truncated": 47,
|
| 84 |
+
"non_truncated": 3,
|
| 85 |
+
"padded": 33,
|
| 86 |
+
"non_padded": 17,
|
| 87 |
+
"num_truncated_few_shots": 0
|
| 88 |
+
}
|
| 89 |
+
}
|