Upload eval_results/AI-MO/qwen-14b-sft/aimo_v00.03/aimo_kaggle_medium/results_2024-06-10T06-58-11.462911.json with huggingface_hub
Browse files
eval_results/AI-MO/qwen-14b-sft/aimo_v00.03/aimo_kaggle_medium/results_2024-06-10T06-58-11.462911.json
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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": 1363519.275138334,
|
| 9 |
+
"end_time": 1364403.119699325,
|
| 10 |
+
"total_evaluation_time_secondes": "883.8445609910414",
|
| 11 |
+
"model_name": "AI-MO/qwen-14b-sft",
|
| 12 |
+
"model_sha": "8a792f6fc445c55099998d2e2698fd8243732862",
|
| 13 |
+
"model_dtype": "torch.bfloat16",
|
| 14 |
+
"model_size": "26.39 GB",
|
| 15 |
+
"config": null
|
| 16 |
+
},
|
| 17 |
+
"results": {
|
| 18 |
+
"custom|aimo_kaggle_medium:v0|0": {
|
| 19 |
+
"qem": 0.175,
|
| 20 |
+
"qem_stderr": 0.060843430844447564
|
| 21 |
+
},
|
| 22 |
+
"all": {
|
| 23 |
+
"qem": 0.175,
|
| 24 |
+
"qem_stderr": 0.060843430844447564
|
| 25 |
+
}
|
| 26 |
+
},
|
| 27 |
+
"versions": {
|
| 28 |
+
"custom|aimo_kaggle_medium:v0|0": 0
|
| 29 |
+
},
|
| 30 |
+
"config_tasks": {
|
| 31 |
+
"custom|aimo_kaggle_medium:v0": {
|
| 32 |
+
"name": "aimo_kaggle_medium:v0",
|
| 33 |
+
"prompt_function": "kaggle_medium_prompt_fn",
|
| 34 |
+
"hf_repo": "AI-MO/kaggle-validation-set-medium",
|
| 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 |
+
"num_samples": null,
|
| 51 |
+
"frozen": false,
|
| 52 |
+
"suite": [
|
| 53 |
+
"custom"
|
| 54 |
+
],
|
| 55 |
+
"original_num_docs": 40,
|
| 56 |
+
"effective_num_docs": 40,
|
| 57 |
+
"trust_dataset": null,
|
| 58 |
+
"must_remove_duplicate_docs": null,
|
| 59 |
+
"version": 0
|
| 60 |
+
}
|
| 61 |
+
},
|
| 62 |
+
"summary_tasks": {
|
| 63 |
+
"custom|aimo_kaggle_medium:v0|0": {
|
| 64 |
+
"hashes": {
|
| 65 |
+
"hash_examples": "3401efda8b0cbcb5",
|
| 66 |
+
"hash_full_prompts": "14cb6646c78f810c",
|
| 67 |
+
"hash_input_tokens": "a1582bb5fbe8f005",
|
| 68 |
+
"hash_cont_tokens": "f16e32b9cc233319"
|
| 69 |
+
},
|
| 70 |
+
"truncated": 40,
|
| 71 |
+
"non_truncated": 0,
|
| 72 |
+
"padded": 32,
|
| 73 |
+
"non_padded": 8,
|
| 74 |
+
"effective_few_shots": 0.0,
|
| 75 |
+
"num_truncated_few_shots": 0
|
| 76 |
+
}
|
| 77 |
+
},
|
| 78 |
+
"summary_general": {
|
| 79 |
+
"hashes": {
|
| 80 |
+
"hash_examples": "4c81d27cfdb9d737",
|
| 81 |
+
"hash_full_prompts": "0d4787c840fc98b7",
|
| 82 |
+
"hash_input_tokens": "57c332fd31d33589",
|
| 83 |
+
"hash_cont_tokens": "986802f88896bcd2"
|
| 84 |
+
},
|
| 85 |
+
"truncated": 40,
|
| 86 |
+
"non_truncated": 0,
|
| 87 |
+
"padded": 32,
|
| 88 |
+
"non_padded": 8,
|
| 89 |
+
"num_truncated_few_shots": 0
|
| 90 |
+
}
|
| 91 |
+
}
|