Upload eval_results/AI-MO/deepseek-math-7b-sft/aimo_v01.43/aimo_kaggle/results_2024-04-26T00-15-00.009211.json with huggingface_hub
Browse files
eval_results/AI-MO/deepseek-math-7b-sft/aimo_v01.43/aimo_kaggle/results_2024-04-26T00-15-00.009211.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": 591317.777299186,
|
9 |
+
"end_time": 591592.202243397,
|
10 |
+
"total_evaluation_time_secondes": "274.4249442110304",
|
11 |
+
"model_name": "AI-MO/deepseek-math-7b-sft",
|
12 |
+
"model_sha": "7384467f01d7770ba37c1cad7a97ef45a9da00cb",
|
13 |
+
"model_dtype": "torch.bfloat16",
|
14 |
+
"model_size": "12.93 GB",
|
15 |
+
"config": null
|
16 |
+
},
|
17 |
+
"results": {
|
18 |
+
"custom|aimo_kaggle:v0|0": {
|
19 |
+
"qem": 0.14,
|
20 |
+
"qem_stderr": 0.049569575922564196
|
21 |
+
},
|
22 |
+
"all": {
|
23 |
+
"qem": 0.14,
|
24 |
+
"qem_stderr": 0.049569575922564196
|
25 |
+
}
|
26 |
+
},
|
27 |
+
"versions": {
|
28 |
+
"custom|aimo_kaggle:v0|0": 0
|
29 |
+
},
|
30 |
+
"config_tasks": {
|
31 |
+
"custom|aimo_kaggle:v0": {
|
32 |
+
"name": "aimo_kaggle: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:v0|0": {
|
62 |
+
"hashes": {
|
63 |
+
"hash_examples": "2a919e6b839e921a",
|
64 |
+
"hash_full_prompts": "e0424ade9e31a0fe",
|
65 |
+
"hash_input_tokens": "4002f9b277fa4d5e",
|
66 |
+
"hash_cont_tokens": "c2f4f0ed157e4a01"
|
67 |
+
},
|
68 |
+
"truncated": 47,
|
69 |
+
"non_truncated": 3,
|
70 |
+
"padded": 36,
|
71 |
+
"non_padded": 14,
|
72 |
+
"effective_few_shots": 0.0,
|
73 |
+
"num_truncated_few_shots": 0
|
74 |
+
}
|
75 |
+
},
|
76 |
+
"summary_general": {
|
77 |
+
"hashes": {
|
78 |
+
"hash_examples": "f59ec6c4e24c1226",
|
79 |
+
"hash_full_prompts": "35e1e25418ccb6f7",
|
80 |
+
"hash_input_tokens": "40119909fe745426",
|
81 |
+
"hash_cont_tokens": "0a20471ef0a14309"
|
82 |
+
},
|
83 |
+
"truncated": 47,
|
84 |
+
"non_truncated": 3,
|
85 |
+
"padded": 36,
|
86 |
+
"non_padded": 14,
|
87 |
+
"num_truncated_few_shots": 0
|
88 |
+
}
|
89 |
+
}
|