Upload eval_results/AI-MO/deepseek-math-7b-sft/aimo_v03.00/aimo_kaggle/results_2024-05-07T12-15-37.241134.json with huggingface_hub
Browse files
eval_results/AI-MO/deepseek-math-7b-sft/aimo_v03.00/aimo_kaggle/results_2024-05-07T12-15-37.241134.json
ADDED
@@ -0,0 +1,144 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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": 437176.178939583,
|
9 |
+
"end_time": 437485.420318817,
|
10 |
+
"total_evaluation_time_secondes": "309.2413792340085",
|
11 |
+
"model_name": "AI-MO/deepseek-math-7b-sft",
|
12 |
+
"model_sha": "f500a523198914c8fa81f759aed450551064ab3a",
|
13 |
+
"model_dtype": "torch.bfloat16",
|
14 |
+
"model_size": "12.93 GB",
|
15 |
+
"config": null
|
16 |
+
},
|
17 |
+
"results": {
|
18 |
+
"custom|aimo_kaggle:train|0": {
|
19 |
+
"qem": 0.0,
|
20 |
+
"qem_stderr": 0.0
|
21 |
+
},
|
22 |
+
"custom|aimo_kaggle:valid|0": {
|
23 |
+
"qem": 0.28,
|
24 |
+
"qem_stderr": 0.06414269805898185
|
25 |
+
},
|
26 |
+
"custom|aimo_kaggle:_average|0": {
|
27 |
+
"qem": 0.14,
|
28 |
+
"qem_stderr": 0.03207134902949092
|
29 |
+
},
|
30 |
+
"all": {
|
31 |
+
"qem": 0.14,
|
32 |
+
"qem_stderr": 0.03207134902949092
|
33 |
+
}
|
34 |
+
},
|
35 |
+
"versions": {
|
36 |
+
"custom|aimo_kaggle:train|0": 0,
|
37 |
+
"custom|aimo_kaggle:valid|0": 0
|
38 |
+
},
|
39 |
+
"config_tasks": {
|
40 |
+
"custom|aimo_kaggle:train": {
|
41 |
+
"name": "aimo_kaggle:train",
|
42 |
+
"prompt_function": "kaggle_prompt_fn",
|
43 |
+
"hf_repo": "AI-MO/kaggle-train-set",
|
44 |
+
"hf_subset": "",
|
45 |
+
"metric": [
|
46 |
+
"quasi_exact_match_math"
|
47 |
+
],
|
48 |
+
"hf_avail_splits": [
|
49 |
+
"train"
|
50 |
+
],
|
51 |
+
"evaluation_splits": [
|
52 |
+
"train"
|
53 |
+
],
|
54 |
+
"few_shots_split": null,
|
55 |
+
"few_shots_select": null,
|
56 |
+
"generation_size": 2048,
|
57 |
+
"stop_sequence": null,
|
58 |
+
"output_regex": null,
|
59 |
+
"num_samples": null,
|
60 |
+
"frozen": false,
|
61 |
+
"suite": [
|
62 |
+
"custom"
|
63 |
+
],
|
64 |
+
"original_num_docs": 10,
|
65 |
+
"effective_num_docs": 10,
|
66 |
+
"trust_dataset": null,
|
67 |
+
"must_remove_duplicate_docs": null,
|
68 |
+
"version": 0
|
69 |
+
},
|
70 |
+
"custom|aimo_kaggle:valid": {
|
71 |
+
"name": "aimo_kaggle:valid",
|
72 |
+
"prompt_function": "kaggle_prompt_fn",
|
73 |
+
"hf_repo": "AI-MO/kaggle-validation-set",
|
74 |
+
"hf_subset": "v0",
|
75 |
+
"metric": [
|
76 |
+
"quasi_exact_match_math"
|
77 |
+
],
|
78 |
+
"hf_avail_splits": [
|
79 |
+
"train"
|
80 |
+
],
|
81 |
+
"evaluation_splits": [
|
82 |
+
"train"
|
83 |
+
],
|
84 |
+
"few_shots_split": null,
|
85 |
+
"few_shots_select": null,
|
86 |
+
"generation_size": 2048,
|
87 |
+
"stop_sequence": null,
|
88 |
+
"output_regex": null,
|
89 |
+
"num_samples": null,
|
90 |
+
"frozen": false,
|
91 |
+
"suite": [
|
92 |
+
"custom"
|
93 |
+
],
|
94 |
+
"original_num_docs": 50,
|
95 |
+
"effective_num_docs": 50,
|
96 |
+
"trust_dataset": null,
|
97 |
+
"must_remove_duplicate_docs": null,
|
98 |
+
"version": 0
|
99 |
+
}
|
100 |
+
},
|
101 |
+
"summary_tasks": {
|
102 |
+
"custom|aimo_kaggle:train|0": {
|
103 |
+
"hashes": {
|
104 |
+
"hash_examples": "3ef01feac22949ce",
|
105 |
+
"hash_full_prompts": "74c9baf89033f76a",
|
106 |
+
"hash_input_tokens": "2f2b429412d88494",
|
107 |
+
"hash_cont_tokens": "28f1d3fdc9ea3afd"
|
108 |
+
},
|
109 |
+
"truncated": 10,
|
110 |
+
"non_truncated": 0,
|
111 |
+
"padded": 6,
|
112 |
+
"non_padded": 4,
|
113 |
+
"effective_few_shots": 0.0,
|
114 |
+
"num_truncated_few_shots": 0
|
115 |
+
},
|
116 |
+
"custom|aimo_kaggle:valid|0": {
|
117 |
+
"hashes": {
|
118 |
+
"hash_examples": "2a919e6b839e921a",
|
119 |
+
"hash_full_prompts": "e0424ade9e31a0fe",
|
120 |
+
"hash_input_tokens": "7da1857b6831dd82",
|
121 |
+
"hash_cont_tokens": "9139787617b042e2"
|
122 |
+
},
|
123 |
+
"truncated": 48,
|
124 |
+
"non_truncated": 2,
|
125 |
+
"padded": 34,
|
126 |
+
"non_padded": 16,
|
127 |
+
"effective_few_shots": 0.0,
|
128 |
+
"num_truncated_few_shots": 0
|
129 |
+
}
|
130 |
+
},
|
131 |
+
"summary_general": {
|
132 |
+
"hashes": {
|
133 |
+
"hash_examples": "42ad458c3603ccb6",
|
134 |
+
"hash_full_prompts": "51040dc54a721b0c",
|
135 |
+
"hash_input_tokens": "1337f29495dfd28d",
|
136 |
+
"hash_cont_tokens": "a4f1caa708b78677"
|
137 |
+
},
|
138 |
+
"truncated": 58,
|
139 |
+
"non_truncated": 2,
|
140 |
+
"padded": 40,
|
141 |
+
"non_padded": 20,
|
142 |
+
"num_truncated_few_shots": 0
|
143 |
+
}
|
144 |
+
}
|