Upload eval_results/abhishek/autotrain-llama3-70b-orpo-v1/main/gsm8k/results_2024-05-03T08-32-22.135929.json with huggingface_hub
Browse files
eval_results/abhishek/autotrain-llama3-70b-orpo-v1/main/gsm8k/results_2024-05-03T08-32-22.135929.json
ADDED
@@ -0,0 +1,92 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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": 4834263.36921878,
|
9 |
+
"end_time": 4840096.580476612,
|
10 |
+
"total_evaluation_time_secondes": "5833.211257832125",
|
11 |
+
"model_name": "abhishek/autotrain-llama3-70b-orpo-v1",
|
12 |
+
"model_sha": "053236c6846cc561c1503ba05e2b28c94855a432",
|
13 |
+
"model_dtype": "torch.float16",
|
14 |
+
"model_size": "131.73 GB",
|
15 |
+
"config": null
|
16 |
+
},
|
17 |
+
"results": {
|
18 |
+
"lighteval|gsm8k|5": {
|
19 |
+
"qem": 0.8893100833965125,
|
20 |
+
"qem_stderr": 0.00864217255139248
|
21 |
+
},
|
22 |
+
"all": {
|
23 |
+
"qem": 0.8893100833965125,
|
24 |
+
"qem_stderr": 0.00864217255139248
|
25 |
+
}
|
26 |
+
},
|
27 |
+
"versions": {
|
28 |
+
"lighteval|gsm8k|5": 0
|
29 |
+
},
|
30 |
+
"config_tasks": {
|
31 |
+
"lighteval|gsm8k": {
|
32 |
+
"name": "gsm8k",
|
33 |
+
"prompt_function": "gsm8k",
|
34 |
+
"hf_repo": "gsm8k",
|
35 |
+
"hf_subset": "main",
|
36 |
+
"metric": [
|
37 |
+
"quasi_exact_match_gsm8k"
|
38 |
+
],
|
39 |
+
"hf_avail_splits": [
|
40 |
+
"train",
|
41 |
+
"test"
|
42 |
+
],
|
43 |
+
"evaluation_splits": [
|
44 |
+
"test"
|
45 |
+
],
|
46 |
+
"few_shots_split": null,
|
47 |
+
"few_shots_select": "random_sampling_from_train",
|
48 |
+
"generation_size": 256,
|
49 |
+
"stop_sequence": [
|
50 |
+
"Question:"
|
51 |
+
],
|
52 |
+
"output_regex": null,
|
53 |
+
"frozen": false,
|
54 |
+
"suite": [
|
55 |
+
"lighteval"
|
56 |
+
],
|
57 |
+
"original_num_docs": 1319,
|
58 |
+
"effective_num_docs": 1319,
|
59 |
+
"trust_dataset": true,
|
60 |
+
"must_remove_duplicate_docs": null
|
61 |
+
}
|
62 |
+
},
|
63 |
+
"summary_tasks": {
|
64 |
+
"lighteval|gsm8k|5": {
|
65 |
+
"hashes": {
|
66 |
+
"hash_examples": "0ed016e24e7512fd",
|
67 |
+
"hash_full_prompts": "a9c6515c2ecd2d07",
|
68 |
+
"hash_input_tokens": "a2f1c23e78a778da",
|
69 |
+
"hash_cont_tokens": "1eb0b65b73c78b80"
|
70 |
+
},
|
71 |
+
"truncated": 1319,
|
72 |
+
"non_truncated": 0,
|
73 |
+
"padded": 623,
|
74 |
+
"non_padded": 696,
|
75 |
+
"effective_few_shots": 5.0,
|
76 |
+
"num_truncated_few_shots": 0
|
77 |
+
}
|
78 |
+
},
|
79 |
+
"summary_general": {
|
80 |
+
"hashes": {
|
81 |
+
"hash_examples": "bc71463e88551d0e",
|
82 |
+
"hash_full_prompts": "f49691ec139bfeac",
|
83 |
+
"hash_input_tokens": "df014b250093da23",
|
84 |
+
"hash_cont_tokens": "38ec4cee483801e9"
|
85 |
+
},
|
86 |
+
"truncated": 1319,
|
87 |
+
"non_truncated": 0,
|
88 |
+
"padded": 623,
|
89 |
+
"non_padded": 696,
|
90 |
+
"num_truncated_few_shots": 0
|
91 |
+
}
|
92 |
+
}
|