Upload results_2025-05-11T03-57-31.270083.json with huggingface_hub
Browse files
results_2025-05-11T03-57-31.270083.json
ADDED
@@ -0,0 +1,120 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"config_general": {
|
3 |
+
"lighteval_sha": "?",
|
4 |
+
"num_fewshot_seeds": 1,
|
5 |
+
"max_samples": null,
|
6 |
+
"job_id": 0,
|
7 |
+
"start_time": 7391.813776305,
|
8 |
+
"end_time": 7504.824690543,
|
9 |
+
"total_evaluation_time_secondes": "113.01091423799971",
|
10 |
+
"model_name": "Qwen/Qwen3-0.6B",
|
11 |
+
"model_sha": "",
|
12 |
+
"model_dtype": null,
|
13 |
+
"model_size": null,
|
14 |
+
"generation_parameters": {
|
15 |
+
"early_stopping": null,
|
16 |
+
"repetition_penalty": null,
|
17 |
+
"frequency_penalty": null,
|
18 |
+
"length_penalty": null,
|
19 |
+
"presence_penalty": null,
|
20 |
+
"max_new_tokens": null,
|
21 |
+
"min_new_tokens": null,
|
22 |
+
"seed": null,
|
23 |
+
"stop_tokens": null,
|
24 |
+
"temperature": null,
|
25 |
+
"top_k": null,
|
26 |
+
"min_p": null,
|
27 |
+
"top_p": null,
|
28 |
+
"truncate_prompt": null,
|
29 |
+
"response_format": null
|
30 |
+
}
|
31 |
+
},
|
32 |
+
"results": {
|
33 |
+
"original|mmlu:anatomy|0": {
|
34 |
+
"acc": 0.1925925925925926,
|
35 |
+
"acc_stderr": 0.03406542058502653
|
36 |
+
},
|
37 |
+
"all": {
|
38 |
+
"acc": 0.1925925925925926,
|
39 |
+
"acc_stderr": 0.03406542058502653
|
40 |
+
}
|
41 |
+
},
|
42 |
+
"versions": {
|
43 |
+
"original|mmlu:anatomy|0": 0
|
44 |
+
},
|
45 |
+
"config_tasks": {
|
46 |
+
"original|mmlu:anatomy": {
|
47 |
+
"name": "mmlu:anatomy",
|
48 |
+
"prompt_function": "mmlu_anatomy",
|
49 |
+
"hf_repo": "cais/mmlu",
|
50 |
+
"hf_subset": "anatomy",
|
51 |
+
"metric": [
|
52 |
+
{
|
53 |
+
"metric_name": "acc",
|
54 |
+
"higher_is_better": true,
|
55 |
+
"category": "8",
|
56 |
+
"use_case": "1",
|
57 |
+
"sample_level_fn": "compute",
|
58 |
+
"corpus_level_fn": "mean"
|
59 |
+
}
|
60 |
+
],
|
61 |
+
"hf_revision": null,
|
62 |
+
"hf_filter": null,
|
63 |
+
"hf_avail_splits": [
|
64 |
+
"auxiliary_train",
|
65 |
+
"test",
|
66 |
+
"validation",
|
67 |
+
"dev"
|
68 |
+
],
|
69 |
+
"trust_dataset": true,
|
70 |
+
"evaluation_splits": [
|
71 |
+
"test"
|
72 |
+
],
|
73 |
+
"few_shots_split": "dev",
|
74 |
+
"few_shots_select": "sequential",
|
75 |
+
"generation_size": 1,
|
76 |
+
"generation_grammar": null,
|
77 |
+
"stop_sequence": [
|
78 |
+
"\n"
|
79 |
+
],
|
80 |
+
"num_samples": null,
|
81 |
+
"suite": [
|
82 |
+
"original",
|
83 |
+
"mmlu"
|
84 |
+
],
|
85 |
+
"original_num_docs": 135,
|
86 |
+
"effective_num_docs": 135,
|
87 |
+
"must_remove_duplicate_docs": false,
|
88 |
+
"version": 0
|
89 |
+
}
|
90 |
+
},
|
91 |
+
"summary_tasks": {
|
92 |
+
"original|mmlu:anatomy|0": {
|
93 |
+
"hashes": {
|
94 |
+
"hash_examples": "2ace6ded4afc2a5e",
|
95 |
+
"hash_full_prompts": "2ace6ded4afc2a5e",
|
96 |
+
"hash_input_tokens": "2208f6c9b1418fb3",
|
97 |
+
"hash_cont_tokens": "b6f29f2efe9d60e5"
|
98 |
+
},
|
99 |
+
"truncated": 0,
|
100 |
+
"non_truncated": 135,
|
101 |
+
"padded": 0,
|
102 |
+
"non_padded": 540,
|
103 |
+
"effective_few_shots": 0.0,
|
104 |
+
"num_truncated_few_shots": 0
|
105 |
+
}
|
106 |
+
},
|
107 |
+
"summary_general": {
|
108 |
+
"hashes": {
|
109 |
+
"hash_examples": "b4397c7c7be2716f",
|
110 |
+
"hash_full_prompts": "b4397c7c7be2716f",
|
111 |
+
"hash_input_tokens": "b6d70b8b90cdad6b",
|
112 |
+
"hash_cont_tokens": "d0f0c11bb15a1909"
|
113 |
+
},
|
114 |
+
"truncated": 0,
|
115 |
+
"non_truncated": 135,
|
116 |
+
"padded": 0,
|
117 |
+
"non_padded": 540,
|
118 |
+
"num_truncated_few_shots": 0
|
119 |
+
}
|
120 |
+
}
|