Clean
Browse files
eval_results/Qwen/Qwen1.5-0.5B-Chat/main/gsm8k/results_2024-02-27T19-10-10.196610.json
DELETED
@@ -1,86 +0,0 @@
|
|
1 |
-
{
|
2 |
-
"config_general": {
|
3 |
-
"lighteval_sha": "?",
|
4 |
-
"num_fewshot_seeds": 1,
|
5 |
-
"override_batch_size": 1,
|
6 |
-
"max_samples": null,
|
7 |
-
"job_id": "",
|
8 |
-
"start_time": 341340.439719465,
|
9 |
-
"end_time": 341888.227747698,
|
10 |
-
"total_evaluation_time_secondes": "547.7880282330443",
|
11 |
-
"model_name": "Qwen/Qwen1.5-0.5B-Chat",
|
12 |
-
"model_sha": "7f630fd18dccab574ab1b78411a8753f989a55ac",
|
13 |
-
"model_dtype": "torch.bfloat16",
|
14 |
-
"model_size": "1.05 GB",
|
15 |
-
"config": null
|
16 |
-
},
|
17 |
-
"results": {
|
18 |
-
"lighteval|gsm8k|5": {
|
19 |
-
"qem": 0.03866565579984837,
|
20 |
-
"qem_stderr": 0.005310583162098074
|
21 |
-
}
|
22 |
-
},
|
23 |
-
"versions": {
|
24 |
-
"lighteval|gsm8k|5": 0
|
25 |
-
},
|
26 |
-
"config_tasks": {
|
27 |
-
"lighteval|gsm8k": {
|
28 |
-
"name": "gsm8k",
|
29 |
-
"prompt_function": "gsm8k",
|
30 |
-
"hf_repo": "gsm8k",
|
31 |
-
"hf_subset": "main",
|
32 |
-
"metric": [
|
33 |
-
"quasi_exact_match_gsm8k"
|
34 |
-
],
|
35 |
-
"hf_avail_splits": [
|
36 |
-
"train",
|
37 |
-
"test"
|
38 |
-
],
|
39 |
-
"evaluation_splits": [
|
40 |
-
"test"
|
41 |
-
],
|
42 |
-
"few_shots_split": null,
|
43 |
-
"few_shots_select": "random_sampling_from_train",
|
44 |
-
"generation_size": 256,
|
45 |
-
"stop_sequence": [
|
46 |
-
":",
|
47 |
-
"Question:",
|
48 |
-
"Question"
|
49 |
-
],
|
50 |
-
"output_regex": null,
|
51 |
-
"frozen": false,
|
52 |
-
"suite": [
|
53 |
-
"lighteval"
|
54 |
-
]
|
55 |
-
}
|
56 |
-
},
|
57 |
-
"summary_tasks": {
|
58 |
-
"lighteval|gsm8k|5": {
|
59 |
-
"hashes": {
|
60 |
-
"hash_examples": "0ed016e24e7512fd",
|
61 |
-
"hash_full_prompts": "66ff2108d151c6de",
|
62 |
-
"hash_input_tokens": "aa57736f8c45c33c",
|
63 |
-
"hash_cont_tokens": "4385070600734386"
|
64 |
-
},
|
65 |
-
"truncated": 0,
|
66 |
-
"non_truncated": 1319,
|
67 |
-
"padded": 0,
|
68 |
-
"non_padded": 1319,
|
69 |
-
"effective_few_shots": 5.0,
|
70 |
-
"num_truncated_few_shots": 0
|
71 |
-
}
|
72 |
-
},
|
73 |
-
"summary_general": {
|
74 |
-
"hashes": {
|
75 |
-
"hash_examples": "bc71463e88551d0e",
|
76 |
-
"hash_full_prompts": "84c6a9cfa115dda5",
|
77 |
-
"hash_input_tokens": "5aac43a2850ca464",
|
78 |
-
"hash_cont_tokens": "7d429d7833abb236"
|
79 |
-
},
|
80 |
-
"truncated": 0,
|
81 |
-
"non_truncated": 1319,
|
82 |
-
"padded": 0,
|
83 |
-
"non_padded": 1319,
|
84 |
-
"num_truncated_few_shots": 0
|
85 |
-
}
|
86 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
eval_results/Qwen/Qwen1.5-0.5B/main/gsm8k/details_lighteval|gsm8k|5_2024-02-27T20-01-30.199372.parquet
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:2c8888cb93fef108e23cac9b2bda6c01c573e5916b2aa26beeb947fb5ee9318a
|
3 |
-
size 4979350
|
|
|
|
|
|
|
|
eval_results/Qwen/Qwen1.5-0.5B/main/gsm8k/results_2024-02-27T20-01-30.199372.json
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:a1eb89fa50fc56c5ee362d873ca679c0fcab06271c59ab07ea77e341b9cda9f5
|
3 |
-
size 2089
|
|
|
|
|
|
|
|