Upload results/meta-llama/Meta-Llama-3.1-70B-Instruct/results_2024-09-23T13-26-25.903852.json with huggingface_hub
Browse files
results/meta-llama/Meta-Llama-3.1-70B-Instruct/results_2024-09-23T13-26-25.903852.json
ADDED
|
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config_general": {
|
| 3 |
+
"lighteval_sha": "d3fad8bc430df2c96132a56decf820e760575531",
|
| 4 |
+
"num_fewshot_seeds": 1,
|
| 5 |
+
"override_batch_size": -1,
|
| 6 |
+
"max_samples": null,
|
| 7 |
+
"job_id": "",
|
| 8 |
+
"start_time": 1651620.991836474,
|
| 9 |
+
"end_time": 1652240.273853538,
|
| 10 |
+
"total_evaluation_time_secondes": "619.282017064048",
|
| 11 |
+
"model_name": "meta-llama/Meta-Llama-3.1-70B-Instruct",
|
| 12 |
+
"model_sha": "",
|
| 13 |
+
"model_dtype": null,
|
| 14 |
+
"model_size": null
|
| 15 |
+
},
|
| 16 |
+
"results": {
|
| 17 |
+
"custom|musr:murder_mysteries|0": {
|
| 18 |
+
"pem": 0.0,
|
| 19 |
+
"pem_stderr": 0.0,
|
| 20 |
+
"acc": 0.536,
|
| 21 |
+
"acc_stderr": 0.031603975145223735
|
| 22 |
+
},
|
| 23 |
+
"all": {
|
| 24 |
+
"pem": 0.0,
|
| 25 |
+
"pem_stderr": 0.0,
|
| 26 |
+
"acc": 0.536,
|
| 27 |
+
"acc_stderr": 0.031603975145223735
|
| 28 |
+
}
|
| 29 |
+
},
|
| 30 |
+
"versions": {
|
| 31 |
+
"custom|musr:murder_mysteries|0": 0
|
| 32 |
+
},
|
| 33 |
+
"config_tasks": {
|
| 34 |
+
"custom|musr:murder_mysteries": {
|
| 35 |
+
"name": "musr:murder_mysteries",
|
| 36 |
+
"prompt_function": "musr_prompt_mm",
|
| 37 |
+
"hf_repo": "TAUR-Lab/MuSR",
|
| 38 |
+
"hf_subset": "default",
|
| 39 |
+
"metric": [
|
| 40 |
+
{
|
| 41 |
+
"metric_name": "acc",
|
| 42 |
+
"higher_is_better": true,
|
| 43 |
+
"category": "8",
|
| 44 |
+
"use_case": "1",
|
| 45 |
+
"sample_level_fn": "compute",
|
| 46 |
+
"corpus_level_fn": "mean"
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"metric_name": "pem",
|
| 50 |
+
"higher_is_better": true,
|
| 51 |
+
"category": "3",
|
| 52 |
+
"use_case": "1",
|
| 53 |
+
"sample_level_fn": "compute",
|
| 54 |
+
"corpus_level_fn": "mean"
|
| 55 |
+
}
|
| 56 |
+
],
|
| 57 |
+
"hf_avail_splits": [
|
| 58 |
+
"murder_mysteries"
|
| 59 |
+
],
|
| 60 |
+
"evaluation_splits": [
|
| 61 |
+
"murder_mysteries"
|
| 62 |
+
],
|
| 63 |
+
"few_shots_split": "",
|
| 64 |
+
"few_shots_select": "random_sampling",
|
| 65 |
+
"generation_size": 10000,
|
| 66 |
+
"generation_grammar": null,
|
| 67 |
+
"stop_sequence": [],
|
| 68 |
+
"output_regex": null,
|
| 69 |
+
"num_samples": null,
|
| 70 |
+
"frozen": false,
|
| 71 |
+
"suite": [
|
| 72 |
+
"custom"
|
| 73 |
+
],
|
| 74 |
+
"original_num_docs": 250,
|
| 75 |
+
"effective_num_docs": 250,
|
| 76 |
+
"trust_dataset": null,
|
| 77 |
+
"must_remove_duplicate_docs": null,
|
| 78 |
+
"version": 0
|
| 79 |
+
}
|
| 80 |
+
},
|
| 81 |
+
"summary_tasks": {
|
| 82 |
+
"custom|musr:murder_mysteries|0": {
|
| 83 |
+
"hashes": {
|
| 84 |
+
"hash_examples": "e57d289b06a7b768",
|
| 85 |
+
"hash_full_prompts": "ef87de16b29e1ad6",
|
| 86 |
+
"hash_input_tokens": "261a3362fb1e3a49",
|
| 87 |
+
"hash_cont_tokens": "4ccd93a3cc7758cc"
|
| 88 |
+
},
|
| 89 |
+
"truncated": 0,
|
| 90 |
+
"non_truncated": 500,
|
| 91 |
+
"padded": 0,
|
| 92 |
+
"non_padded": 750,
|
| 93 |
+
"effective_few_shots": 0.0,
|
| 94 |
+
"num_truncated_few_shots": 0
|
| 95 |
+
}
|
| 96 |
+
},
|
| 97 |
+
"summary_general": {
|
| 98 |
+
"hashes": {
|
| 99 |
+
"hash_examples": "12857cef73fec968",
|
| 100 |
+
"hash_full_prompts": "b4d2c9f7b84e6c86",
|
| 101 |
+
"hash_input_tokens": "9492dc004d54bb27",
|
| 102 |
+
"hash_cont_tokens": "b9e0dc01cae2cba0"
|
| 103 |
+
},
|
| 104 |
+
"truncated": 0,
|
| 105 |
+
"non_truncated": 500,
|
| 106 |
+
"padded": 0,
|
| 107 |
+
"non_padded": 750,
|
| 108 |
+
"num_truncated_few_shots": 0
|
| 109 |
+
}
|
| 110 |
+
}
|