Upload eval_results/alignment-handbook/zephyr-2b-gemma-dpo-mix7-beta-0.05-epoch-6/main/hellaswag/results_2024-03-07T11-05-53.241131.json with huggingface_hub
Browse files
eval_results/alignment-handbook/zephyr-2b-gemma-dpo-mix7-beta-0.05-epoch-6/main/hellaswag/results_2024-03-07T11-05-53.241131.json
ADDED
|
@@ -0,0 +1,90 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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": 2313055.015010817,
|
| 9 |
+
"end_time": 2313519.405629995,
|
| 10 |
+
"total_evaluation_time_secondes": "464.3906191778369",
|
| 11 |
+
"model_name": "alignment-handbook/zephyr-2b-gemma-dpo-mix7-beta-0.05-epoch-6",
|
| 12 |
+
"model_sha": "1ecb6ade923a19d209a5ffe161833e4fe42c3580",
|
| 13 |
+
"model_dtype": "torch.bfloat16",
|
| 14 |
+
"model_size": "4.68 GB",
|
| 15 |
+
"config": null
|
| 16 |
+
},
|
| 17 |
+
"results": {
|
| 18 |
+
"lighteval|hellaswag|10": {
|
| 19 |
+
"acc": 0.5403306114319857,
|
| 20 |
+
"acc_stderr": 0.004973522582431218,
|
| 21 |
+
"acc_norm": 0.7210714997012547,
|
| 22 |
+
"acc_norm_stderr": 0.004475557360359704
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"versions": {
|
| 26 |
+
"lighteval|hellaswag|10": 0
|
| 27 |
+
},
|
| 28 |
+
"config_tasks": {
|
| 29 |
+
"lighteval|hellaswag": {
|
| 30 |
+
"name": "hellaswag",
|
| 31 |
+
"prompt_function": "hellaswag_harness",
|
| 32 |
+
"hf_repo": "hellaswag",
|
| 33 |
+
"hf_subset": "default",
|
| 34 |
+
"metric": [
|
| 35 |
+
"loglikelihood_acc",
|
| 36 |
+
"loglikelihood_acc_norm"
|
| 37 |
+
],
|
| 38 |
+
"hf_avail_splits": [
|
| 39 |
+
"train",
|
| 40 |
+
"test",
|
| 41 |
+
"validation"
|
| 42 |
+
],
|
| 43 |
+
"evaluation_splits": [
|
| 44 |
+
"validation"
|
| 45 |
+
],
|
| 46 |
+
"few_shots_split": null,
|
| 47 |
+
"few_shots_select": "random_sampling_from_train",
|
| 48 |
+
"generation_size": -1,
|
| 49 |
+
"stop_sequence": [
|
| 50 |
+
"\n"
|
| 51 |
+
],
|
| 52 |
+
"output_regex": null,
|
| 53 |
+
"frozen": false,
|
| 54 |
+
"suite": [
|
| 55 |
+
"lighteval"
|
| 56 |
+
],
|
| 57 |
+
"original_num_docs": 10042,
|
| 58 |
+
"effective_num_docs": 10042
|
| 59 |
+
}
|
| 60 |
+
},
|
| 61 |
+
"summary_tasks": {
|
| 62 |
+
"lighteval|hellaswag|10": {
|
| 63 |
+
"hashes": {
|
| 64 |
+
"hash_examples": "31985c805c3a737e",
|
| 65 |
+
"hash_full_prompts": "48de8fbcf45195b3",
|
| 66 |
+
"hash_input_tokens": "08c832518b7403b1",
|
| 67 |
+
"hash_cont_tokens": "6232829aec74d42d"
|
| 68 |
+
},
|
| 69 |
+
"truncated": 0,
|
| 70 |
+
"non_truncated": 10042,
|
| 71 |
+
"padded": 39905,
|
| 72 |
+
"non_padded": 263,
|
| 73 |
+
"effective_few_shots": 10.0,
|
| 74 |
+
"num_truncated_few_shots": 0
|
| 75 |
+
}
|
| 76 |
+
},
|
| 77 |
+
"summary_general": {
|
| 78 |
+
"hashes": {
|
| 79 |
+
"hash_examples": "63bc2cf8bae03fbc",
|
| 80 |
+
"hash_full_prompts": "20d63c990d3e8572",
|
| 81 |
+
"hash_input_tokens": "e93858038f602554",
|
| 82 |
+
"hash_cont_tokens": "c2b17126697982c3"
|
| 83 |
+
},
|
| 84 |
+
"truncated": 0,
|
| 85 |
+
"non_truncated": 10042,
|
| 86 |
+
"padded": 39905,
|
| 87 |
+
"non_padded": 263,
|
| 88 |
+
"num_truncated_few_shots": 0
|
| 89 |
+
}
|
| 90 |
+
}
|