Spaces:
Runtime error
Runtime error
data_only_hallucination_leaderboard
/
meta-llama
/Llama-2-13b-hf
/results_2023-09-07T15-27-15.010124.json
| { | |
| "config_general": { | |
| "model_name": "meta-llama/Llama-2-13b-hf", | |
| "model_sha": "db6b8eb1feabb38985fdf785a89895959e944936", | |
| "model_size": "24.32 GB", | |
| "model_dtype": "torch.float16", | |
| "lighteval_sha": "457ac5672c5fdebfd6bc95bb94bda825c148eccf", | |
| "num_few_shot_default": 0, | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "" | |
| }, | |
| "results": { | |
| "harness|drop|3": { | |
| "em": 0.0014681208053691276, | |
| "em_stderr": 0.00039210421902982666, | |
| "f1": 0.0607822986577181, | |
| "f1_stderr": 0.0013583957676382913 | |
| }, | |
| "harness|gsm8k|5": { | |
| "acc": 0.10841546626231995, | |
| "acc_stderr": 0.008563852506627487 | |
| }, | |
| "harness|winogrande|5": { | |
| "acc": 0.7663772691397001, | |
| "acc_stderr": 0.011892194477183524 | |
| }, | |
| "all": { | |
| "em": 0.0014681208053691276, | |
| "em_stderr": 0.00039210421902982666, | |
| "f1": 0.0607822986577181, | |
| "f1_stderr": 0.0013583957676382913, | |
| "acc": 0.43739636770101, | |
| "acc_stderr": 0.010228023491905505 | |
| } | |
| }, | |
| "versions": { | |
| "harness|drop|3": 1, | |
| "harness|gsm8k|5": 0, | |
| "harness|winogrande|5": 0, | |
| "all": 0 | |
| }, | |
| "config_tasks": { | |
| "harness|drop": "LM Harness task", | |
| "harness|gsm8k": "LM Harness task", | |
| "harness|winogrande": "LM Harness task" | |
| }, | |
| "summary_tasks": { | |
| "harness|drop|3": { | |
| "hashes": { | |
| "hash_examples": "1d27416e8324e9a3", | |
| "hash_full_prompts": "a5513ff9a741b385", | |
| "hash_input_tokens": "42076f0efbb50aa6", | |
| "hash_cont_tokens": "c9346ec21b7560de" | |
| }, | |
| "truncated": 3, | |
| "non-truncated": 9533, | |
| "padded": 0, | |
| "non-padded": 9536, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|gsm8k|5": { | |
| "hashes": { | |
| "hash_examples": "4c0843a5d99bcfdc", | |
| "hash_full_prompts": "41d55e83abc0e02d", | |
| "hash_input_tokens": "bda342e47b5099b2", | |
| "hash_cont_tokens": "32cafa77d8a3f04e" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1319, | |
| "padded": 0, | |
| "non-padded": 1319, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|winogrande|5": { | |
| "hashes": { | |
| "hash_examples": "aada0a176fd81218", | |
| "hash_full_prompts": "c8655cbd12de8409", | |
| "hash_input_tokens": "c0bedf98cb040854", | |
| "hash_cont_tokens": "f08975ad6f2d5864" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 2534, | |
| "padded": 2432, | |
| "non-padded": 102, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "9b4d8993161e637d", | |
| "hash_full_prompts": "08215e527b7e60a5", | |
| "hash_input_tokens": "a12f3e3c934bd78b", | |
| "hash_cont_tokens": "4d8f1e04b1d56e40" | |
| }, | |
| "total_evaluation_time_secondes": "6066.877633810043", | |
| "truncated": 3, | |
| "non-truncated": 13386, | |
| "padded": 2432, | |
| "non-padded": 10957, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |