{ "config_general": { "model_name": "TheBloke/WizardLM-13B-V1-1-SuperHOT-8K-GPTQ", "model_sha": "085eb5cd394f30d72bf5efcf83a580e87264b3e8", "model_size": "6.92 GB", "model_dtype": "None", "lighteval_sha": "0f318ecf002208468154899217b3ba7c6ae09374", "num_few_shot_default": 0, "num_fewshot_seeds": 1, "override_batch_size": 1, "max_samples": null, "job_id": "" }, "results": { "harness|drop|3": { "em": 0.22158137583892618, "em_stderr": 0.004253171428083824, "f1": 0.28616296140939684, "f1_stderr": 0.004276937020149761 }, "harness|gsm8k|5": { "acc": 0.006823351023502654, "acc_stderr": 0.0022675371022544783 }, "harness|winogrande|5": { "acc": 0.7434885556432518, "acc_stderr": 0.012273648008759979 }, "all": { "em": 0.22158137583892618, "em_stderr": 0.004253171428083824, "f1": 0.28616296140939684, "f1_stderr": 0.004276937020149761, "acc": 0.3751559533333772, "acc_stderr": 0.007270592555507228 } }, "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": "fbe785c62d941897", "hash_cont_tokens": "765bf0d9fe7f36b6" }, "truncated": 0, "non-truncated": 9536, "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": "c3e21d310721c449" }, "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": "9f72a6bc6743d18f", "hash_cont_tokens": "1eb50782b30a77e9" }, "total_evaluation_time_secondes": "6928.959567308426", "truncated": 0, "non-truncated": 13389, "padded": 2432, "non-padded": 10957, "num_truncated_few_shots": 0 } }