Upload eval_results/orpo-explorers/kaist-mistral-orpo-OHP-15k-Stratified-1-beta-0.1-2epoch/main/gsm8k/results_2024-05-09T18-08-08.741899.json with huggingface_hub

eval_results/orpo-explorers/kaist-mistral-orpo-OHP-15k-Stratified-1-beta-0.1-2epoch/main/gsm8k/results_2024-05-09T18-08-08.741899.json

@@ -0,0 +1,99 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 599274.603479495,
+    "end_time": 600070.020875774,
+    "total_evaluation_time_secondes": "795.4173962789355",
+    "model_name": "orpo-explorers/kaist-mistral-orpo-OHP-15k-Stratified-1-beta-0.1-2epoch",
+    "model_sha": "3c9953262edf0986a57cbf6aff79956c154e84dc",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "13.99 GB",
+    "config": null
+  },
+  "results": {
+    "lighteval|gsm8k|5": {
+      "qem": 0.3866565579984837,
+      "qem_stderr": 0.0134139550959653,
+      "maj@8": 0.574677786201668,
+      "maj@8_stderr": 0.013618006363084797
+    },
+    "all": {
+      "qem": 0.3866565579984837,
+      "qem_stderr": 0.0134139550959653,
+      "maj@8": 0.574677786201668,
+      "maj@8_stderr": 0.013618006363084797
+    }
+  },
+  "versions": {
+    "lighteval|gsm8k|5": 0
+  },
+  "config_tasks": {
+    "lighteval|gsm8k": {
+      "name": "gsm8k",
+      "prompt_function": "gsm8k",
+      "hf_repo": "gsm8k",
+      "hf_subset": "main",
+      "metric": [
+        "quasi_exact_match_gsm8k",
+        "maj_at_8_gsm8k"
+      ],
+      "hf_avail_splits": [
+        "train",
+        "test"
+      ],
+      "evaluation_splits": [
+        "test"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": "random_sampling_from_train",
+      "generation_size": 256,
+      "stop_sequence": [
+        "Question:"
+      ],
+      "output_regex": null,
+      "num_samples": null,
+      "frozen": false,
+      "suite": [
+        "lighteval"
+      ],
+      "original_num_docs": 1319,
+      "effective_num_docs": 1319,
+      "trust_dataset": true,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "lighteval|gsm8k|5": {
+      "hashes": {
+        "hash_examples": "0ed016e24e7512fd",
+        "hash_full_prompts": "faa09535944c3187",
+        "hash_input_tokens": "61e7bfb27a50fc48",
+        "hash_cont_tokens": "2ec520e5d85f2c8b"
+      },
+      "truncated": 1319,
+      "non_truncated": 0,
+      "padded": 652,
+      "non_padded": 667,
+      "effective_few_shots": 5.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "bc71463e88551d0e",
+      "hash_full_prompts": "24555d96d1b06364",
+      "hash_input_tokens": "fbc91ddb76fcfe8f",
+      "hash_cont_tokens": "88c4930837aa252a"
+    },
+    "truncated": 1319,
+    "non_truncated": 0,
+    "padded": 652,
+    "non_padded": 667,
+    "num_truncated_few_shots": 0
+  }
+}