Upload eval_results/orpo-explorers/hf-mixtral-47b-orpo-CapybaraPreferences-bs-64/main/ifeval/results_2024-05-12T08-07-23.943802.json with huggingface_hub

eval_results/orpo-explorers/hf-mixtral-47b-orpo-CapybaraPreferences-bs-64/main/ifeval/results_2024-05-12T08-07-23.943802.json

@@ -0,0 +1,103 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 1025854.660279755,
+    "end_time": 1031980.949205193,
+    "total_evaluation_time_secondes": "6126.288925437955",
+    "model_name": "orpo-explorers/hf-mixtral-47b-orpo-CapybaraPreferences-bs-64",
+    "model_sha": "a1c03e57f25f392a8b9a221ef219c0f2960dc0af",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "87.49 GB",
+    "config": null
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.4898336414048059,
+      "prompt_level_strict_acc_stderr": 0.02151212600135079,
+      "inst_level_strict_acc": 0.6007194244604317,
+      "inst_level_strict_acc_stderr": 0.0005103966952808571,
+      "prompt_level_loose_acc": 0.5249537892791127,
+      "prompt_level_loose_acc_stderr": 0.021489761058967338,
+      "inst_level_loose_acc": 0.6282973621103117,
+      "inst_level_loose_acc_stderr": 0.0005194209463555523
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.4898336414048059,
+      "prompt_level_strict_acc_stderr": 0.02151212600135079,
+      "inst_level_strict_acc": 0.6007194244604317,
+      "inst_level_strict_acc_stderr": 0.0005103966952808571,
+      "prompt_level_loose_acc": 0.5249537892791127,
+      "prompt_level_loose_acc_stderr": 0.021489761058967338,
+      "inst_level_loose_acc": 0.6282973621103117,
+      "inst_level_loose_acc_stderr": 0.0005194209463555523
+    }
+  },
+  "versions": {
+    "extended|ifeval|0": 0
+  },
+  "config_tasks": {
+    "extended|ifeval": {
+      "name": "ifeval",
+      "prompt_function": "ifeval_prompt",
+      "hf_repo": "wis-k/instruction-following-eval",
+      "hf_subset": "default",
+      "metric": [
+        "ifeval_metric"
+      ],
+      "hf_avail_splits": [
+        "train"
+      ],
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": "train",
+      "few_shots_select": "random_sampling",
+      "generation_size": 1280,
+      "stop_sequence": [],
+      "output_regex": null,
+      "num_samples": null,
+      "frozen": false,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "46932ca113ba6b0c",
+        "hash_input_tokens": "8b2e0300fbfb70bc",
+        "hash_cont_tokens": "b899880e9c9dcbe1"
+      },
+      "truncated": 541,
+      "non_truncated": 0,
+      "padded": 123,
+      "non_padded": 418,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "8ed52f7b74eb0590",
+      "hash_input_tokens": "9f4b7e9e7ff95d4a",
+      "hash_cont_tokens": "46135e9af7fa4efe"
+    },
+    "truncated": 541,
+    "non_truncated": 0,
+    "padded": 123,
+    "non_padded": 418,
+    "num_truncated_few_shots": 0
+  }
+}