Upload eval_results/alvarobartt/mistral-7b-orpo-capybara-reproduction/main/ifeval/results_2024-03-28T17-38-53.532479.json with huggingface_hub

eval_results/alvarobartt/mistral-7b-orpo-capybara-reproduction/main/ifeval/results_2024-03-28T17-38-53.532479.json

@@ -0,0 +1,101 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 835951.878234298,
+    "end_time": 836742.095991933,
+    "total_evaluation_time_secondes": "790.2177576350514",
+    "model_name": "alvarobartt/mistral-7b-orpo-capybara-reproduction",
+    "model_sha": "0a87f28dcb8483abd79ba9c6c168fc0c78fa78af",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "13.99 GB",
+    "config": null
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.30868761552680224,
+      "prompt_level_strict_acc_stderr": 0.019879245251116444,
+      "inst_level_strict_acc": 0.4352517985611511,
+      "inst_level_strict_acc_stderr": 0.0005363021850641,
+      "prompt_level_loose_acc": 0.3364140480591497,
+      "prompt_level_loose_acc_stderr": 0.02033240600470129,
+      "inst_level_loose_acc": 0.4628297362110312,
+      "inst_level_loose_acc_stderr": 0.0005370563972627654
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.30868761552680224,
+      "prompt_level_strict_acc_stderr": 0.019879245251116444,
+      "inst_level_strict_acc": 0.4352517985611511,
+      "inst_level_strict_acc_stderr": 0.0005363021850641,
+      "prompt_level_loose_acc": 0.3364140480591497,
+      "prompt_level_loose_acc_stderr": 0.02033240600470129,
+      "inst_level_loose_acc": 0.4628297362110312,
+      "inst_level_loose_acc_stderr": 0.0005370563972627654
+    }
+  },
+  "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,
+      "frozen": false,
+      "suite": [
+        "extended"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null
+    }
+  },
+  "summary_tasks": {
+    "extended|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "1674fbca8428210a",
+        "hash_cont_tokens": "08529dcc809e91f1"
+      },
+      "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": "45f8422f6ad2da79",
+      "hash_input_tokens": "211ca010114f2960",
+      "hash_cont_tokens": "4bb4f3415f191af3"
+    },
+    "truncated": 541,
+    "non_truncated": 0,
+    "padded": 123,
+    "non_padded": 418,
+    "num_truncated_few_shots": 0
+  }
+}