Upload eval_results/mistralai/Mistral-7B-Instruct-v0.2/main/ifeval/results_2024-03-02T15-46-44.707676.json with huggingface_hub

eval_results/mistralai/Mistral-7B-Instruct-v0.2/main/ifeval/results_2024-03-02T15-46-44.707676.json

@@ -0,0 +1,89 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 2865534.91015085,
+    "end_time": 2866324.183763554,
+    "total_evaluation_time_secondes": "789.2736127036624",
+    "model_name": "mistralai/Mistral-7B-Instruct-v0.2",
+    "model_sha": "cf47bb3e18fe41a5351bc36eef76e9c900847c89",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "13.99 GB",
+    "config": null
+  },
+  "results": {
+    "custom|ifeval|0": {
+      "prompt_level_strict_acc": 0.49722735674676527,
+      "prompt_level_strict_acc_stderr": 0.02151624332354812,
+      "inst_level_strict_acc": 0.6139088729016786,
+      "inst_level_strict_acc_stderr": 0.0004971694286686962,
+      "prompt_level_loose_acc": 0.5397412199630314,
+      "prompt_level_loose_acc_stderr": 0.021448501434135046,
+      "inst_level_loose_acc": 0.6534772182254197,
+      "inst_level_loose_acc_stderr": 0.0004908858224509614
+    }
+  },
+  "versions": {
+    "custom|ifeval|0": 0
+  },
+  "config_tasks": {
+    "custom|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": [
+        "custom"
+      ],
+      "original_num_docs": 541,
+      "effective_num_docs": 541
+    }
+  },
+  "summary_tasks": {
+    "custom|ifeval|0": {
+      "hashes": {
+        "hash_examples": "e99cbf567588d7c6",
+        "hash_full_prompts": "50f8a8f5a08077cd",
+        "hash_input_tokens": "fb1ed90207cf652b",
+        "hash_cont_tokens": "769b23a4669e8f2b"
+      },
+      "truncated": 0,
+      "non_truncated": 541,
+      "padded": 0,
+      "non_padded": 541,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "9c04db679f2c9132",
+      "hash_input_tokens": "af9a63ef7d2de8ba",
+      "hash_cont_tokens": "975d0719b9bfe6a6"
+    },
+    "truncated": 0,
+    "non_truncated": 541,
+    "padded": 0,
+    "non_padded": 541,
+    "num_truncated_few_shots": 0
+  }
+}