Upload eval_results/HuggingFaceH4/llama-3.1-70b-sft/v00.03-step-000001941/ifeval/results_2024-09-21T12-01-17.855595.json with huggingface_hub

eval_results/HuggingFaceH4/llama-3.1-70b-sft/v00.03-step-000001941/ifeval/results_2024-09-21T12-01-17.855595.json

@@ -0,0 +1,103 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 95321.51246627,
+    "end_time": 100926.256391966,
+    "total_evaluation_time_secondes": "5604.743925696006",
+    "model_name": "HuggingFaceH4/llama-3.1-70b-sft",
+    "model_sha": "f13dd58cf8c8b73b000234ec2b0a4a979b63efa0",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "131.42 GB",
+    "config": null
+  },
+  "results": {
+    "extended|ifeval|0": {
+      "prompt_level_strict_acc": 0.6358595194085028,
+      "prompt_level_strict_acc_stderr": 0.02070704795859199,
+      "inst_level_strict_acc": 0.7266187050359713,
+      "inst_level_strict_acc_stderr": 0.0004994264632868299,
+      "prompt_level_loose_acc": 0.7282809611829945,
+      "prompt_level_loose_acc_stderr": 0.019143116099594022,
+      "inst_level_loose_acc": 0.8009592326139089,
+      "inst_level_loose_acc_stderr": 0.0004409319659503015
+    },
+    "all": {
+      "prompt_level_strict_acc": 0.6358595194085028,
+      "prompt_level_strict_acc_stderr": 0.02070704795859199,
+      "inst_level_strict_acc": 0.7266187050359713,
+      "inst_level_strict_acc_stderr": 0.0004994264632868299,
+      "prompt_level_loose_acc": 0.7282809611829945,
+      "prompt_level_loose_acc_stderr": 0.019143116099594022,
+      "inst_level_loose_acc": 0.8009592326139089,
+      "inst_level_loose_acc_stderr": 0.0004409319659503015
+    }
+  },
+  "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": "7ea7bf2a8edba8f4",
+        "hash_input_tokens": "ca4a056d1ce1b77c",
+        "hash_cont_tokens": "2033d72f0011931d"
+      },
+      "truncated": 541,
+      "non_truncated": 0,
+      "padded": 115,
+      "non_padded": 426,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "ea046ab2c6fc5928",
+      "hash_full_prompts": "45f8422f6ad2da79",
+      "hash_input_tokens": "05d4422f76889c35",
+      "hash_cont_tokens": "17c9224c9f1156a1"
+    },
+    "truncated": 541,
+    "non_truncated": 0,
+    "padded": 115,
+    "non_padded": 426,
+    "num_truncated_few_shots": 0
+  }
+}