Upload eval_results/HuggingFaceH4/qwen-1.5-1.8b-dpo/v0.2/gsm8k/results_2024-03-22T22-43-48.017523.json with huggingface_hub

eval_results/HuggingFaceH4/qwen-1.5-1.8b-dpo/v0.2/gsm8k/results_2024-03-22T22-43-48.017523.json

@@ -0,0 +1,87 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 1,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 728442.148091353,
+    "end_time": 729110.453855462,
+    "total_evaluation_time_secondes": "668.3057641090127",
+    "model_name": "HuggingFaceH4/qwen-1.5-1.8b-dpo",
+    "model_sha": "d8c1d30de4a597868755e6cb3e48d2e5abef047f",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "3.79 GB",
+    "config": null
+  },
+  "results": {
+    "lighteval|gsm8k|5": {
+      "qem": 0.3237300985595148,
+      "qem_stderr": 0.01288824739737114
+    }
+  },
+  "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"
+      ],
+      "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,
+      "frozen": false,
+      "suite": [
+        "lighteval"
+      ],
+      "original_num_docs": 1319,
+      "effective_num_docs": 1319,
+      "trust_dataset": true
+    }
+  },
+  "summary_tasks": {
+    "lighteval|gsm8k|5": {
+      "hashes": {
+        "hash_examples": "0ed016e24e7512fd",
+        "hash_full_prompts": "a9c6515c2ecd2d07",
+        "hash_input_tokens": "cf5e91d1c1d6ca10",
+        "hash_cont_tokens": "9fd55a5447c61cf0"
+      },
+      "truncated": 0,
+      "non_truncated": 1319,
+      "padded": 0,
+      "non_padded": 1319,
+      "effective_few_shots": 5.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "bc71463e88551d0e",
+      "hash_full_prompts": "f49691ec139bfeac",
+      "hash_input_tokens": "4db9bb980ae67af7",
+      "hash_cont_tokens": "e015129377dea4f9"
+    },
+    "truncated": 0,
+    "non_truncated": 1319,
+    "padded": 0,
+    "non_padded": 1319,
+    "num_truncated_few_shots": 0
+  }
+}