Upload eval_results/AI-MO/deepseek-math-7b-sft/aimo_v01.37/gsm8k/results_2024-04-25T21-20-27.450702.json with huggingface_hub

eval_results/AI-MO/deepseek-math-7b-sft/aimo_v01.37/gsm8k/results_2024-04-25T21-20-27.450702.json

@@ -0,0 +1,92 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 182785.535231981,
+    "end_time": 183203.595512553,
+    "total_evaluation_time_secondes": "418.06028057201183",
+    "model_name": "AI-MO/deepseek-math-7b-sft",
+    "model_sha": "26c302bd121ff11f89fe70faa3bf3ef91b776cf1",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "12.93 GB",
+    "config": null
+  },
+  "results": {
+    "lighteval|gsm8k|5": {
+      "qem": 0.645185746777862,
+      "qem_stderr": 0.013179083387979205
+    },
+    "all": {
+      "qem": 0.645185746777862,
+      "qem_stderr": 0.013179083387979205
+    }
+  },
+  "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,
+      "must_remove_duplicate_docs": null
+    }
+  },
+  "summary_tasks": {
+    "lighteval|gsm8k|5": {
+      "hashes": {
+        "hash_examples": "0ed016e24e7512fd",
+        "hash_full_prompts": "b6d97958f0ed1df8",
+        "hash_input_tokens": "3f58ee5e568409ec",
+        "hash_cont_tokens": "2203b3127ec088f6"
+      },
+      "truncated": 1319,
+      "non_truncated": 0,
+      "padded": 611,
+      "non_padded": 708,
+      "effective_few_shots": 5.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "bc71463e88551d0e",
+      "hash_full_prompts": "b5a5dacaed0816ff",
+      "hash_input_tokens": "e4e0fd8eba82bd47",
+      "hash_cont_tokens": "924dcaf34158d1fe"
+    },
+    "truncated": 1319,
+    "non_truncated": 0,
+    "padded": 611,
+    "non_padded": 708,
+    "num_truncated_few_shots": 0
+  }
+}