Upload eval_results/deepseek-ai/deepseek-math-7b-rl/main/aimo_kaggle_hard_pot/results_2024-05-28T09-06-30.789083.json with huggingface_hub

eval_results/deepseek-ai/deepseek-math-7b-rl/main/aimo_kaggle_hard_pot/results_2024-05-28T09-06-30.789083.json

@@ -0,0 +1,193 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 2154508.660483309,
+    "end_time": 2155107.785841121,
+    "total_evaluation_time_secondes": "599.1253578118049",
+    "model_name": "deepseek-ai/deepseek-math-7b-rl",
+    "model_sha": "f3cd419a172ff5a100e9a563ebfe7900d78e9740",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "12.93 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle_hard_pot:v0|0": {
+      "qem": 0.02,
+      "qem_stderr": 0.02
+    },
+    "custom|aimo_kaggle_hard_pot:v1|0": {
+      "qem": 0.02,
+      "qem_stderr": 0.02
+    },
+    "custom|aimo_kaggle_hard_pot:v2|0": {
+      "qem": 0.04,
+      "qem_stderr": 0.02799416848895062
+    },
+    "custom|aimo_kaggle_hard_pot:_average|0": {
+      "qem": 0.02666666666666667,
+      "qem_stderr": 0.022664722829650208
+    },
+    "all": {
+      "qem": 0.02666666666666667,
+      "qem_stderr": 0.022664722829650208
+    }
+  },
+  "versions": {
+    "custom|aimo_kaggle_hard_pot:v0|0": 0,
+    "custom|aimo_kaggle_hard_pot:v1|0": 0,
+    "custom|aimo_kaggle_hard_pot:v2|0": 0
+  },
+  "config_tasks": {
+    "custom|aimo_kaggle_hard_pot:v0": {
+      "name": "aimo_kaggle_hard_pot:v0",
+      "prompt_function": "kaggle_hard_pot_prompt_fn_v0",
+      "hf_repo": "AI-MO/kaggle-validation-set-hard",
+      "hf_subset": "v0",
+      "metric": [
+        "quasi_exact_match_code_and_math"
+      ],
+      "hf_avail_splits": [
+        "train"
+      ],
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "num_samples": null,
+      "frozen": false,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 50,
+      "effective_num_docs": 50,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    },
+    "custom|aimo_kaggle_hard_pot:v1": {
+      "name": "aimo_kaggle_hard_pot:v1",
+      "prompt_function": "kaggle_hard_pot_prompt_fn_v1",
+      "hf_repo": "AI-MO/kaggle-validation-set-hard",
+      "hf_subset": "v0",
+      "metric": [
+        "quasi_exact_match_code_and_math"
+      ],
+      "hf_avail_splits": [
+        "train"
+      ],
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "num_samples": null,
+      "frozen": false,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 50,
+      "effective_num_docs": 50,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    },
+    "custom|aimo_kaggle_hard_pot:v2": {
+      "name": "aimo_kaggle_hard_pot:v2",
+      "prompt_function": "kaggle_hard_pot_prompt_fn_v2",
+      "hf_repo": "AI-MO/kaggle-validation-set-hard",
+      "hf_subset": "v0",
+      "metric": [
+        "quasi_exact_match_code_and_math"
+      ],
+      "hf_avail_splits": [
+        "train"
+      ],
+      "evaluation_splits": [
+        "train"
+      ],
+      "few_shots_split": null,
+      "few_shots_select": null,
+      "generation_size": 2048,
+      "stop_sequence": null,
+      "output_regex": null,
+      "num_samples": null,
+      "frozen": false,
+      "suite": [
+        "custom"
+      ],
+      "original_num_docs": 50,
+      "effective_num_docs": 50,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "custom|aimo_kaggle_hard_pot:v0|0": {
+      "hashes": {
+        "hash_examples": "303213a38d9f7512",
+        "hash_full_prompts": "3c670fa0e80bc301",
+        "hash_input_tokens": "1bd4187ad6963415",
+        "hash_cont_tokens": "4a5131314ac71945"
+      },
+      "truncated": 50,
+      "non_truncated": 0,
+      "padded": 37,
+      "non_padded": 13,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|aimo_kaggle_hard_pot:v1|0": {
+      "hashes": {
+        "hash_examples": "e4234b97ad92862f",
+        "hash_full_prompts": "09e7759a96f64e59",
+        "hash_input_tokens": "dc7f68cf14be3d61",
+        "hash_cont_tokens": "835e43e6085c79ec"
+      },
+      "truncated": 50,
+      "non_truncated": 0,
+      "padded": 32,
+      "non_padded": 18,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|aimo_kaggle_hard_pot:v2|0": {
+      "hashes": {
+        "hash_examples": "6396eb8833e13ba0",
+        "hash_full_prompts": "65f8b95ea99c7087",
+        "hash_input_tokens": "bddb67459e7601c5",
+        "hash_cont_tokens": "16c7b73b8b022a8d"
+      },
+      "truncated": 50,
+      "non_truncated": 0,
+      "padded": 30,
+      "non_padded": 20,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "648c9a107d279e1e",
+      "hash_full_prompts": "7f4c38eb08b3bb41",
+      "hash_input_tokens": "16f0fd76e0a2a165",
+      "hash_cont_tokens": "c21eba382df462d7"
+    },
+    "truncated": 150,
+    "non_truncated": 0,
+    "padded": 99,
+    "non_padded": 51,
+    "num_truncated_few_shots": 0
+  }
+}