Upload eval_results/AI-MO/CodeLlama-13b-Python-hf-sft/aimo_v01.21/aimo_kaggle_hard_pot/results_2024-06-06T04-50-49.035447.json with huggingface_hub

eval_results/AI-MO/CodeLlama-13b-Python-hf-sft/aimo_v01.21/aimo_kaggle_hard_pot/results_2024-06-06T04-50-49.035447.json

@@ -0,0 +1,193 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 2323298.264689084,
+    "end_time": 2324324.875178094,
+    "total_evaluation_time_secondes": "1026.6104890098795",
+    "model_name": "AI-MO/CodeLlama-13b-Python-hf-sft",
+    "model_sha": "a4f5bedc01bc294426aaed69ca13f1c19abf945b",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "24.56 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle_hard_pot:v0|0": {
+      "qem": 0.06,
+      "qem_stderr": 0.033926691677251195
+    },
+    "custom|aimo_kaggle_hard_pot:v1|0": {
+      "qem": 0.02,
+      "qem_stderr": 0.02
+    },
+    "custom|aimo_kaggle_hard_pot:v2|0": {
+      "qem": 0.12,
+      "qem_stderr": 0.046423076597919784
+    },
+    "custom|aimo_kaggle_hard_pot:_average|0": {
+      "qem": 0.06666666666666667,
+      "qem_stderr": 0.03344992275839032
+    },
+    "all": {
+      "qem": 0.06666666666666667,
+      "qem_stderr": 0.03344992275839032
+    }
+  },
+  "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": "bfd073e62f180246",
+        "hash_input_tokens": "2e8d8fe2c6d12abc",
+        "hash_cont_tokens": "1b279a63e82b48a8"
+      },
+      "truncated": 50,
+      "non_truncated": 0,
+      "padded": 35,
+      "non_padded": 15,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|aimo_kaggle_hard_pot:v1|0": {
+      "hashes": {
+        "hash_examples": "e4234b97ad92862f",
+        "hash_full_prompts": "4747f0cd9a10355c",
+        "hash_input_tokens": "837c14488ae0fba1",
+        "hash_cont_tokens": "39472db4e3db23e1"
+      },
+      "truncated": 50,
+      "non_truncated": 0,
+      "padded": 26,
+      "non_padded": 24,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|aimo_kaggle_hard_pot:v2|0": {
+      "hashes": {
+        "hash_examples": "6396eb8833e13ba0",
+        "hash_full_prompts": "0584a2707d3a5d56",
+        "hash_input_tokens": "e16a853ef5822c84",
+        "hash_cont_tokens": "32a5d6d1a9b9244a"
+      },
+      "truncated": 50,
+      "non_truncated": 0,
+      "padded": 34,
+      "non_padded": 16,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "648c9a107d279e1e",
+      "hash_full_prompts": "c13504bf8e62491b",
+      "hash_input_tokens": "115bce65a38d1991",
+      "hash_cont_tokens": "f1533e4d65056a8c"
+    },
+    "truncated": 150,
+    "non_truncated": 0,
+    "padded": 95,
+    "non_padded": 55,
+    "num_truncated_few_shots": 0
+  }
+}