Upload eval_results/AI-MO/CodeLlama-7b-Python-hf-sft/aimo_v01.20/aimo_kaggle_medium_pot/results_2024-06-02T17-49-44.917833.json with huggingface_hub

eval_results/AI-MO/CodeLlama-7b-Python-hf-sft/aimo_v01.20/aimo_kaggle_medium_pot/results_2024-06-02T17-49-44.917833.json

@@ -0,0 +1,193 @@
+{
+  "config_general": {
+    "lighteval_sha": "?",
+    "num_fewshot_seeds": 1,
+    "override_batch_size": 4,
+    "max_samples": null,
+    "job_id": "",
+    "start_time": 430831.734051874,
+    "end_time": 431275.77680128,
+    "total_evaluation_time_secondes": "444.04274940595496",
+    "model_name": "AI-MO/CodeLlama-7b-Python-hf-sft",
+    "model_sha": "a4b1a02cbd15a07cda4e5bf3f7f5d2dab94fa1db",
+    "model_dtype": "torch.bfloat16",
+    "model_size": "12.8 GB",
+    "config": null
+  },
+  "results": {
+    "custom|aimo_kaggle_medium_pot:v0|0": {
+      "score": 0.075,
+      "score_stderr": 0.04217636961434867
+    },
+    "custom|aimo_kaggle_medium_pot:v1|0": {
+      "score": 0.075,
+      "score_stderr": 0.04217636961434867
+    },
+    "custom|aimo_kaggle_medium_pot:v2|0": {
+      "score": 0.075,
+      "score_stderr": 0.04217636961434867
+    },
+    "custom|aimo_kaggle_medium_pot:_average|0": {
+      "score": 0.075,
+      "score_stderr": 0.04217636961434867
+    },
+    "all": {
+      "score": 0.075,
+      "score_stderr": 0.04217636961434867
+    }
+  },
+  "versions": {
+    "custom|aimo_kaggle_medium_pot:v0|0": 0,
+    "custom|aimo_kaggle_medium_pot:v1|0": 0,
+    "custom|aimo_kaggle_medium_pot:v2|0": 0
+  },
+  "config_tasks": {
+    "custom|aimo_kaggle_medium_pot:v0": {
+      "name": "aimo_kaggle_medium_pot:v0",
+      "prompt_function": "kaggle_medium_pot_prompt_fn_v0",
+      "hf_repo": "AI-MO/kaggle-validation-set-medium",
+      "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": 40,
+      "effective_num_docs": 40,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    },
+    "custom|aimo_kaggle_medium_pot:v1": {
+      "name": "aimo_kaggle_medium_pot:v1",
+      "prompt_function": "kaggle_medium_pot_prompt_fn_v1",
+      "hf_repo": "AI-MO/kaggle-validation-set-medium",
+      "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": 40,
+      "effective_num_docs": 40,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    },
+    "custom|aimo_kaggle_medium_pot:v2": {
+      "name": "aimo_kaggle_medium_pot:v2",
+      "prompt_function": "kaggle_medium_pot_prompt_fn_v2",
+      "hf_repo": "AI-MO/kaggle-validation-set-medium",
+      "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": 40,
+      "effective_num_docs": 40,
+      "trust_dataset": null,
+      "must_remove_duplicate_docs": null,
+      "version": 0
+    }
+  },
+  "summary_tasks": {
+    "custom|aimo_kaggle_medium_pot:v0|0": {
+      "hashes": {
+        "hash_examples": "2799c24461029dc3",
+        "hash_full_prompts": "2af864fbfc2e0a79",
+        "hash_input_tokens": "e868d382e92028d6",
+        "hash_cont_tokens": "43047b5d44706b24"
+      },
+      "truncated": 40,
+      "non_truncated": 0,
+      "padded": 29,
+      "non_padded": 11,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|aimo_kaggle_medium_pot:v1|0": {
+      "hashes": {
+        "hash_examples": "806b2e2056b41f84",
+        "hash_full_prompts": "8123a0d96a6ceb9d",
+        "hash_input_tokens": "44f80b07cbbc1202",
+        "hash_cont_tokens": "8884f1ca090130fc"
+      },
+      "truncated": 40,
+      "non_truncated": 0,
+      "padded": 28,
+      "non_padded": 12,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    },
+    "custom|aimo_kaggle_medium_pot:v2|0": {
+      "hashes": {
+        "hash_examples": "d8534375acc5d427",
+        "hash_full_prompts": "71ba7c8172fec45a",
+        "hash_input_tokens": "9fe3382a1eef295c",
+        "hash_cont_tokens": "47b120f15484e4ba"
+      },
+      "truncated": 40,
+      "non_truncated": 0,
+      "padded": 28,
+      "non_padded": 12,
+      "effective_few_shots": 0.0,
+      "num_truncated_few_shots": 0
+    }
+  },
+  "summary_general": {
+    "hashes": {
+      "hash_examples": "623505a45a4910c2",
+      "hash_full_prompts": "0ee7c8ef786b9aa3",
+      "hash_input_tokens": "89b21512945d231a",
+      "hash_cont_tokens": "102fda147903b2f2"
+    },
+    "truncated": 120,
+    "non_truncated": 0,
+    "padded": 85,
+    "non_padded": 35,
+    "num_truncated_few_shots": 0
+  }
+}