Upload eval_results/open-r1/DeepSeek-R1-Distill-Qwen-1.5B-GRPO-LIMO-v02.01/main-step-000001000/math_500/results_2025-02-08T07-04-32.301817.json with huggingface_hub
Browse files
eval_results/open-r1/DeepSeek-R1-Distill-Qwen-1.5B-GRPO-LIMO-v02.01/main-step-000001000/math_500/results_2025-02-08T07-04-32.301817.json
ADDED
|
@@ -0,0 +1,98 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config_general": {
|
| 3 |
+
"lighteval_sha": "?",
|
| 4 |
+
"num_fewshot_seeds": 1,
|
| 5 |
+
"override_batch_size": -1,
|
| 6 |
+
"max_samples": null,
|
| 7 |
+
"job_id": 0,
|
| 8 |
+
"start_time": 294783.918339334,
|
| 9 |
+
"end_time": 295273.188057696,
|
| 10 |
+
"total_evaluation_time_secondes": "489.26971836196026",
|
| 11 |
+
"model_name": "open-r1/DeepSeek-R1-Distill-Qwen-1.5B-GRPO-LIMO-v02.01",
|
| 12 |
+
"model_sha": "",
|
| 13 |
+
"model_dtype": null,
|
| 14 |
+
"model_size": null
|
| 15 |
+
},
|
| 16 |
+
"results": {
|
| 17 |
+
"custom|math_500|0": {
|
| 18 |
+
"extractive_match": 0.77,
|
| 19 |
+
"extractive_match_stderr": 0.018839050391123147
|
| 20 |
+
},
|
| 21 |
+
"all": {
|
| 22 |
+
"extractive_match": 0.77,
|
| 23 |
+
"extractive_match_stderr": 0.018839050391123147
|
| 24 |
+
}
|
| 25 |
+
},
|
| 26 |
+
"versions": {
|
| 27 |
+
"custom|math_500|0": 1
|
| 28 |
+
},
|
| 29 |
+
"config_tasks": {
|
| 30 |
+
"custom|math_500": {
|
| 31 |
+
"name": "math_500",
|
| 32 |
+
"prompt_function": "prompt_fn",
|
| 33 |
+
"hf_repo": "HuggingFaceH4/MATH-500",
|
| 34 |
+
"hf_subset": "default",
|
| 35 |
+
"metric": [
|
| 36 |
+
{
|
| 37 |
+
"metric_name": "extractive_match",
|
| 38 |
+
"higher_is_better": true,
|
| 39 |
+
"category": "3",
|
| 40 |
+
"use_case": "1",
|
| 41 |
+
"sample_level_fn": "sample_level_fn",
|
| 42 |
+
"corpus_level_fn": "mean"
|
| 43 |
+
}
|
| 44 |
+
],
|
| 45 |
+
"hf_revision": null,
|
| 46 |
+
"hf_filter": null,
|
| 47 |
+
"hf_avail_splits": [
|
| 48 |
+
"test"
|
| 49 |
+
],
|
| 50 |
+
"trust_dataset": false,
|
| 51 |
+
"evaluation_splits": [
|
| 52 |
+
"test"
|
| 53 |
+
],
|
| 54 |
+
"few_shots_split": null,
|
| 55 |
+
"few_shots_select": null,
|
| 56 |
+
"generation_size": 32768,
|
| 57 |
+
"generation_grammar": null,
|
| 58 |
+
"stop_sequence": [],
|
| 59 |
+
"num_samples": null,
|
| 60 |
+
"suite": [
|
| 61 |
+
"custom"
|
| 62 |
+
],
|
| 63 |
+
"original_num_docs": 500,
|
| 64 |
+
"effective_num_docs": 500,
|
| 65 |
+
"must_remove_duplicate_docs": false,
|
| 66 |
+
"version": 1
|
| 67 |
+
}
|
| 68 |
+
},
|
| 69 |
+
"summary_tasks": {
|
| 70 |
+
"custom|math_500|0": {
|
| 71 |
+
"hashes": {
|
| 72 |
+
"hash_examples": "eac05bd67b8179c3",
|
| 73 |
+
"hash_full_prompts": "9043592f69431f18",
|
| 74 |
+
"hash_input_tokens": "c5aa3a61e16cb62b",
|
| 75 |
+
"hash_cont_tokens": "62cb80732df428eb"
|
| 76 |
+
},
|
| 77 |
+
"truncated": 0,
|
| 78 |
+
"non_truncated": 500,
|
| 79 |
+
"padded": 0,
|
| 80 |
+
"non_padded": 500,
|
| 81 |
+
"effective_few_shots": 0.0,
|
| 82 |
+
"num_truncated_few_shots": 0
|
| 83 |
+
}
|
| 84 |
+
},
|
| 85 |
+
"summary_general": {
|
| 86 |
+
"hashes": {
|
| 87 |
+
"hash_examples": "d6b3f24200421bb2",
|
| 88 |
+
"hash_full_prompts": "1a4bc197befd9b91",
|
| 89 |
+
"hash_input_tokens": "b703f1639cd56c2a",
|
| 90 |
+
"hash_cont_tokens": "07072c0d54ec5fcf"
|
| 91 |
+
},
|
| 92 |
+
"truncated": 0,
|
| 93 |
+
"non_truncated": 500,
|
| 94 |
+
"padded": 0,
|
| 95 |
+
"non_padded": 500,
|
| 96 |
+
"num_truncated_few_shots": 0
|
| 97 |
+
}
|
| 98 |
+
}
|