chore: upload lm-eval-harness results
Browse files
evals/results_2026-04-15T13-55-33.588896.json
ADDED
|
@@ -0,0 +1,161 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"results": {
|
| 3 |
+
"gsm8k": {
|
| 4 |
+
"alias": "gsm8k",
|
| 5 |
+
"exact_match,strict-match": 0.844,
|
| 6 |
+
"exact_match_stderr,strict-match": 0.016243636028391094,
|
| 7 |
+
"exact_match,flexible-extract": 0.892,
|
| 8 |
+
"exact_match_stderr,flexible-extract": 0.013894535480989173
|
| 9 |
+
}
|
| 10 |
+
},
|
| 11 |
+
"group_subtasks": {
|
| 12 |
+
"gsm8k": []
|
| 13 |
+
},
|
| 14 |
+
"configs": {
|
| 15 |
+
"gsm8k": {
|
| 16 |
+
"task": "gsm8k",
|
| 17 |
+
"tag": [
|
| 18 |
+
"math_word_problems"
|
| 19 |
+
],
|
| 20 |
+
"dataset_path": "openai/gsm8k",
|
| 21 |
+
"dataset_name": "main",
|
| 22 |
+
"training_split": "train",
|
| 23 |
+
"test_split": "test",
|
| 24 |
+
"fewshot_split": "train",
|
| 25 |
+
"doc_to_text": "Question: {{question}}\nAnswer:",
|
| 26 |
+
"doc_to_target": "{{answer}}",
|
| 27 |
+
"unsafe_code": false,
|
| 28 |
+
"description": "",
|
| 29 |
+
"target_delimiter": " ",
|
| 30 |
+
"fewshot_delimiter": "\n\n",
|
| 31 |
+
"fewshot_config": {
|
| 32 |
+
"sampler": "default",
|
| 33 |
+
"split": "train",
|
| 34 |
+
"process_docs": null,
|
| 35 |
+
"fewshot_indices": null,
|
| 36 |
+
"samples": null,
|
| 37 |
+
"doc_to_text": "Question: {{question}}\nAnswer:",
|
| 38 |
+
"doc_to_choice": null,
|
| 39 |
+
"doc_to_target": "{{answer}}",
|
| 40 |
+
"gen_prefix": null,
|
| 41 |
+
"fewshot_delimiter": "\n\n",
|
| 42 |
+
"target_delimiter": " "
|
| 43 |
+
},
|
| 44 |
+
"num_fewshot": 5,
|
| 45 |
+
"metric_list": [
|
| 46 |
+
{
|
| 47 |
+
"metric": "exact_match",
|
| 48 |
+
"aggregation": "mean",
|
| 49 |
+
"higher_is_better": true,
|
| 50 |
+
"ignore_case": true,
|
| 51 |
+
"ignore_punctuation": false,
|
| 52 |
+
"regexes_to_ignore": [
|
| 53 |
+
",",
|
| 54 |
+
"\\$",
|
| 55 |
+
"(?s).*#### ",
|
| 56 |
+
"\\.$"
|
| 57 |
+
]
|
| 58 |
+
}
|
| 59 |
+
],
|
| 60 |
+
"output_type": "generate_until",
|
| 61 |
+
"generation_kwargs": {
|
| 62 |
+
"until": [
|
| 63 |
+
"Question:",
|
| 64 |
+
"</s>",
|
| 65 |
+
"<|im_end|>"
|
| 66 |
+
],
|
| 67 |
+
"do_sample": false,
|
| 68 |
+
"temperature": 0.0
|
| 69 |
+
},
|
| 70 |
+
"repeats": 1,
|
| 71 |
+
"filter_list": [
|
| 72 |
+
{
|
| 73 |
+
"name": "strict-match",
|
| 74 |
+
"filter": [
|
| 75 |
+
{
|
| 76 |
+
"function": "regex",
|
| 77 |
+
"regex_pattern": "#### (\\-?[0-9\\.\\,]+)"
|
| 78 |
+
},
|
| 79 |
+
{
|
| 80 |
+
"function": "take_first"
|
| 81 |
+
}
|
| 82 |
+
]
|
| 83 |
+
},
|
| 84 |
+
{
|
| 85 |
+
"name": "flexible-extract",
|
| 86 |
+
"filter": [
|
| 87 |
+
{
|
| 88 |
+
"function": "regex",
|
| 89 |
+
"group_select": -1,
|
| 90 |
+
"regex_pattern": "(-?[$0-9.,]{2,})|(-?[0-9]+)"
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"function": "take_first"
|
| 94 |
+
}
|
| 95 |
+
]
|
| 96 |
+
}
|
| 97 |
+
],
|
| 98 |
+
"should_decontaminate": false,
|
| 99 |
+
"metadata": {
|
| 100 |
+
"version": 3.0,
|
| 101 |
+
"base_url": "http://localhost:8000/v1/chat/completions",
|
| 102 |
+
"model": "devstral",
|
| 103 |
+
"num_concurrent": 1
|
| 104 |
+
}
|
| 105 |
+
}
|
| 106 |
+
},
|
| 107 |
+
"versions": {
|
| 108 |
+
"gsm8k": 3.0
|
| 109 |
+
},
|
| 110 |
+
"n-shot": {
|
| 111 |
+
"gsm8k": 5
|
| 112 |
+
},
|
| 113 |
+
"higher_is_better": {
|
| 114 |
+
"gsm8k": {
|
| 115 |
+
"exact_match": true
|
| 116 |
+
}
|
| 117 |
+
},
|
| 118 |
+
"n-samples": {
|
| 119 |
+
"gsm8k": {
|
| 120 |
+
"original": 1319,
|
| 121 |
+
"effective": 500
|
| 122 |
+
}
|
| 123 |
+
},
|
| 124 |
+
"config": {
|
| 125 |
+
"model": "local-chat-completions",
|
| 126 |
+
"model_args": {
|
| 127 |
+
"base_url": "http://localhost:8000/v1/chat/completions",
|
| 128 |
+
"model": "devstral",
|
| 129 |
+
"num_concurrent": 1
|
| 130 |
+
},
|
| 131 |
+
"batch_size": "1",
|
| 132 |
+
"batch_sizes": [],
|
| 133 |
+
"device": "cuda:0",
|
| 134 |
+
"use_cache": null,
|
| 135 |
+
"limit": 500.0,
|
| 136 |
+
"bootstrap_iters": 100000,
|
| 137 |
+
"gen_kwargs": {},
|
| 138 |
+
"random_seed": 0,
|
| 139 |
+
"numpy_seed": 1234,
|
| 140 |
+
"torch_seed": 1234,
|
| 141 |
+
"fewshot_seed": 1234
|
| 142 |
+
},
|
| 143 |
+
"git_hash": null,
|
| 144 |
+
"date": 1776252935.2190523,
|
| 145 |
+
"pretty_env_info": "N/A (torch not installed)",
|
| 146 |
+
"transformers_version": "N/A",
|
| 147 |
+
"lm_eval_version": "0.4.11",
|
| 148 |
+
"upper_git_hash": null,
|
| 149 |
+
"task_hashes": {
|
| 150 |
+
"gsm8k": "d7e2c3373f72b8e34c540589c2c10df09e2f7b703350791f9e62af69c7f59ab0"
|
| 151 |
+
},
|
| 152 |
+
"model_source": "local-chat-completions",
|
| 153 |
+
"model_name": "devstral",
|
| 154 |
+
"model_name_sanitized": "devstral",
|
| 155 |
+
"system_instruction": null,
|
| 156 |
+
"system_instruction_sha": null,
|
| 157 |
+
"fewshot_as_multiturn": true,
|
| 158 |
+
"chat_template": "",
|
| 159 |
+
"chat_template_sha": null,
|
| 160 |
+
"total_evaluation_time_seconds": "1199.729275439866"
|
| 161 |
+
}
|