every_eval_ever/gsm8k_platinum.json
| 1 | { |
| 2 | "schema_version": "0.2.2", |
| 3 | "evaluation_id": "gsm8k_platinum_cot_llama/RedHatAI/gemma-4-31B-it-FP8-block/1781646636.119754", |
| 4 | "evaluation_timestamp": "1781532932", |
| 5 | "retrieved_timestamp": "1781646636.119754", |
| 6 | "source_metadata": { |
| 7 | "source_name": "lm-evaluation-harness", |
| 8 | "source_type": "evaluation_run", |
| 9 | "source_organization_name": "RedHatAI", |
| 10 | "evaluator_relationship": "third_party" |
| 11 | }, |
| 12 | "eval_library": { |
| 13 | "name": "lm_eval", |
| 14 | "version": "0.4.13.dev0" |
| 15 | }, |
| 16 | "model_info": { |
| 17 | "name": "RedHatAI/gemma-4-31B-it-FP8-block", |
| 18 | "id": "RedHatAI/gemma-4-31B-it-FP8-block", |
| 19 | "developer": "RedHatAI", |
| 20 | "additional_details": { |
| 21 | "model_args": "{'model': 'RedHatAI/gemma-4-31B-it-FP8-block', 'max_length': 69632, 'base_url': 'http://127.0.0.1:8000/v1/chat/completions', 'num_concurrent': 32, 'max_retries': 3, 'tokenized_requests': False, 'tokenizer_backend': None, 'timeout': 3600}", |
| 22 | "num_seeds_merged": "3" |
| 23 | } |
| 24 | }, |
| 25 | "evaluation_results": [ |
| 26 | { |
| 27 | "evaluation_name": "gsm8k_platinum_cot_llama/strict-match", |
| 28 | "source_data": { |
| 29 | "dataset_name": "gsm8k_platinum_cot_llama", |
| 30 | "source_type": "hf_dataset", |
| 31 | "hf_repo": "madrylab/gsm8k-platinum", |
| 32 | "hf_split": "test" |
| 33 | }, |
| 34 | "metric_config": { |
| 35 | "evaluation_description": "exact_match (filter: strict-match)", |
| 36 | "lower_is_better": false, |
| 37 | "score_type": "continuous", |
| 38 | "min_score": 0.0, |
| 39 | "max_score": 1.0 |
| 40 | }, |
| 41 | "score_details": { |
| 42 | "score": 0.9578163771712159, |
| 43 | "details": { |
| 44 | "seed_scores": "[0.9586435070306039, 0.956989247311828, 0.9578163771712159]", |
| 45 | "evaluation_timestamps": "[1781532932, 1781533049, 1781533165]", |
| 46 | "seed_values": "[1234, 2345, 3456]" |
| 47 | }, |
| 48 | "uncertainty": { |
| 49 | "standard_error": { |
| 50 | "value": 0.00047754364697239973, |
| 51 | "method": "across_seeds" |
| 52 | }, |
| 53 | "num_samples": 3 |
| 54 | } |
| 55 | }, |
| 56 | "generation_config": { |
| 57 | "generation_args": { |
| 58 | "temperature": 1.0, |
| 59 | "top_p": 0.95, |
| 60 | "top_k": 64.0, |
| 61 | "max_tokens": 32000, |
| 62 | "max_attempts": 1 |
| 63 | }, |
| 64 | "additional_details": { |
| 65 | "do_sample": "true", |
| 66 | "until": "[\"<|eot_id|>\", \"<|start_header_id|>user<|end_header_id|>\", \"Q:\", \"</s>\", \"<|im_end|>\"]", |
| 67 | "seed": "1234", |
| 68 | "num_fewshot": "0" |
| 69 | } |
| 70 | } |
| 71 | }, |
| 72 | { |
| 73 | "evaluation_name": "gsm8k_platinum_cot_llama/flexible-extract", |
| 74 | "source_data": { |
| 75 | "dataset_name": "gsm8k_platinum_cot_llama", |
| 76 | "source_type": "hf_dataset", |
| 77 | "hf_repo": "madrylab/gsm8k-platinum", |
| 78 | "hf_split": "test" |
| 79 | }, |
| 80 | "metric_config": { |
| 81 | "evaluation_description": "exact_match (filter: flexible-extract)", |
| 82 | "lower_is_better": false, |
| 83 | "score_type": "continuous", |
| 84 | "min_score": 0.0, |
| 85 | "max_score": 1.0 |
| 86 | }, |
| 87 | "score_details": { |
| 88 | "score": 0.9583677970774745, |
| 89 | "details": { |
| 90 | "seed_scores": "[0.9586435070306039, 0.9578163771712159, 0.9586435070306039]", |
| 91 | "evaluation_timestamps": "[1781532932, 1781533049, 1781533165]", |
| 92 | "seed_values": "[1234, 2345, 3456]" |
| 93 | }, |
| 94 | "uncertainty": { |
| 95 | "standard_error": { |
| 96 | "value": 0.00027570995312931057, |
| 97 | "method": "across_seeds" |
| 98 | }, |
| 99 | "num_samples": 3 |
| 100 | } |
| 101 | }, |
| 102 | "generation_config": { |
| 103 | "generation_args": { |
| 104 | "temperature": 1.0, |
| 105 | "top_p": 0.95, |
| 106 | "top_k": 64.0, |
| 107 | "max_tokens": 32000, |
| 108 | "max_attempts": 1 |
| 109 | }, |
| 110 | "additional_details": { |
| 111 | "do_sample": "true", |
| 112 | "until": "[\"<|eot_id|>\", \"<|start_header_id|>user<|end_header_id|>\", \"Q:\", \"</s>\", \"<|im_end|>\"]", |
| 113 | "seed": "1234", |
| 114 | "num_fewshot": "0" |
| 115 | } |
| 116 | } |
| 117 | } |
| 118 | ] |
| 119 | } |