Update README.md
Browse files
README.md
CHANGED
|
@@ -8,10 +8,7 @@ language:
|
|
| 8 |
|
| 9 |
---
|
| 10 |
|
| 11 |
-
|
| 12 |
-
### Open LLM Leaderboard Average Score: 0.3057
|
| 13 |
-
|
| 14 |
-
This is just above base gpt2 only because of truthfulqa score bringing the average up, it has a higher truthfulqa score than any base gpt2 model. It is also just under pythia-160m for average score (0.01%) and
|
| 15 |
|
| 16 |
| model | avg | arc | hellaswag | mmlu | truthfulqa |
|
| 17 |
| --- | --- | --- | --- | --- | --- |
|
|
@@ -23,139 +20,5 @@ This is just above base gpt2 only because of truthfulqa score bringing the avera
|
|
| 23 |
| pythia 160m | 30.58 | 22.78 | 30.34 | 24.95 | 44.26 |
|
| 24 |
|
| 25 |
|
|
|
|
| 26 |
|
| 27 |
-
| Task |Version| Metric |Value | |Stderr|
|
| 28 |
-
|-------------|------:|--------|-----:|---|-----:|
|
| 29 |
-
|arc_challenge| 0|acc |0.1741|± |0.0111|
|
| 30 |
-
| | |acc_norm|**0.2176**|± |0.0121|
|
| 31 |
-
|
| 32 |
-
| Task |Version| Metric |Value | |Stderr|
|
| 33 |
-
|---------|------:|--------|-----:|---|-----:|
|
| 34 |
-
|hellaswag| 0|acc |0.2698|± |0.0044|
|
| 35 |
-
| | |acc_norm|**0.2735**|± |0.0044|
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
| Task |Version|Metric|Value | |Stderr|
|
| 39 |
-
|-------------|------:|------|-----:|---|-----:|
|
| 40 |
-
|truthfulqa_mc| 1|mc1 |0.2803|± |0.0157|
|
| 41 |
-
| | |mc2 |**0.4766**|± |0.0156|
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
| Task |Version| Metric |Value | |Stderr|
|
| 45 |
-
|-------------------------------------------------|------:|--------|-----:|---|-----:|
|
| 46 |
-
|hendrycksTest-abstract_algebra | 1|acc |0.2200|± |0.0416|
|
| 47 |
-
| | |acc_norm|0.2200|± |0.0416|
|
| 48 |
-
|hendrycksTest-anatomy | 1|acc |0.3333|± |0.0407|
|
| 49 |
-
| | |acc_norm|0.3333|± |0.0407|
|
| 50 |
-
|hendrycksTest-astronomy | 1|acc |0.2237|± |0.0339|
|
| 51 |
-
| | |acc_norm|0.2237|± |0.0339|
|
| 52 |
-
|hendrycksTest-business_ethics | 1|acc |0.2000|± |0.0402|
|
| 53 |
-
| | |acc_norm|0.2000|± |0.0402|
|
| 54 |
-
|hendrycksTest-clinical_knowledge | 1|acc |0.2189|± |0.0254|
|
| 55 |
-
| | |acc_norm|0.2189|± |0.0254|
|
| 56 |
-
|hendrycksTest-college_biology | 1|acc |0.2083|± |0.0340|
|
| 57 |
-
| | |acc_norm|0.2083|± |0.0340|
|
| 58 |
-
|hendrycksTest-college_chemistry | 1|acc |0.3400|± |0.0476|
|
| 59 |
-
| | |acc_norm|0.3400|± |0.0476|
|
| 60 |
-
|hendrycksTest-college_computer_science | 1|acc |0.3100|± |0.0465|
|
| 61 |
-
| | |acc_norm|0.3100|± |0.0465|
|
| 62 |
-
|hendrycksTest-college_mathematics | 1|acc |0.3100|± |0.0465|
|
| 63 |
-
| | |acc_norm|0.3100|± |0.0465|
|
| 64 |
-
|hendrycksTest-college_medicine | 1|acc |0.2197|± |0.0316|
|
| 65 |
-
| | |acc_norm|0.2197|± |0.0316|
|
| 66 |
-
|hendrycksTest-college_physics | 1|acc |0.3431|± |0.0472|
|
| 67 |
-
| | |acc_norm|0.3431|± |0.0472|
|
| 68 |
-
|hendrycksTest-computer_security | 1|acc |0.2000|± |0.0402|
|
| 69 |
-
| | |acc_norm|0.2000|± |0.0402|
|
| 70 |
-
|hendrycksTest-conceptual_physics | 1|acc |0.2809|± |0.0294|
|
| 71 |
-
| | |acc_norm|0.2809|± |0.0294|
|
| 72 |
-
|hendrycksTest-econometrics | 1|acc |0.2544|± |0.0410|
|
| 73 |
-
| | |acc_norm|0.2544|± |0.0410|
|
| 74 |
-
|hendrycksTest-electrical_engineering | 1|acc |0.2414|± |0.0357|
|
| 75 |
-
| | |acc_norm|0.2414|± |0.0357|
|
| 76 |
-
|hendrycksTest-elementary_mathematics | 1|acc |0.2566|± |0.0225|
|
| 77 |
-
| | |acc_norm|0.2566|± |0.0225|
|
| 78 |
-
|hendrycksTest-formal_logic | 1|acc |0.1825|± |0.0346|
|
| 79 |
-
| | |acc_norm|0.1825|± |0.0346|
|
| 80 |
-
|hendrycksTest-global_facts | 1|acc |0.2000|± |0.0402|
|
| 81 |
-
| | |acc_norm|0.2000|± |0.0402|
|
| 82 |
-
|hendrycksTest-high_school_biology | 1|acc |0.3161|± |0.0265|
|
| 83 |
-
| | |acc_norm|0.3161|± |0.0265|
|
| 84 |
-
|hendrycksTest-high_school_chemistry | 1|acc |0.2759|± |0.0314|
|
| 85 |
-
| | |acc_norm|0.2759|± |0.0314|
|
| 86 |
-
|hendrycksTest-high_school_computer_science | 1|acc |0.2400|± |0.0429|
|
| 87 |
-
| | |acc_norm|0.2400|± |0.0429|
|
| 88 |
-
|hendrycksTest-high_school_european_history | 1|acc |0.2909|± |0.0355|
|
| 89 |
-
| | |acc_norm|0.2909|± |0.0355|
|
| 90 |
-
|hendrycksTest-high_school_geography | 1|acc |0.3535|± |0.0341|
|
| 91 |
-
| | |acc_norm|0.3535|± |0.0341|
|
| 92 |
-
|hendrycksTest-high_school_government_and_politics| 1|acc |0.2280|± |0.0303|
|
| 93 |
-
| | |acc_norm|0.2280|± |0.0303|
|
| 94 |
-
|hendrycksTest-high_school_macroeconomics | 1|acc |0.2051|± |0.0205|
|
| 95 |
-
| | |acc_norm|0.2051|± |0.0205|
|
| 96 |
-
|hendrycksTest-high_school_mathematics | 1|acc |0.2630|± |0.0268|
|
| 97 |
-
| | |acc_norm|0.2630|± |0.0268|
|
| 98 |
-
|hendrycksTest-high_school_microeconomics | 1|acc |0.3403|± |0.0308|
|
| 99 |
-
| | |acc_norm|0.3403|± |0.0308|
|
| 100 |
-
|hendrycksTest-high_school_physics | 1|acc |0.2384|± |0.0348|
|
| 101 |
-
| | |acc_norm|0.2384|± |0.0348|
|
| 102 |
-
|hendrycksTest-high_school_psychology | 1|acc |0.2257|± |0.0179|
|
| 103 |
-
| | |acc_norm|0.2257|± |0.0179|
|
| 104 |
-
|hendrycksTest-high_school_statistics | 1|acc |0.4722|± |0.0340|
|
| 105 |
-
| | |acc_norm|0.4722|± |0.0340|
|
| 106 |
-
|hendrycksTest-high_school_us_history | 1|acc |0.2206|± |0.0291|
|
| 107 |
-
| | |acc_norm|0.2206|± |0.0291|
|
| 108 |
-
|hendrycksTest-high_school_world_history | 1|acc |0.2658|± |0.0288|
|
| 109 |
-
| | |acc_norm|0.2658|± |0.0288|
|
| 110 |
-
|hendrycksTest-human_aging | 1|acc |0.2063|± |0.0272|
|
| 111 |
-
| | |acc_norm|0.2063|± |0.0272|
|
| 112 |
-
|hendrycksTest-human_sexuality | 1|acc |0.2366|± |0.0373|
|
| 113 |
-
| | |acc_norm|0.2366|± |0.0373|
|
| 114 |
-
|hendrycksTest-international_law | 1|acc |0.2562|± |0.0398|
|
| 115 |
-
| | |acc_norm|0.2562|± |0.0398|
|
| 116 |
-
|hendrycksTest-jurisprudence | 1|acc |0.2130|± |0.0396|
|
| 117 |
-
| | |acc_norm|0.2130|± |0.0396|
|
| 118 |
-
|hendrycksTest-logical_fallacies | 1|acc |0.2393|± |0.0335|
|
| 119 |
-
| | |acc_norm|0.2393|± |0.0335|
|
| 120 |
-
|hendrycksTest-machine_learning | 1|acc |0.2054|± |0.0383|
|
| 121 |
-
| | |acc_norm|0.2054|± |0.0383|
|
| 122 |
-
|hendrycksTest-management | 1|acc |0.1942|± |0.0392|
|
| 123 |
-
| | |acc_norm|0.1942|± |0.0392|
|
| 124 |
-
|hendrycksTest-marketing | 1|acc |0.1923|± |0.0258|
|
| 125 |
-
| | |acc_norm|0.1923|± |0.0258|
|
| 126 |
-
|hendrycksTest-medical_genetics | 1|acc |0.3000|± |0.0461|
|
| 127 |
-
| | |acc_norm|0.3000|± |0.0461|
|
| 128 |
-
|hendrycksTest-miscellaneous | 1|acc |0.2708|± |0.0159|
|
| 129 |
-
| | |acc_norm|0.2708|± |0.0159|
|
| 130 |
-
|hendrycksTest-moral_disputes | 1|acc |0.2168|± |0.0222|
|
| 131 |
-
| | |acc_norm|0.2168|± |0.0222|
|
| 132 |
-
|hendrycksTest-moral_scenarios | 1|acc |0.2313|± |0.0141|
|
| 133 |
-
| | |acc_norm|0.2313|± |0.0141|
|
| 134 |
-
|hendrycksTest-nutrition | 1|acc |0.2222|± |0.0238|
|
| 135 |
-
| | |acc_norm|0.2222|± |0.0238|
|
| 136 |
-
|hendrycksTest-philosophy | 1|acc |0.2315|± |0.0240|
|
| 137 |
-
| | |acc_norm|0.2315|± |0.0240|
|
| 138 |
-
|hendrycksTest-prehistory | 1|acc |0.2963|± |0.0254|
|
| 139 |
-
| | |acc_norm|0.2963|± |0.0254|
|
| 140 |
-
|hendrycksTest-professional_accounting | 1|acc |0.2589|± |0.0261|
|
| 141 |
-
| | |acc_norm|0.2589|± |0.0261|
|
| 142 |
-
|hendrycksTest-professional_law | 1|acc |0.2490|± |0.0110|
|
| 143 |
-
| | |acc_norm|0.2490|± |0.0110|
|
| 144 |
-
|hendrycksTest-professional_medicine | 1|acc |0.4375|± |0.0301|
|
| 145 |
-
| | |acc_norm|0.4375|± |0.0301|
|
| 146 |
-
|hendrycksTest-professional_psychology | 1|acc |0.2271|± |0.0169|
|
| 147 |
-
| | |acc_norm|0.2271|± |0.0169|
|
| 148 |
-
|hendrycksTest-public_relations | 1|acc |0.2455|± |0.0412|
|
| 149 |
-
| | |acc_norm|0.2455|± |0.0412|
|
| 150 |
-
|hendrycksTest-security_studies | 1|acc |0.2367|± |0.0272|
|
| 151 |
-
| | |acc_norm|0.2367|± |0.0272|
|
| 152 |
-
|hendrycksTest-sociology | 1|acc |0.2438|± |0.0304|
|
| 153 |
-
| | |acc_norm|0.2438|± |0.0304|
|
| 154 |
-
|hendrycksTest-us_foreign_policy | 1|acc |0.2900|± |0.0456|
|
| 155 |
-
| | |acc_norm|0.2900|± |0.0456|
|
| 156 |
-
|hendrycksTest-virology | 1|acc |0.1928|± |0.0307|
|
| 157 |
-
| | |acc_norm|0.1928|± |0.0307|
|
| 158 |
-
|hendrycksTest-world_religions | 1|acc |0.1813|± |0.0295|
|
| 159 |
-
| | |acc_norm|0.1813|± |0.0295|
|
| 160 |
-
|
| 161 |
-
average mmlu is 0.2553175438596491 ??
|
|
|
|
| 8 |
|
| 9 |
---
|
| 10 |
|
| 11 |
+
A modified GPT-2 model with only 25 million non-embedding params that outbenches GPT-2(124m), Pythia-70m/160m, and Cerebras-111m, it has ScaledSinusoidal position embeddings, embedding layernorm, no biases, and was trained on only 8 billion tokens of the SlimPajama dataset at home on 2xA6000.
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
| model | avg | arc | hellaswag | mmlu | truthfulqa |
|
| 14 |
| --- | --- | --- | --- | --- | --- |
|
|
|
|
| 20 |
| pythia 160m | 30.58 | 22.78 | 30.34 | 24.95 | 44.26 |
|
| 21 |
|
| 22 |
|
| 23 |
+
|
| 24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|