Honest OOD FineWeb benchmark: lead AR +13.5% vs GPT-4o, lose EN by 6.2%
Browse files- fair_benchmark_results.json +170 -0
fair_benchmark_results.json
ADDED
|
@@ -0,0 +1,170 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"corpus": {
|
| 3 |
+
"ar_source": "HuggingFaceFW/fineweb-2/arb_Arab",
|
| 4 |
+
"en_source": "HuggingFaceFW/fineweb-edu/sample-10BT",
|
| 5 |
+
"num_samples_per_lang": 300,
|
| 6 |
+
"max_chars": 2000,
|
| 7 |
+
"ar_chars": 400273,
|
| 8 |
+
"en_chars": 507749,
|
| 9 |
+
"ar_words": 68345,
|
| 10 |
+
"en_words": 82573
|
| 11 |
+
},
|
| 12 |
+
"results": [
|
| 13 |
+
{
|
| 14 |
+
"name": "SARFTokenizer v0.2",
|
| 15 |
+
"vocab": 65000,
|
| 16 |
+
"ar_cpt": 3.597183528946564,
|
| 17 |
+
"en_cpt": 4.514207222745782,
|
| 18 |
+
"parity": 0.7968583078821457,
|
| 19 |
+
"ar_tpw": 1.6281220279464481,
|
| 20 |
+
"en_tpw": 1.3621643878749712,
|
| 21 |
+
"ar_tokens": 111274,
|
| 22 |
+
"en_tokens": 112478,
|
| 23 |
+
"encode_time": 0.83054518699646
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"name": "tiktoken/o200k_base",
|
| 27 |
+
"vocab": 200019,
|
| 28 |
+
"ar_cpt": 3.1684714636270086,
|
| 29 |
+
"en_cpt": 4.796329183276341,
|
| 30 |
+
"parity": 0.660603420356283,
|
| 31 |
+
"ar_tpw": 1.8484161240763772,
|
| 32 |
+
"en_tpw": 1.2820413452339143,
|
| 33 |
+
"ar_tokens": 126330,
|
| 34 |
+
"en_tokens": 105862,
|
| 35 |
+
"encode_time": 0.25562620162963867
|
| 36 |
+
},
|
| 37 |
+
{
|
| 38 |
+
"name": "tiktoken/cl100k_base",
|
| 39 |
+
"vocab": 100277,
|
| 40 |
+
"ar_cpt": 1.4619227973805602,
|
| 41 |
+
"en_cpt": 4.741108361734908,
|
| 42 |
+
"parity": 0.30835042902195986,
|
| 43 |
+
"ar_tpw": 4.006130660618918,
|
| 44 |
+
"en_tpw": 1.2969735870078598,
|
| 45 |
+
"ar_tokens": 273799,
|
| 46 |
+
"en_tokens": 107095,
|
| 47 |
+
"encode_time": 0.17181658744812012
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"name": "Qwen/Qwen2.5-0.5B",
|
| 51 |
+
"vocab": 151665,
|
| 52 |
+
"ar_cpt": 2.6675619118705516,
|
| 53 |
+
"en_cpt": 4.649290358025822,
|
| 54 |
+
"parity": 0.5737567900584402,
|
| 55 |
+
"ar_tpw": 2.195508083985661,
|
| 56 |
+
"en_tpw": 1.3225872864011239,
|
| 57 |
+
"ar_tokens": 150052,
|
| 58 |
+
"en_tokens": 109210,
|
| 59 |
+
"encode_time": 0.7926328182220459
|
| 60 |
+
},
|
| 61 |
+
{
|
| 62 |
+
"name": "Qwen/Qwen3.6-35B-A3B",
|
| 63 |
+
"vocab": 248077,
|
| 64 |
+
"ar_cpt": 3.229727112818113,
|
| 65 |
+
"en_cpt": 4.595803803368905,
|
| 66 |
+
"parity": 0.7027556551588638,
|
| 67 |
+
"ar_tpw": 1.8133586948569755,
|
| 68 |
+
"en_tpw": 1.3379797270294165,
|
| 69 |
+
"ar_tokens": 123934,
|
| 70 |
+
"en_tokens": 110481,
|
| 71 |
+
"encode_time": 0.7811288833618164
|
| 72 |
+
},
|
| 73 |
+
{
|
| 74 |
+
"name": "google/gemma-2-2b",
|
| 75 |
+
"vocab": 256000,
|
| 76 |
+
"ar_cpt": 2.8638181571020755,
|
| 77 |
+
"en_cpt": 4.679024291348741,
|
| 78 |
+
"parity": 0.6120545606905948,
|
| 79 |
+
"ar_tpw": 2.0450508449776867,
|
| 80 |
+
"en_tpw": 1.3141826020612064,
|
| 81 |
+
"ar_tokens": 139769,
|
| 82 |
+
"en_tokens": 108516,
|
| 83 |
+
"encode_time": 0.5027248859405518
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"name": "google/gemma-3-1b-pt",
|
| 87 |
+
"vocab": 262145,
|
| 88 |
+
"ar_cpt": 2.9128346565563215,
|
| 89 |
+
"en_cpt": 4.660299948601219,
|
| 90 |
+
"parity": 0.6250315835208426,
|
| 91 |
+
"ar_tpw": 2.010637208281513,
|
| 92 |
+
"en_tpw": 1.3194627783900306,
|
| 93 |
+
"ar_tokens": 137417,
|
| 94 |
+
"en_tokens": 108952,
|
| 95 |
+
"encode_time": 0.49225592613220215
|
| 96 |
+
},
|
| 97 |
+
{
|
| 98 |
+
"name": "google/gemma-4-31B-it",
|
| 99 |
+
"vocab": 262144,
|
| 100 |
+
"ar_cpt": 2.9128346565563215,
|
| 101 |
+
"en_cpt": 4.660299948601219,
|
| 102 |
+
"parity": 0.6250315835208426,
|
| 103 |
+
"ar_tpw": 2.010637208281513,
|
| 104 |
+
"en_tpw": 1.3194627783900306,
|
| 105 |
+
"ar_tokens": 137417,
|
| 106 |
+
"en_tokens": 108952,
|
| 107 |
+
"encode_time": 0.47115373611450195
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"name": "tiiuae/falcon-7b",
|
| 111 |
+
"vocab": 65024,
|
| 112 |
+
"ar_cpt": 1.0156944641805892,
|
| 113 |
+
"en_cpt": 4.558954513620773,
|
| 114 |
+
"parity": 0.22279109413046394,
|
| 115 |
+
"ar_tpw": 5.766156997585778,
|
| 116 |
+
"en_tpw": 1.3487944001065724,
|
| 117 |
+
"ar_tokens": 394088,
|
| 118 |
+
"en_tokens": 111374,
|
| 119 |
+
"encode_time": 1.0502738952636719
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"name": "ALLaM-AI/ALLaM-7B-Instruct-preview",
|
| 123 |
+
"vocab": 64000,
|
| 124 |
+
"ar_cpt": 2.928541117939713,
|
| 125 |
+
"en_cpt": 3.4529473369239976,
|
| 126 |
+
"parity": 0.8481279417798931,
|
| 127 |
+
"ar_tpw": 1.9998536835174483,
|
| 128 |
+
"en_tpw": 1.780824240369128,
|
| 129 |
+
"ar_tokens": 136680,
|
| 130 |
+
"en_tokens": 147048,
|
| 131 |
+
"encode_time": 0.3760392665863037
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"name": "QCRI/Fanar-1-9B-Instruct",
|
| 135 |
+
"vocab": 128256,
|
| 136 |
+
"ar_cpt": 2.861709277053306,
|
| 137 |
+
"en_cpt": 4.564241089487168,
|
| 138 |
+
"parity": 0.6269846883515182,
|
| 139 |
+
"ar_tpw": 2.04655790474797,
|
| 140 |
+
"en_tpw": 1.3472321461010257,
|
| 141 |
+
"ar_tokens": 139872,
|
| 142 |
+
"en_tokens": 111245,
|
| 143 |
+
"encode_time": 0.46768975257873535
|
| 144 |
+
},
|
| 145 |
+
{
|
| 146 |
+
"name": "hammh0a/Hala-350M",
|
| 147 |
+
"vocab": 64400,
|
| 148 |
+
"ar_cpt": 2.2519508284339924,
|
| 149 |
+
"en_cpt": 4.648353962208876,
|
| 150 |
+
"parity": 0.48446199380304417,
|
| 151 |
+
"ar_tpw": 2.6007023191162486,
|
| 152 |
+
"en_tpw": 1.3228537173167985,
|
| 153 |
+
"ar_tokens": 177745,
|
| 154 |
+
"en_tokens": 109232,
|
| 155 |
+
"encode_time": 0.6467485427856445
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"name": "moonshotai/Kimi-K2.6",
|
| 159 |
+
"vocab": 163840,
|
| 160 |
+
"ar_cpt": 2.114814208198826,
|
| 161 |
+
"en_cpt": 4.74310135450724,
|
| 162 |
+
"parity": 0.4458716038587655,
|
| 163 |
+
"ar_tpw": 2.7693466969054064,
|
| 164 |
+
"en_tpw": 1.2964286146803434,
|
| 165 |
+
"ar_tokens": 189271,
|
| 166 |
+
"en_tokens": 107050,
|
| 167 |
+
"encode_time": 3.198082685470581
|
| 168 |
+
}
|
| 169 |
+
]
|
| 170 |
+
}
|