AksharaLP-MultiIndic-demo / benchmark.json
sourojitdotcom's picture
Private review: measured five-language release, Hindi sentence mode and custom GGUFs
5e783af verified
Raw History Blame Contribute Delete
7.9 kB
{
"languages": {
"hi": {
"ours": {
"n": 9991,
"top1": 70.04,
"top1_ci95": [
69.16,
70.9
],
"top3": 88.31,
"top8": 93.63
},
"indicxlit_full": {
"top1": 69.33,
"top1_ci95": [
68.44,
70.23
],
"top3": 85.4,
"top8": 88.29
},
"paired_ours_minus_indicxlit": {
"top1": [
0.71,
-0.04,
1.4
]
},
"download_mb": 7.905041
},
"bn": {
"ours": {
"n": 13367,
"top1": 64.64,
"top1_ci95": [
63.83,
65.44
],
"top3": 75.66,
"top3_ci95": [
74.9,
76.38
],
"top8": 77.97,
"top8_ci95": [
77.23,
78.66
]
},
"indicxlit_full": {
"n": 13367,
"top1": 70.52,
"top1_ci95": [
69.76,
71.34
],
"top3": 82.25,
"top3_ci95": [
81.61,
82.89
],
"top8": 85.44,
"top8_ci95": [
84.86,
86.06
]
},
"paired_ours_minus_indicxlit": {
"top1": [
-5.88,
-6.52,
-5.22
],
"top3": [
-6.59,
-7.18,
-5.97
],
"top8": [
-7.47,
-8.05,
-6.89
]
},
"frozen_sha256": {
"xlit.q4": "9f099f0b9524f446debf1d3f0ae1dc7ed1e88625c33f6dbb7b45e0348bf76376",
"dict.tsv.gz": "edb2ad0bd7de4e87bd4d49ac406a9ef01011451bf23ac82d4c063bb9cce1dc2b",
"params.json": "e0bd564ad61f6dcef8eca6bf3d64f8b231146c9885aa4c5b282a2b2420606c4f"
},
"test_rows_sha256": "de2dc6a45db144a32074c586b2b92f6af07e9b3b4fef495150d62c59c31c9dfd",
"baseline_predictions_sha256": "e3e72158035ae04c46691f75eb85df3b59959e9d801cd9099af26148cf7a5c9e",
"baseline_missing_empty_inputs": 1,
"baseline_note": "Existing IndicXlit PyTorch port beam-8 + full dictionary predictions, rescored on exactly the same rows; no new baseline inference.",
"size": {
"parameters": 989056,
"model_raw": 589040,
"model_gz9": 540107,
"dict_raw": 14068340,
"dict_gz9": 2508610,
"dict_words": 500000
},
"download_mb": 3.049313
},
"te": {
"ours": {
"n": 9936,
"top1": 78.74,
"top1_ci95": [
77.93,
79.53
],
"top3": 86.87,
"top3_ci95": [
86.2,
87.52
],
"top8": 88.63,
"top8_ci95": [
88.0,
89.25
]
},
"indicxlit_full": {
"n": 9936,
"top1": 82.42,
"top1_ci95": [
81.65,
83.14
],
"top3": 90.63,
"top3_ci95": [
90.04,
91.18
],
"top8": 92.75,
"top8_ci95": [
92.21,
93.26
]
},
"paired_ours_minus_indicxlit": {
"top1": [
-3.67,
-4.27,
-3.07
],
"top3": [
-3.76,
-4.27,
-3.21
],
"top8": [
-4.13,
-4.59,
-3.62
]
},
"frozen_sha256": {
"xlit.q4": "aea094e46a3c61eb9487d1aaba3824d3d9dc60f3f1cf6b6f99cadf04fd4318f9",
"dict.tsv.gz": "20612a52ccf39c816d3068e78e9e9a3aa9aed9156f10de45ebbd543363943640",
"params.json": "653f897efe1af78bb52149f2a18dc804b49ecac9c93add867871e64f56e07470"
},
"test_rows_sha256": "698de748186eb92a43baed29e1e0aa9a15acbd5b20a79b3a4fd7db4b336ceda1",
"baseline_predictions_sha256": "6c05dfc219ebc012fd96bac5f5b9586fd5ac35d0856a36af28c7362b04a31c82",
"baseline_note": "Existing IndicXlit PyTorch port beam-8 + full dictionary predictions, rescored on exactly the same rows; no new baseline inference.",
"size": {
"parameters": 988928,
"model_raw": 588904,
"model_gz9": 541242,
"dict_raw": 15989177,
"dict_gz9": 2680527,
"dict_words": 500000
},
"download_mb": 3.222365
},
"kn": {
"ours": {
"n": 11165,
"top1": 77.37,
"top1_ci95": [
76.61,
78.13
],
"top3": 85.8,
"top3_ci95": [
85.16,
86.46
],
"top8": 87.68,
"top8_ci95": [
87.08,
88.32
]
},
"indicxlit_full": {
"n": 11165,
"top1": 82.61,
"top1_ci95": [
81.92,
83.29
],
"top3": 90.68,
"top3_ci95": [
90.16,
91.2
],
"top8": 92.41,
"top8_ci95": [
91.94,
92.9
]
},
"paired_ours_minus_indicxlit": {
"top1": [
-5.24,
-5.83,
-4.67
],
"top3": [
-4.87,
-5.39,
-4.39
],
"top8": [
-4.73,
-5.19,
-4.26
]
},
"frozen_sha256": {
"xlit.q4": "3788fd545c297febb7dbbdbdd780e831042395b4ec99260b95c963607afc8664",
"dict.tsv.gz": "97035264d30eddb5810467dc5bbf088614221ea6ba2c3fff349c6bbd196bb644",
"params.json": "6076d9126ffce66630b634e47ac5b38d03cda3bce7ab4d07162aeb0cbc925356"
},
"test_rows_sha256": "5bfe000c741316d700f4caae046ed05c4cdcaf485ef31707f6254bcc4c9abbd9",
"baseline_predictions_sha256": "928c3375ffdbca7811a36284ac4afb70b8b5192d28aa35f88a6291f941373407",
"baseline_note": "Existing IndicXlit PyTorch port beam-8 + full dictionary predictions, rescored on exactly the same rows; no new baseline inference.",
"size": {
"parameters": 988928,
"model_raw": 588904,
"model_gz9": 539543,
"dict_raw": 16786145,
"dict_gz9": 2684083,
"dict_words": 500000
},
"download_mb": 3.224222
},
"ta": {
"ours": {
"n": 11405,
"top1": 71.78,
"top1_ci95": [
71.01,
72.61
],
"top3": 80.05,
"top3_ci95": [
79.37,
80.78
],
"top8": 82.61,
"top8_ci95": [
81.95,
83.31
]
},
"indicxlit_full": {
"n": 11405,
"top1": 75.95,
"top1_ci95": [
75.2,
76.73
],
"top3": 84.21,
"top3_ci95": [
83.56,
84.89
],
"top8": 87.26,
"top8_ci95": [
86.66,
87.87
]
},
"paired_ours_minus_indicxlit": {
"top1": [
-4.16,
-4.75,
-3.59
],
"top3": [
-4.16,
-4.66,
-3.63
],
"top8": [
-4.65,
-5.12,
-4.16
]
},
"frozen_sha256": {
"xlit.q4": "5dd185a033db79835fede66ab41a25f5897f24ad5d7c8ecc9864ca6ec167fdc0",
"dict.tsv.gz": "4d14f514020e7737144c22991e9324a29734b5ab203bef3b68ec6d9f51e7cddb",
"params.json": "f4520a8320f02573e6b8b6530a372c980903fbf757999540698957e8eec58385"
},
"test_rows_sha256": "464c11fec7cfa71680a01c57899aeb691d88f1a871b1c7976184b0a21e2d75a6",
"baseline_predictions_sha256": "0c2272e9423eb22e3fa04037a9973a777d50070cb9f3bcd03d4037865fd71bc3",
"baseline_note": "Existing IndicXlit PyTorch port beam-8 + full dictionary predictions, rescored on exactly the same rows; no new baseline inference.",
"size": {
"parameters": 988928,
"model_raw": 588904,
"model_gz9": 540818,
"dict_raw": 18105290,
"dict_gz9": 2588634,
"dict_words": 500000
},
"download_mb": 3.130045
}
}
}