5-shot,FineWeb-1.5T,Ours-Base,Ours-Upsampling2,All-Upsampling1 time: 22 min,Llama-8x8B-seq8192,Llama-8x8B-seq8192,Llama-8x8B-seq8192,Llama-8x8B-seq8192 5k,0.0341,0.0416,0.0565,0.0526 10k,0.0715,,0.0931,0.0767 15k,0.0765,,0.1061,0.1127 20k,0.0787,,0.1183,0.1247 25k,0.0892,0.115,0.1352,0.1343 30k,0.0911,0.1366,0.1271,0.1421 35k,0.097,0.1488,0.1485,0.1524 40k,0.1028,0.1355,0.1488,0.1562 45k,0.1078,0.1488,0.162,0.1598 50k,0.105,0.154,0.159,0.1698 55k,0.1097,0.1607,0.1662,0.1704 60k,0.1211,0.1654,0.1612,0.1801 65k,0.1089,0.1573,0.1693,0.1823 70k,0.1222,0.1634,0.1679,0.1767 75k,0.1097,0.1709,0.1881,0.1762 80k,0.1277,0.1573,0.1776,0.1964 85k,0.128,0.1776,0.1889,0.1889 90k,0.1158,0.1598,0.1806,0.1773 95k,0.1235,0.1762,0.1781,0.1917 100k,0.1258,,0.1928,0.1947 105k,0.1366,,0.1814,0.2094 110k,0.1377,0.1756,0.1859, 115k,0.1346,0.1831,0.1947,0.2119 120k,0.1402,0.2014,,0.2119 125k,0.1307,0.203,0.1992,0.1787 130k,0.1368,0.1997,0.1994,0.2086 135k,0.1363,,0.2014,0.2069 140k,0.1435,,0.1986,0.2058 145k,0.1532,,0.1953,0.2102 150k,0.1404,,,0.2075 155k,0.1418,,0.1931,0.2205 160k,0.1346,,0.2116,0.2208 165k,0.1524,,0.2139,0.2213 170k,0.1388,,,0.2169 175k,0.1438,,0.2222,0.2321 180k,0.1471,,0.2249,0.236 185k,0.1499,,0.2222,0.2366 190k,0.1504,,,0.2274 195k,0.1554,,,0.2454 200k,0.1565,,,0.2346 205k,0.1726,,,0.2316 210k,0.1623,,,0.2493 215k,0.1576,,,0.2355 220k,0.1693,,,0.2427 225k,0.1596,,,0.244 230k,0.1693,,,0.2554 235k,0.172,,,0.2535 240k,0.1712,,, 245k,0.1704,,, 250k,0.1784,,, 255k,0.174,,, 260k,0.1756,,, 265k,0.1886,,, 270k,0.182,,, 275k,0.187,,, 280k,0.1704,,, 285k,0.1903,,, 290k,,,, 300k,,,, 305k,,,, 310k,,,, 315k,,,, 320k,,,, 325k,,,, 330k,,,, 335k,,,,