Training checkpoint at step 6000
Browse files- trainer_state.json +365 -5
trainer_state.json
CHANGED
|
@@ -1,10 +1,10 @@
|
|
| 1 |
{
|
| 2 |
-
"best_global_step":
|
| 3 |
-
"best_metric": 2.
|
| 4 |
"best_model_checkpoint": "../runs/karpathy/fineweb-edu-100b-shuffle/meta-llama/Llama-3.2-1B/linear_adamw_wd1e-03_7x1024_mem32_bs64_hf_armt_dmem64/run_30/checkpoint-5000",
|
| 5 |
-
"epoch": 0.
|
| 6 |
"eval_steps": 100,
|
| 7 |
-
"global_step":
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
|
@@ -1808,6 +1808,366 @@
|
|
| 1808 |
"eval_samples_per_second": 3.544,
|
| 1809 |
"eval_steps_per_second": 1.787,
|
| 1810 |
"step": 5000
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1811 |
}
|
| 1812 |
],
|
| 1813 |
"logging_steps": 25,
|
|
@@ -1827,7 +2187,7 @@
|
|
| 1827 |
"attributes": {}
|
| 1828 |
}
|
| 1829 |
},
|
| 1830 |
-
"total_flos": 1.
|
| 1831 |
"train_batch_size": 1,
|
| 1832 |
"trial_name": null,
|
| 1833 |
"trial_params": null
|
|
|
|
| 1 |
{
|
| 2 |
+
"best_global_step": 5900,
|
| 3 |
+
"best_metric": 2.4210917949676514,
|
| 4 |
"best_model_checkpoint": "../runs/karpathy/fineweb-edu-100b-shuffle/meta-llama/Llama-3.2-1B/linear_adamw_wd1e-03_7x1024_mem32_bs64_hf_armt_dmem64/run_30/checkpoint-5000",
|
| 5 |
+
"epoch": 0.12,
|
| 6 |
"eval_steps": 100,
|
| 7 |
+
"global_step": 6000,
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
|
|
|
| 1808 |
"eval_samples_per_second": 3.544,
|
| 1809 |
"eval_steps_per_second": 1.787,
|
| 1810 |
"step": 5000
|
| 1811 |
+
},
|
| 1812 |
+
{
|
| 1813 |
+
"epoch": 0.1005,
|
| 1814 |
+
"grad_norm": 0.0187590945164155,
|
| 1815 |
+
"learning_rate": 9.994666666666668e-06,
|
| 1816 |
+
"loss": 2.4164,
|
| 1817 |
+
"step": 5025
|
| 1818 |
+
},
|
| 1819 |
+
{
|
| 1820 |
+
"epoch": 0.101,
|
| 1821 |
+
"grad_norm": 0.018683158146542603,
|
| 1822 |
+
"learning_rate": 9.989111111111111e-06,
|
| 1823 |
+
"loss": 2.4082,
|
| 1824 |
+
"step": 5050
|
| 1825 |
+
},
|
| 1826 |
+
{
|
| 1827 |
+
"epoch": 0.1015,
|
| 1828 |
+
"grad_norm": 0.017610949419625762,
|
| 1829 |
+
"learning_rate": 9.983555555555556e-06,
|
| 1830 |
+
"loss": 2.4124,
|
| 1831 |
+
"step": 5075
|
| 1832 |
+
},
|
| 1833 |
+
{
|
| 1834 |
+
"epoch": 0.102,
|
| 1835 |
+
"grad_norm": 0.01862298073358942,
|
| 1836 |
+
"learning_rate": 9.978000000000002e-06,
|
| 1837 |
+
"loss": 2.409,
|
| 1838 |
+
"step": 5100
|
| 1839 |
+
},
|
| 1840 |
+
{
|
| 1841 |
+
"epoch": 0.102,
|
| 1842 |
+
"eval_loss": 2.425841808319092,
|
| 1843 |
+
"eval_runtime": 33.063,
|
| 1844 |
+
"eval_samples_per_second": 3.539,
|
| 1845 |
+
"eval_steps_per_second": 1.784,
|
| 1846 |
+
"step": 5100
|
| 1847 |
+
},
|
| 1848 |
+
{
|
| 1849 |
+
"epoch": 0.1025,
|
| 1850 |
+
"grad_norm": 0.025407800531065724,
|
| 1851 |
+
"learning_rate": 9.972444444444445e-06,
|
| 1852 |
+
"loss": 2.4051,
|
| 1853 |
+
"step": 5125
|
| 1854 |
+
},
|
| 1855 |
+
{
|
| 1856 |
+
"epoch": 0.103,
|
| 1857 |
+
"grad_norm": 0.01838713779514561,
|
| 1858 |
+
"learning_rate": 9.966888888888889e-06,
|
| 1859 |
+
"loss": 2.4105,
|
| 1860 |
+
"step": 5150
|
| 1861 |
+
},
|
| 1862 |
+
{
|
| 1863 |
+
"epoch": 0.1035,
|
| 1864 |
+
"grad_norm": 0.018921321521659856,
|
| 1865 |
+
"learning_rate": 9.961333333333334e-06,
|
| 1866 |
+
"loss": 2.4191,
|
| 1867 |
+
"step": 5175
|
| 1868 |
+
},
|
| 1869 |
+
{
|
| 1870 |
+
"epoch": 0.104,
|
| 1871 |
+
"grad_norm": 0.01824666535901335,
|
| 1872 |
+
"learning_rate": 9.95577777777778e-06,
|
| 1873 |
+
"loss": 2.4115,
|
| 1874 |
+
"step": 5200
|
| 1875 |
+
},
|
| 1876 |
+
{
|
| 1877 |
+
"epoch": 0.104,
|
| 1878 |
+
"eval_loss": 2.4254310131073,
|
| 1879 |
+
"eval_runtime": 33.141,
|
| 1880 |
+
"eval_samples_per_second": 3.53,
|
| 1881 |
+
"eval_steps_per_second": 1.78,
|
| 1882 |
+
"step": 5200
|
| 1883 |
+
},
|
| 1884 |
+
{
|
| 1885 |
+
"epoch": 0.1045,
|
| 1886 |
+
"grad_norm": 0.018794067362196056,
|
| 1887 |
+
"learning_rate": 9.950222222222223e-06,
|
| 1888 |
+
"loss": 2.4062,
|
| 1889 |
+
"step": 5225
|
| 1890 |
+
},
|
| 1891 |
+
{
|
| 1892 |
+
"epoch": 0.105,
|
| 1893 |
+
"grad_norm": 0.01825837669653065,
|
| 1894 |
+
"learning_rate": 9.944666666666668e-06,
|
| 1895 |
+
"loss": 2.4154,
|
| 1896 |
+
"step": 5250
|
| 1897 |
+
},
|
| 1898 |
+
{
|
| 1899 |
+
"epoch": 0.1055,
|
| 1900 |
+
"grad_norm": 0.01843310767671649,
|
| 1901 |
+
"learning_rate": 9.939111111111112e-06,
|
| 1902 |
+
"loss": 2.4201,
|
| 1903 |
+
"step": 5275
|
| 1904 |
+
},
|
| 1905 |
+
{
|
| 1906 |
+
"epoch": 0.106,
|
| 1907 |
+
"grad_norm": 0.018304681522005508,
|
| 1908 |
+
"learning_rate": 9.933555555555557e-06,
|
| 1909 |
+
"loss": 2.4089,
|
| 1910 |
+
"step": 5300
|
| 1911 |
+
},
|
| 1912 |
+
{
|
| 1913 |
+
"epoch": 0.106,
|
| 1914 |
+
"eval_loss": 2.424731492996216,
|
| 1915 |
+
"eval_runtime": 33.0325,
|
| 1916 |
+
"eval_samples_per_second": 3.542,
|
| 1917 |
+
"eval_steps_per_second": 1.786,
|
| 1918 |
+
"step": 5300
|
| 1919 |
+
},
|
| 1920 |
+
{
|
| 1921 |
+
"epoch": 0.1065,
|
| 1922 |
+
"grad_norm": 0.01846362790517963,
|
| 1923 |
+
"learning_rate": 9.928e-06,
|
| 1924 |
+
"loss": 2.4118,
|
| 1925 |
+
"step": 5325
|
| 1926 |
+
},
|
| 1927 |
+
{
|
| 1928 |
+
"epoch": 0.107,
|
| 1929 |
+
"grad_norm": 0.01872825463357926,
|
| 1930 |
+
"learning_rate": 9.922444444444446e-06,
|
| 1931 |
+
"loss": 2.4045,
|
| 1932 |
+
"step": 5350
|
| 1933 |
+
},
|
| 1934 |
+
{
|
| 1935 |
+
"epoch": 0.1075,
|
| 1936 |
+
"grad_norm": 0.017781011104963246,
|
| 1937 |
+
"learning_rate": 9.91688888888889e-06,
|
| 1938 |
+
"loss": 2.4145,
|
| 1939 |
+
"step": 5375
|
| 1940 |
+
},
|
| 1941 |
+
{
|
| 1942 |
+
"epoch": 0.108,
|
| 1943 |
+
"grad_norm": 0.018840752543683545,
|
| 1944 |
+
"learning_rate": 9.911333333333335e-06,
|
| 1945 |
+
"loss": 2.416,
|
| 1946 |
+
"step": 5400
|
| 1947 |
+
},
|
| 1948 |
+
{
|
| 1949 |
+
"epoch": 0.108,
|
| 1950 |
+
"eval_loss": 2.423886775970459,
|
| 1951 |
+
"eval_runtime": 33.1239,
|
| 1952 |
+
"eval_samples_per_second": 3.532,
|
| 1953 |
+
"eval_steps_per_second": 1.781,
|
| 1954 |
+
"step": 5400
|
| 1955 |
+
},
|
| 1956 |
+
{
|
| 1957 |
+
"epoch": 0.1085,
|
| 1958 |
+
"grad_norm": 0.019278786947294697,
|
| 1959 |
+
"learning_rate": 9.905777777777778e-06,
|
| 1960 |
+
"loss": 2.4117,
|
| 1961 |
+
"step": 5425
|
| 1962 |
+
},
|
| 1963 |
+
{
|
| 1964 |
+
"epoch": 0.109,
|
| 1965 |
+
"grad_norm": 0.018430470806705172,
|
| 1966 |
+
"learning_rate": 9.900222222222223e-06,
|
| 1967 |
+
"loss": 2.4114,
|
| 1968 |
+
"step": 5450
|
| 1969 |
+
},
|
| 1970 |
+
{
|
| 1971 |
+
"epoch": 0.1095,
|
| 1972 |
+
"grad_norm": 0.018464088455141334,
|
| 1973 |
+
"learning_rate": 9.894666666666669e-06,
|
| 1974 |
+
"loss": 2.4185,
|
| 1975 |
+
"step": 5475
|
| 1976 |
+
},
|
| 1977 |
+
{
|
| 1978 |
+
"epoch": 0.11,
|
| 1979 |
+
"grad_norm": 0.01866239126789079,
|
| 1980 |
+
"learning_rate": 9.889111111111112e-06,
|
| 1981 |
+
"loss": 2.4099,
|
| 1982 |
+
"step": 5500
|
| 1983 |
+
},
|
| 1984 |
+
{
|
| 1985 |
+
"epoch": 0.11,
|
| 1986 |
+
"eval_loss": 2.423039197921753,
|
| 1987 |
+
"eval_runtime": 35.4471,
|
| 1988 |
+
"eval_samples_per_second": 3.301,
|
| 1989 |
+
"eval_steps_per_second": 1.664,
|
| 1990 |
+
"step": 5500
|
| 1991 |
+
},
|
| 1992 |
+
{
|
| 1993 |
+
"epoch": 0.1105,
|
| 1994 |
+
"grad_norm": 0.01827370320895024,
|
| 1995 |
+
"learning_rate": 9.883555555555556e-06,
|
| 1996 |
+
"loss": 2.4078,
|
| 1997 |
+
"step": 5525
|
| 1998 |
+
},
|
| 1999 |
+
{
|
| 2000 |
+
"epoch": 0.111,
|
| 2001 |
+
"grad_norm": 0.01863057836209491,
|
| 2002 |
+
"learning_rate": 9.878000000000001e-06,
|
| 2003 |
+
"loss": 2.4044,
|
| 2004 |
+
"step": 5550
|
| 2005 |
+
},
|
| 2006 |
+
{
|
| 2007 |
+
"epoch": 0.1115,
|
| 2008 |
+
"grad_norm": 0.018262835671926946,
|
| 2009 |
+
"learning_rate": 9.872444444444446e-06,
|
| 2010 |
+
"loss": 2.4123,
|
| 2011 |
+
"step": 5575
|
| 2012 |
+
},
|
| 2013 |
+
{
|
| 2014 |
+
"epoch": 0.112,
|
| 2015 |
+
"grad_norm": 0.017655227692766756,
|
| 2016 |
+
"learning_rate": 9.86688888888889e-06,
|
| 2017 |
+
"loss": 2.4118,
|
| 2018 |
+
"step": 5600
|
| 2019 |
+
},
|
| 2020 |
+
{
|
| 2021 |
+
"epoch": 0.112,
|
| 2022 |
+
"eval_loss": 2.4225943088531494,
|
| 2023 |
+
"eval_runtime": 33.2709,
|
| 2024 |
+
"eval_samples_per_second": 3.517,
|
| 2025 |
+
"eval_steps_per_second": 1.773,
|
| 2026 |
+
"step": 5600
|
| 2027 |
+
},
|
| 2028 |
+
{
|
| 2029 |
+
"epoch": 0.1125,
|
| 2030 |
+
"grad_norm": 0.01812962067528887,
|
| 2031 |
+
"learning_rate": 9.861333333333333e-06,
|
| 2032 |
+
"loss": 2.4017,
|
| 2033 |
+
"step": 5625
|
| 2034 |
+
},
|
| 2035 |
+
{
|
| 2036 |
+
"epoch": 0.113,
|
| 2037 |
+
"grad_norm": 0.018265397582930686,
|
| 2038 |
+
"learning_rate": 9.855777777777779e-06,
|
| 2039 |
+
"loss": 2.4166,
|
| 2040 |
+
"step": 5650
|
| 2041 |
+
},
|
| 2042 |
+
{
|
| 2043 |
+
"epoch": 0.1135,
|
| 2044 |
+
"grad_norm": 0.018207114017877214,
|
| 2045 |
+
"learning_rate": 9.850222222222224e-06,
|
| 2046 |
+
"loss": 2.413,
|
| 2047 |
+
"step": 5675
|
| 2048 |
+
},
|
| 2049 |
+
{
|
| 2050 |
+
"epoch": 0.114,
|
| 2051 |
+
"grad_norm": 0.01952225079171619,
|
| 2052 |
+
"learning_rate": 9.844666666666667e-06,
|
| 2053 |
+
"loss": 2.4022,
|
| 2054 |
+
"step": 5700
|
| 2055 |
+
},
|
| 2056 |
+
{
|
| 2057 |
+
"epoch": 0.114,
|
| 2058 |
+
"eval_loss": 2.42179274559021,
|
| 2059 |
+
"eval_runtime": 33.0648,
|
| 2060 |
+
"eval_samples_per_second": 3.539,
|
| 2061 |
+
"eval_steps_per_second": 1.784,
|
| 2062 |
+
"step": 5700
|
| 2063 |
+
},
|
| 2064 |
+
{
|
| 2065 |
+
"epoch": 0.1145,
|
| 2066 |
+
"grad_norm": 0.01780836124763766,
|
| 2067 |
+
"learning_rate": 9.839111111111111e-06,
|
| 2068 |
+
"loss": 2.4128,
|
| 2069 |
+
"step": 5725
|
| 2070 |
+
},
|
| 2071 |
+
{
|
| 2072 |
+
"epoch": 0.115,
|
| 2073 |
+
"grad_norm": 0.018290904429709265,
|
| 2074 |
+
"learning_rate": 9.833555555555556e-06,
|
| 2075 |
+
"loss": 2.4119,
|
| 2076 |
+
"step": 5750
|
| 2077 |
+
},
|
| 2078 |
+
{
|
| 2079 |
+
"epoch": 0.1155,
|
| 2080 |
+
"grad_norm": 0.019359740861514655,
|
| 2081 |
+
"learning_rate": 9.828000000000001e-06,
|
| 2082 |
+
"loss": 2.4019,
|
| 2083 |
+
"step": 5775
|
| 2084 |
+
},
|
| 2085 |
+
{
|
| 2086 |
+
"epoch": 0.116,
|
| 2087 |
+
"grad_norm": 0.018278231474623628,
|
| 2088 |
+
"learning_rate": 9.822444444444445e-06,
|
| 2089 |
+
"loss": 2.4072,
|
| 2090 |
+
"step": 5800
|
| 2091 |
+
},
|
| 2092 |
+
{
|
| 2093 |
+
"epoch": 0.116,
|
| 2094 |
+
"eval_loss": 2.4214675426483154,
|
| 2095 |
+
"eval_runtime": 33.0642,
|
| 2096 |
+
"eval_samples_per_second": 3.539,
|
| 2097 |
+
"eval_steps_per_second": 1.784,
|
| 2098 |
+
"step": 5800
|
| 2099 |
+
},
|
| 2100 |
+
{
|
| 2101 |
+
"epoch": 0.1165,
|
| 2102 |
+
"grad_norm": 0.017493007146383306,
|
| 2103 |
+
"learning_rate": 9.81688888888889e-06,
|
| 2104 |
+
"loss": 2.4134,
|
| 2105 |
+
"step": 5825
|
| 2106 |
+
},
|
| 2107 |
+
{
|
| 2108 |
+
"epoch": 0.117,
|
| 2109 |
+
"grad_norm": 0.018399348008473985,
|
| 2110 |
+
"learning_rate": 9.811333333333334e-06,
|
| 2111 |
+
"loss": 2.4082,
|
| 2112 |
+
"step": 5850
|
| 2113 |
+
},
|
| 2114 |
+
{
|
| 2115 |
+
"epoch": 0.1175,
|
| 2116 |
+
"grad_norm": 0.0186494867742927,
|
| 2117 |
+
"learning_rate": 9.805777777777779e-06,
|
| 2118 |
+
"loss": 2.4131,
|
| 2119 |
+
"step": 5875
|
| 2120 |
+
},
|
| 2121 |
+
{
|
| 2122 |
+
"epoch": 0.118,
|
| 2123 |
+
"grad_norm": 0.017842605036949514,
|
| 2124 |
+
"learning_rate": 9.800222222222223e-06,
|
| 2125 |
+
"loss": 2.4134,
|
| 2126 |
+
"step": 5900
|
| 2127 |
+
},
|
| 2128 |
+
{
|
| 2129 |
+
"epoch": 0.118,
|
| 2130 |
+
"eval_loss": 2.4210917949676514,
|
| 2131 |
+
"eval_runtime": 33.1318,
|
| 2132 |
+
"eval_samples_per_second": 3.531,
|
| 2133 |
+
"eval_steps_per_second": 1.781,
|
| 2134 |
+
"step": 5900
|
| 2135 |
+
},
|
| 2136 |
+
{
|
| 2137 |
+
"epoch": 0.1185,
|
| 2138 |
+
"grad_norm": 0.01835138877842204,
|
| 2139 |
+
"learning_rate": 9.794666666666668e-06,
|
| 2140 |
+
"loss": 2.4017,
|
| 2141 |
+
"step": 5925
|
| 2142 |
+
},
|
| 2143 |
+
{
|
| 2144 |
+
"epoch": 0.119,
|
| 2145 |
+
"grad_norm": 0.018202303746487493,
|
| 2146 |
+
"learning_rate": 9.789111111111111e-06,
|
| 2147 |
+
"loss": 2.4103,
|
| 2148 |
+
"step": 5950
|
| 2149 |
+
},
|
| 2150 |
+
{
|
| 2151 |
+
"epoch": 0.1195,
|
| 2152 |
+
"grad_norm": 0.0176777777086958,
|
| 2153 |
+
"learning_rate": 9.783555555555557e-06,
|
| 2154 |
+
"loss": 2.4023,
|
| 2155 |
+
"step": 5975
|
| 2156 |
+
},
|
| 2157 |
+
{
|
| 2158 |
+
"epoch": 0.12,
|
| 2159 |
+
"grad_norm": 0.019351209333625233,
|
| 2160 |
+
"learning_rate": 9.778e-06,
|
| 2161 |
+
"loss": 2.4053,
|
| 2162 |
+
"step": 6000
|
| 2163 |
+
},
|
| 2164 |
+
{
|
| 2165 |
+
"epoch": 0.12,
|
| 2166 |
+
"eval_loss": 2.421157121658325,
|
| 2167 |
+
"eval_runtime": 33.0891,
|
| 2168 |
+
"eval_samples_per_second": 3.536,
|
| 2169 |
+
"eval_steps_per_second": 1.783,
|
| 2170 |
+
"step": 6000
|
| 2171 |
}
|
| 2172 |
],
|
| 2173 |
"logging_steps": 25,
|
|
|
|
| 2187 |
"attributes": {}
|
| 2188 |
}
|
| 2189 |
},
|
| 2190 |
+
"total_flos": 1.6711811550821745e+19,
|
| 2191 |
"train_batch_size": 1,
|
| 2192 |
"trial_name": null,
|
| 2193 |
"trial_params": null
|