Training in progress, epoch 5, checkpoint
Browse files
last-checkpoint/adapter_model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 228140600
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c14b7d8a1648c56d9f25f88d48454e081b6bc178d61bd9f0aadd582257678003
|
| 3 |
size 228140600
|
last-checkpoint/optimizer.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 117931203
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ae997529af0fecd00cf9ea60649b8488f2fcad93e7d57149ed2055f7e443e81c
|
| 3 |
size 117931203
|
last-checkpoint/rng_state.pth
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a806988fecdee5121c06d7240dec6e61421fb0008f39bed17de1e2ca05215f14
|
| 3 |
+
size 14645
|
last-checkpoint/scaler.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 1383
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:dfca50dfc66d4be0e8bab60e1bfd495197005d876487c7e37b847562cfa51471
|
| 3 |
size 1383
|
last-checkpoint/scheduler.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 1465
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7c8cc8d8f0165185e683fffb0aab5024d4cdc129dcf7f2bae80e3717e00f0c4e
|
| 3 |
size 1465
|
last-checkpoint/trainer_state.json
CHANGED
|
@@ -2,9 +2,9 @@
|
|
| 2 |
"best_global_step": 438,
|
| 3 |
"best_metric": 1.2615772485733032,
|
| 4 |
"best_model_checkpoint": "./adapter-phase2/checkpoint-438",
|
| 5 |
-
"epoch":
|
| 6 |
"eval_steps": 500,
|
| 7 |
-
"global_step":
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
|
@@ -1806,6 +1806,458 @@
|
|
| 1806 |
"eval_samples_per_second": 3.172,
|
| 1807 |
"eval_steps_per_second": 0.795,
|
| 1808 |
"step": 1752
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1809 |
}
|
| 1810 |
],
|
| 1811 |
"logging_steps": 10,
|
|
@@ -1820,12 +2272,12 @@
|
|
| 1820 |
"should_evaluate": false,
|
| 1821 |
"should_log": false,
|
| 1822 |
"should_save": true,
|
| 1823 |
-
"should_training_stop":
|
| 1824 |
},
|
| 1825 |
"attributes": {}
|
| 1826 |
}
|
| 1827 |
},
|
| 1828 |
-
"total_flos": 1.
|
| 1829 |
"train_batch_size": 1,
|
| 1830 |
"trial_name": null,
|
| 1831 |
"trial_params": null
|
|
|
|
| 2 |
"best_global_step": 438,
|
| 3 |
"best_metric": 1.2615772485733032,
|
| 4 |
"best_model_checkpoint": "./adapter-phase2/checkpoint-438",
|
| 5 |
+
"epoch": 5.0,
|
| 6 |
"eval_steps": 500,
|
| 7 |
+
"global_step": 2190,
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
|
|
|
| 1806 |
"eval_samples_per_second": 3.172,
|
| 1807 |
"eval_steps_per_second": 0.795,
|
| 1808 |
"step": 1752
|
| 1809 |
+
},
|
| 1810 |
+
{
|
| 1811 |
+
"entropy": 0.5481636302643701,
|
| 1812 |
+
"epoch": 4.018285714285715,
|
| 1813 |
+
"grad_norm": 1.847604513168335,
|
| 1814 |
+
"learning_rate": 1.9771689497716896e-06,
|
| 1815 |
+
"loss": 0.5007,
|
| 1816 |
+
"mean_token_accuracy": 0.8671465257280752,
|
| 1817 |
+
"num_tokens": 434791.0,
|
| 1818 |
+
"step": 1760
|
| 1819 |
+
},
|
| 1820 |
+
{
|
| 1821 |
+
"entropy": 0.5742504514753819,
|
| 1822 |
+
"epoch": 4.041142857142857,
|
| 1823 |
+
"grad_norm": 2.2244632244110107,
|
| 1824 |
+
"learning_rate": 1.931506849315069e-06,
|
| 1825 |
+
"loss": 0.5271,
|
| 1826 |
+
"mean_token_accuracy": 0.8591625761240721,
|
| 1827 |
+
"num_tokens": 447092.0,
|
| 1828 |
+
"step": 1770
|
| 1829 |
+
},
|
| 1830 |
+
{
|
| 1831 |
+
"entropy": 0.6953230138868094,
|
| 1832 |
+
"epoch": 4.064,
|
| 1833 |
+
"grad_norm": 3.42199444770813,
|
| 1834 |
+
"learning_rate": 1.8858447488584477e-06,
|
| 1835 |
+
"loss": 0.6469,
|
| 1836 |
+
"mean_token_accuracy": 0.8285580322146415,
|
| 1837 |
+
"num_tokens": 456254.0,
|
| 1838 |
+
"step": 1780
|
| 1839 |
+
},
|
| 1840 |
+
{
|
| 1841 |
+
"entropy": 0.7736836820840836,
|
| 1842 |
+
"epoch": 4.086857142857143,
|
| 1843 |
+
"grad_norm": 3.351454257965088,
|
| 1844 |
+
"learning_rate": 1.8401826484018268e-06,
|
| 1845 |
+
"loss": 0.6909,
|
| 1846 |
+
"mean_token_accuracy": 0.8115375626832246,
|
| 1847 |
+
"num_tokens": 463064.0,
|
| 1848 |
+
"step": 1790
|
| 1849 |
+
},
|
| 1850 |
+
{
|
| 1851 |
+
"entropy": 0.771463468298316,
|
| 1852 |
+
"epoch": 4.109714285714285,
|
| 1853 |
+
"grad_norm": 4.134479522705078,
|
| 1854 |
+
"learning_rate": 1.7945205479452058e-06,
|
| 1855 |
+
"loss": 0.6807,
|
| 1856 |
+
"mean_token_accuracy": 0.8126782298088073,
|
| 1857 |
+
"num_tokens": 468431.0,
|
| 1858 |
+
"step": 1800
|
| 1859 |
+
},
|
| 1860 |
+
{
|
| 1861 |
+
"entropy": 0.5649335160851479,
|
| 1862 |
+
"epoch": 4.132571428571429,
|
| 1863 |
+
"grad_norm": 2.1762540340423584,
|
| 1864 |
+
"learning_rate": 1.7488584474885847e-06,
|
| 1865 |
+
"loss": 0.5221,
|
| 1866 |
+
"mean_token_accuracy": 0.8567749988287687,
|
| 1867 |
+
"num_tokens": 482534.0,
|
| 1868 |
+
"step": 1810
|
| 1869 |
+
},
|
| 1870 |
+
{
|
| 1871 |
+
"entropy": 0.5837410872802138,
|
| 1872 |
+
"epoch": 4.155428571428572,
|
| 1873 |
+
"grad_norm": 2.349236011505127,
|
| 1874 |
+
"learning_rate": 1.7031963470319637e-06,
|
| 1875 |
+
"loss": 0.5371,
|
| 1876 |
+
"mean_token_accuracy": 0.8581233065575361,
|
| 1877 |
+
"num_tokens": 494845.0,
|
| 1878 |
+
"step": 1820
|
| 1879 |
+
},
|
| 1880 |
+
{
|
| 1881 |
+
"entropy": 0.6922593496739864,
|
| 1882 |
+
"epoch": 4.178285714285714,
|
| 1883 |
+
"grad_norm": 2.9896738529205322,
|
| 1884 |
+
"learning_rate": 1.6575342465753428e-06,
|
| 1885 |
+
"loss": 0.6648,
|
| 1886 |
+
"mean_token_accuracy": 0.8241277992725372,
|
| 1887 |
+
"num_tokens": 504153.0,
|
| 1888 |
+
"step": 1830
|
| 1889 |
+
},
|
| 1890 |
+
{
|
| 1891 |
+
"entropy": 0.7702124075964093,
|
| 1892 |
+
"epoch": 4.201142857142857,
|
| 1893 |
+
"grad_norm": 3.322385549545288,
|
| 1894 |
+
"learning_rate": 1.6118721461187218e-06,
|
| 1895 |
+
"loss": 0.6712,
|
| 1896 |
+
"mean_token_accuracy": 0.8188040722161531,
|
| 1897 |
+
"num_tokens": 511101.0,
|
| 1898 |
+
"step": 1840
|
| 1899 |
+
},
|
| 1900 |
+
{
|
| 1901 |
+
"entropy": 0.8012370727956295,
|
| 1902 |
+
"epoch": 4.224,
|
| 1903 |
+
"grad_norm": 4.359086036682129,
|
| 1904 |
+
"learning_rate": 1.5662100456621007e-06,
|
| 1905 |
+
"loss": 0.6748,
|
| 1906 |
+
"mean_token_accuracy": 0.8118188168853522,
|
| 1907 |
+
"num_tokens": 516460.0,
|
| 1908 |
+
"step": 1850
|
| 1909 |
+
},
|
| 1910 |
+
{
|
| 1911 |
+
"entropy": 0.5523576781153678,
|
| 1912 |
+
"epoch": 4.246857142857142,
|
| 1913 |
+
"grad_norm": 2.107539176940918,
|
| 1914 |
+
"learning_rate": 1.5205479452054797e-06,
|
| 1915 |
+
"loss": 0.5081,
|
| 1916 |
+
"mean_token_accuracy": 0.8673421230167151,
|
| 1917 |
+
"num_tokens": 530848.0,
|
| 1918 |
+
"step": 1860
|
| 1919 |
+
},
|
| 1920 |
+
{
|
| 1921 |
+
"entropy": 0.5729602897539735,
|
| 1922 |
+
"epoch": 4.269714285714286,
|
| 1923 |
+
"grad_norm": 2.5580902099609375,
|
| 1924 |
+
"learning_rate": 1.4748858447488584e-06,
|
| 1925 |
+
"loss": 0.5319,
|
| 1926 |
+
"mean_token_accuracy": 0.8585442833602428,
|
| 1927 |
+
"num_tokens": 543148.0,
|
| 1928 |
+
"step": 1870
|
| 1929 |
+
},
|
| 1930 |
+
{
|
| 1931 |
+
"entropy": 0.6956869766116143,
|
| 1932 |
+
"epoch": 4.292571428571429,
|
| 1933 |
+
"grad_norm": 3.1137397289276123,
|
| 1934 |
+
"learning_rate": 1.4292237442922373e-06,
|
| 1935 |
+
"loss": 0.6509,
|
| 1936 |
+
"mean_token_accuracy": 0.8259295519441366,
|
| 1937 |
+
"num_tokens": 552615.0,
|
| 1938 |
+
"step": 1880
|
| 1939 |
+
},
|
| 1940 |
+
{
|
| 1941 |
+
"entropy": 0.7814504994079471,
|
| 1942 |
+
"epoch": 4.315428571428572,
|
| 1943 |
+
"grad_norm": 3.9837899208068848,
|
| 1944 |
+
"learning_rate": 1.3835616438356165e-06,
|
| 1945 |
+
"loss": 0.6732,
|
| 1946 |
+
"mean_token_accuracy": 0.8206901982426643,
|
| 1947 |
+
"num_tokens": 559644.0,
|
| 1948 |
+
"step": 1890
|
| 1949 |
+
},
|
| 1950 |
+
{
|
| 1951 |
+
"entropy": 0.8006520505994559,
|
| 1952 |
+
"epoch": 4.338285714285714,
|
| 1953 |
+
"grad_norm": 4.293622016906738,
|
| 1954 |
+
"learning_rate": 1.3378995433789954e-06,
|
| 1955 |
+
"loss": 0.687,
|
| 1956 |
+
"mean_token_accuracy": 0.8162542518228293,
|
| 1957 |
+
"num_tokens": 565049.0,
|
| 1958 |
+
"step": 1900
|
| 1959 |
+
},
|
| 1960 |
+
{
|
| 1961 |
+
"entropy": 0.5543251828290522,
|
| 1962 |
+
"epoch": 4.361142857142857,
|
| 1963 |
+
"grad_norm": 2.0004706382751465,
|
| 1964 |
+
"learning_rate": 1.2922374429223744e-06,
|
| 1965 |
+
"loss": 0.5105,
|
| 1966 |
+
"mean_token_accuracy": 0.8651686757802963,
|
| 1967 |
+
"num_tokens": 579511.0,
|
| 1968 |
+
"step": 1910
|
| 1969 |
+
},
|
| 1970 |
+
{
|
| 1971 |
+
"entropy": 0.5778749627992511,
|
| 1972 |
+
"epoch": 4.384,
|
| 1973 |
+
"grad_norm": 2.4198007583618164,
|
| 1974 |
+
"learning_rate": 1.2465753424657535e-06,
|
| 1975 |
+
"loss": 0.5216,
|
| 1976 |
+
"mean_token_accuracy": 0.8559423860162496,
|
| 1977 |
+
"num_tokens": 591873.0,
|
| 1978 |
+
"step": 1920
|
| 1979 |
+
},
|
| 1980 |
+
{
|
| 1981 |
+
"entropy": 0.6606394873932004,
|
| 1982 |
+
"epoch": 4.406857142857143,
|
| 1983 |
+
"grad_norm": 3.485213279724121,
|
| 1984 |
+
"learning_rate": 1.2009132420091325e-06,
|
| 1985 |
+
"loss": 0.6086,
|
| 1986 |
+
"mean_token_accuracy": 0.8372324761003256,
|
| 1987 |
+
"num_tokens": 601337.0,
|
| 1988 |
+
"step": 1930
|
| 1989 |
+
},
|
| 1990 |
+
{
|
| 1991 |
+
"entropy": 0.7576876068487763,
|
| 1992 |
+
"epoch": 4.429714285714286,
|
| 1993 |
+
"grad_norm": 3.3860034942626953,
|
| 1994 |
+
"learning_rate": 1.1552511415525116e-06,
|
| 1995 |
+
"loss": 0.7,
|
| 1996 |
+
"mean_token_accuracy": 0.8150843985378742,
|
| 1997 |
+
"num_tokens": 608326.0,
|
| 1998 |
+
"step": 1940
|
| 1999 |
+
},
|
| 2000 |
+
{
|
| 2001 |
+
"entropy": 0.7779752794653177,
|
| 2002 |
+
"epoch": 4.452571428571429,
|
| 2003 |
+
"grad_norm": 4.422528266906738,
|
| 2004 |
+
"learning_rate": 1.1095890410958906e-06,
|
| 2005 |
+
"loss": 0.6755,
|
| 2006 |
+
"mean_token_accuracy": 0.8102645222097635,
|
| 2007 |
+
"num_tokens": 613860.0,
|
| 2008 |
+
"step": 1950
|
| 2009 |
+
},
|
| 2010 |
+
{
|
| 2011 |
+
"entropy": 0.5644787142053247,
|
| 2012 |
+
"epoch": 4.475428571428571,
|
| 2013 |
+
"grad_norm": 2.2278542518615723,
|
| 2014 |
+
"learning_rate": 1.0639269406392695e-06,
|
| 2015 |
+
"loss": 0.5188,
|
| 2016 |
+
"mean_token_accuracy": 0.8587661664932966,
|
| 2017 |
+
"num_tokens": 627853.0,
|
| 2018 |
+
"step": 1960
|
| 2019 |
+
},
|
| 2020 |
+
{
|
| 2021 |
+
"entropy": 0.5740617036819458,
|
| 2022 |
+
"epoch": 4.498285714285714,
|
| 2023 |
+
"grad_norm": 2.4340105056762695,
|
| 2024 |
+
"learning_rate": 1.0182648401826485e-06,
|
| 2025 |
+
"loss": 0.5167,
|
| 2026 |
+
"mean_token_accuracy": 0.8595852922648192,
|
| 2027 |
+
"num_tokens": 640020.0,
|
| 2028 |
+
"step": 1970
|
| 2029 |
+
},
|
| 2030 |
+
{
|
| 2031 |
+
"entropy": 0.671654068864882,
|
| 2032 |
+
"epoch": 4.521142857142857,
|
| 2033 |
+
"grad_norm": 3.127539873123169,
|
| 2034 |
+
"learning_rate": 9.726027397260274e-07,
|
| 2035 |
+
"loss": 0.6331,
|
| 2036 |
+
"mean_token_accuracy": 0.8256836850196123,
|
| 2037 |
+
"num_tokens": 649058.0,
|
| 2038 |
+
"step": 1980
|
| 2039 |
+
},
|
| 2040 |
+
{
|
| 2041 |
+
"entropy": 0.7583655359223485,
|
| 2042 |
+
"epoch": 4.5440000000000005,
|
| 2043 |
+
"grad_norm": 3.5964298248291016,
|
| 2044 |
+
"learning_rate": 9.269406392694065e-07,
|
| 2045 |
+
"loss": 0.679,
|
| 2046 |
+
"mean_token_accuracy": 0.8139733098447323,
|
| 2047 |
+
"num_tokens": 655831.0,
|
| 2048 |
+
"step": 1990
|
| 2049 |
+
},
|
| 2050 |
+
{
|
| 2051 |
+
"entropy": 0.7816360153257846,
|
| 2052 |
+
"epoch": 4.566857142857143,
|
| 2053 |
+
"grad_norm": 4.389492511749268,
|
| 2054 |
+
"learning_rate": 8.812785388127855e-07,
|
| 2055 |
+
"loss": 0.6784,
|
| 2056 |
+
"mean_token_accuracy": 0.8120625615119934,
|
| 2057 |
+
"num_tokens": 661256.0,
|
| 2058 |
+
"step": 2000
|
| 2059 |
+
},
|
| 2060 |
+
{
|
| 2061 |
+
"entropy": 0.5732687024399639,
|
| 2062 |
+
"epoch": 4.589714285714286,
|
| 2063 |
+
"grad_norm": 2.0767221450805664,
|
| 2064 |
+
"learning_rate": 8.356164383561644e-07,
|
| 2065 |
+
"loss": 0.5335,
|
| 2066 |
+
"mean_token_accuracy": 0.8624513667076826,
|
| 2067 |
+
"num_tokens": 675612.0,
|
| 2068 |
+
"step": 2010
|
| 2069 |
+
},
|
| 2070 |
+
{
|
| 2071 |
+
"entropy": 0.5804870082065463,
|
| 2072 |
+
"epoch": 4.612571428571428,
|
| 2073 |
+
"grad_norm": 2.554534673690796,
|
| 2074 |
+
"learning_rate": 7.899543378995435e-07,
|
| 2075 |
+
"loss": 0.5238,
|
| 2076 |
+
"mean_token_accuracy": 0.8590863507241011,
|
| 2077 |
+
"num_tokens": 687684.0,
|
| 2078 |
+
"step": 2020
|
| 2079 |
+
},
|
| 2080 |
+
{
|
| 2081 |
+
"entropy": 0.6967678766697645,
|
| 2082 |
+
"epoch": 4.635428571428571,
|
| 2083 |
+
"grad_norm": 3.255140542984009,
|
| 2084 |
+
"learning_rate": 7.442922374429224e-07,
|
| 2085 |
+
"loss": 0.6487,
|
| 2086 |
+
"mean_token_accuracy": 0.8244634248316288,
|
| 2087 |
+
"num_tokens": 696675.0,
|
| 2088 |
+
"step": 2030
|
| 2089 |
+
},
|
| 2090 |
+
{
|
| 2091 |
+
"entropy": 0.7526340587064624,
|
| 2092 |
+
"epoch": 4.658285714285714,
|
| 2093 |
+
"grad_norm": 3.69323992729187,
|
| 2094 |
+
"learning_rate": 6.986301369863015e-07,
|
| 2095 |
+
"loss": 0.6719,
|
| 2096 |
+
"mean_token_accuracy": 0.8216490592807532,
|
| 2097 |
+
"num_tokens": 703456.0,
|
| 2098 |
+
"step": 2040
|
| 2099 |
+
},
|
| 2100 |
+
{
|
| 2101 |
+
"entropy": 0.7950498787686229,
|
| 2102 |
+
"epoch": 4.6811428571428575,
|
| 2103 |
+
"grad_norm": 4.715794563293457,
|
| 2104 |
+
"learning_rate": 6.529680365296804e-07,
|
| 2105 |
+
"loss": 0.6808,
|
| 2106 |
+
"mean_token_accuracy": 0.8184644509106874,
|
| 2107 |
+
"num_tokens": 708782.0,
|
| 2108 |
+
"step": 2050
|
| 2109 |
+
},
|
| 2110 |
+
{
|
| 2111 |
+
"entropy": 0.5505135927349329,
|
| 2112 |
+
"epoch": 4.704,
|
| 2113 |
+
"grad_norm": 2.3146073818206787,
|
| 2114 |
+
"learning_rate": 6.073059360730594e-07,
|
| 2115 |
+
"loss": 0.507,
|
| 2116 |
+
"mean_token_accuracy": 0.8652824487537145,
|
| 2117 |
+
"num_tokens": 723247.0,
|
| 2118 |
+
"step": 2060
|
| 2119 |
+
},
|
| 2120 |
+
{
|
| 2121 |
+
"entropy": 0.5807493371888995,
|
| 2122 |
+
"epoch": 4.726857142857143,
|
| 2123 |
+
"grad_norm": 2.615732192993164,
|
| 2124 |
+
"learning_rate": 5.616438356164384e-07,
|
| 2125 |
+
"loss": 0.5342,
|
| 2126 |
+
"mean_token_accuracy": 0.854337964951992,
|
| 2127 |
+
"num_tokens": 735283.0,
|
| 2128 |
+
"step": 2070
|
| 2129 |
+
},
|
| 2130 |
+
{
|
| 2131 |
+
"entropy": 0.7081154704093933,
|
| 2132 |
+
"epoch": 4.749714285714286,
|
| 2133 |
+
"grad_norm": 3.0795960426330566,
|
| 2134 |
+
"learning_rate": 5.159817351598174e-07,
|
| 2135 |
+
"loss": 0.6499,
|
| 2136 |
+
"mean_token_accuracy": 0.8241405732929706,
|
| 2137 |
+
"num_tokens": 744298.0,
|
| 2138 |
+
"step": 2080
|
| 2139 |
+
},
|
| 2140 |
+
{
|
| 2141 |
+
"entropy": 0.783203998953104,
|
| 2142 |
+
"epoch": 4.772571428571428,
|
| 2143 |
+
"grad_norm": 3.7807230949401855,
|
| 2144 |
+
"learning_rate": 4.7031963470319636e-07,
|
| 2145 |
+
"loss": 0.6948,
|
| 2146 |
+
"mean_token_accuracy": 0.8167315106838942,
|
| 2147 |
+
"num_tokens": 751212.0,
|
| 2148 |
+
"step": 2090
|
| 2149 |
+
},
|
| 2150 |
+
{
|
| 2151 |
+
"entropy": 0.7742167858406901,
|
| 2152 |
+
"epoch": 4.795428571428571,
|
| 2153 |
+
"grad_norm": 4.185308933258057,
|
| 2154 |
+
"learning_rate": 4.2465753424657536e-07,
|
| 2155 |
+
"loss": 0.6705,
|
| 2156 |
+
"mean_token_accuracy": 0.8145616598427295,
|
| 2157 |
+
"num_tokens": 756648.0,
|
| 2158 |
+
"step": 2100
|
| 2159 |
+
},
|
| 2160 |
+
{
|
| 2161 |
+
"entropy": 0.5554421614855528,
|
| 2162 |
+
"epoch": 4.8182857142857145,
|
| 2163 |
+
"grad_norm": 2.0456132888793945,
|
| 2164 |
+
"learning_rate": 3.7899543378995436e-07,
|
| 2165 |
+
"loss": 0.4982,
|
| 2166 |
+
"mean_token_accuracy": 0.8656269229948521,
|
| 2167 |
+
"num_tokens": 771047.0,
|
| 2168 |
+
"step": 2110
|
| 2169 |
+
},
|
| 2170 |
+
{
|
| 2171 |
+
"entropy": 0.5584687992930413,
|
| 2172 |
+
"epoch": 4.841142857142858,
|
| 2173 |
+
"grad_norm": 2.591322422027588,
|
| 2174 |
+
"learning_rate": 3.3333333333333335e-07,
|
| 2175 |
+
"loss": 0.5038,
|
| 2176 |
+
"mean_token_accuracy": 0.8639244794845581,
|
| 2177 |
+
"num_tokens": 783484.0,
|
| 2178 |
+
"step": 2120
|
| 2179 |
+
},
|
| 2180 |
+
{
|
| 2181 |
+
"entropy": 0.6443877406418324,
|
| 2182 |
+
"epoch": 4.864,
|
| 2183 |
+
"grad_norm": 3.1148664951324463,
|
| 2184 |
+
"learning_rate": 2.8767123287671235e-07,
|
| 2185 |
+
"loss": 0.5898,
|
| 2186 |
+
"mean_token_accuracy": 0.8393101956695318,
|
| 2187 |
+
"num_tokens": 793159.0,
|
| 2188 |
+
"step": 2130
|
| 2189 |
+
},
|
| 2190 |
+
{
|
| 2191 |
+
"entropy": 0.7694938328117132,
|
| 2192 |
+
"epoch": 4.886857142857143,
|
| 2193 |
+
"grad_norm": 3.860647201538086,
|
| 2194 |
+
"learning_rate": 2.4200913242009135e-07,
|
| 2195 |
+
"loss": 0.6661,
|
| 2196 |
+
"mean_token_accuracy": 0.8193999473005533,
|
| 2197 |
+
"num_tokens": 800322.0,
|
| 2198 |
+
"step": 2140
|
| 2199 |
+
},
|
| 2200 |
+
{
|
| 2201 |
+
"entropy": 0.7594648722559214,
|
| 2202 |
+
"epoch": 4.909714285714285,
|
| 2203 |
+
"grad_norm": 4.108844757080078,
|
| 2204 |
+
"learning_rate": 1.9634703196347034e-07,
|
| 2205 |
+
"loss": 0.6513,
|
| 2206 |
+
"mean_token_accuracy": 0.820786502957344,
|
| 2207 |
+
"num_tokens": 805782.0,
|
| 2208 |
+
"step": 2150
|
| 2209 |
+
},
|
| 2210 |
+
{
|
| 2211 |
+
"entropy": 0.5718974178656936,
|
| 2212 |
+
"epoch": 4.932571428571428,
|
| 2213 |
+
"grad_norm": 2.0961592197418213,
|
| 2214 |
+
"learning_rate": 1.5068493150684934e-07,
|
| 2215 |
+
"loss": 0.5321,
|
| 2216 |
+
"mean_token_accuracy": 0.8587038304656744,
|
| 2217 |
+
"num_tokens": 819399.0,
|
| 2218 |
+
"step": 2160
|
| 2219 |
+
},
|
| 2220 |
+
{
|
| 2221 |
+
"entropy": 0.6284356378018856,
|
| 2222 |
+
"epoch": 4.9554285714285715,
|
| 2223 |
+
"grad_norm": 2.863541841506958,
|
| 2224 |
+
"learning_rate": 1.0502283105022832e-07,
|
| 2225 |
+
"loss": 0.5916,
|
| 2226 |
+
"mean_token_accuracy": 0.8409151379019022,
|
| 2227 |
+
"num_tokens": 830244.0,
|
| 2228 |
+
"step": 2170
|
| 2229 |
+
},
|
| 2230 |
+
{
|
| 2231 |
+
"entropy": 0.7672561943531037,
|
| 2232 |
+
"epoch": 4.978285714285715,
|
| 2233 |
+
"grad_norm": 3.52964186668396,
|
| 2234 |
+
"learning_rate": 5.936073059360731e-08,
|
| 2235 |
+
"loss": 0.6968,
|
| 2236 |
+
"mean_token_accuracy": 0.8189954232424498,
|
| 2237 |
+
"num_tokens": 837492.0,
|
| 2238 |
+
"step": 2180
|
| 2239 |
+
},
|
| 2240 |
+
{
|
| 2241 |
+
"entropy": 0.7868584784630098,
|
| 2242 |
+
"epoch": 5.0,
|
| 2243 |
+
"grad_norm": 6.119595050811768,
|
| 2244 |
+
"learning_rate": 1.3698630136986303e-08,
|
| 2245 |
+
"loss": 0.6771,
|
| 2246 |
+
"mean_token_accuracy": 0.8141911743502868,
|
| 2247 |
+
"num_tokens": 842388.0,
|
| 2248 |
+
"step": 2190
|
| 2249 |
+
},
|
| 2250 |
+
{
|
| 2251 |
+
"epoch": 5.0,
|
| 2252 |
+
"eval_accuracy": 0.009840881272949817,
|
| 2253 |
+
"eval_entropy": 0.8240771901193272,
|
| 2254 |
+
"eval_loss": 1.5346653461456299,
|
| 2255 |
+
"eval_mean_token_accuracy": 0.7217774188656605,
|
| 2256 |
+
"eval_num_tokens": 842388.0,
|
| 2257 |
+
"eval_runtime": 325.5092,
|
| 2258 |
+
"eval_samples_per_second": 3.177,
|
| 2259 |
+
"eval_steps_per_second": 0.796,
|
| 2260 |
+
"step": 2190
|
| 2261 |
}
|
| 2262 |
],
|
| 2263 |
"logging_steps": 10,
|
|
|
|
| 2272 |
"should_evaluate": false,
|
| 2273 |
"should_log": false,
|
| 2274 |
"should_save": true,
|
| 2275 |
+
"should_training_stop": true
|
| 2276 |
},
|
| 2277 |
"attributes": {}
|
| 2278 |
}
|
| 2279 |
},
|
| 2280 |
+
"total_flos": 1.470014665030656e+17,
|
| 2281 |
"train_batch_size": 1,
|
| 2282 |
"trial_name": null,
|
| 2283 |
"trial_params": null
|