Training checkpoint at step 5000
Browse files- trainer_state.json +366 -6
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-
|
| 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,
|
|
@@ -1448,6 +1448,366 @@
|
|
| 1448 |
"eval_samples_per_second": 3.532,
|
| 1449 |
"eval_steps_per_second": 1.781,
|
| 1450 |
"step": 4000
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1451 |
}
|
| 1452 |
],
|
| 1453 |
"logging_steps": 25,
|
|
@@ -1467,7 +1827,7 @@
|
|
| 1467 |
"attributes": {}
|
| 1468 |
}
|
| 1469 |
},
|
| 1470 |
-
"total_flos": 1.
|
| 1471 |
"train_batch_size": 1,
|
| 1472 |
"trial_name": null,
|
| 1473 |
"trial_params": null
|
|
|
|
| 1 |
{
|
| 2 |
+
"best_global_step": 5000,
|
| 3 |
+
"best_metric": 2.426590919494629,
|
| 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": 0.1,
|
| 6 |
"eval_steps": 100,
|
| 7 |
+
"global_step": 5000,
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
|
|
|
| 1448 |
"eval_samples_per_second": 3.532,
|
| 1449 |
"eval_steps_per_second": 1.781,
|
| 1450 |
"step": 4000
|
| 1451 |
+
},
|
| 1452 |
+
{
|
| 1453 |
+
"epoch": 0.0805,
|
| 1454 |
+
"grad_norm": 0.021865944897834468,
|
| 1455 |
+
"learning_rate": 8.048e-06,
|
| 1456 |
+
"loss": 2.4283,
|
| 1457 |
+
"step": 4025
|
| 1458 |
+
},
|
| 1459 |
+
{
|
| 1460 |
+
"epoch": 0.081,
|
| 1461 |
+
"grad_norm": 0.020393010409248808,
|
| 1462 |
+
"learning_rate": 8.098000000000001e-06,
|
| 1463 |
+
"loss": 2.4142,
|
| 1464 |
+
"step": 4050
|
| 1465 |
+
},
|
| 1466 |
+
{
|
| 1467 |
+
"epoch": 0.0815,
|
| 1468 |
+
"grad_norm": 0.02279155824698799,
|
| 1469 |
+
"learning_rate": 8.148e-06,
|
| 1470 |
+
"loss": 2.4208,
|
| 1471 |
+
"step": 4075
|
| 1472 |
+
},
|
| 1473 |
+
{
|
| 1474 |
+
"epoch": 0.082,
|
| 1475 |
+
"grad_norm": 0.021110562493101104,
|
| 1476 |
+
"learning_rate": 8.198e-06,
|
| 1477 |
+
"loss": 2.4093,
|
| 1478 |
+
"step": 4100
|
| 1479 |
+
},
|
| 1480 |
+
{
|
| 1481 |
+
"epoch": 0.082,
|
| 1482 |
+
"eval_loss": 2.4299628734588623,
|
| 1483 |
+
"eval_runtime": 33.2215,
|
| 1484 |
+
"eval_samples_per_second": 3.522,
|
| 1485 |
+
"eval_steps_per_second": 1.776,
|
| 1486 |
+
"step": 4100
|
| 1487 |
+
},
|
| 1488 |
+
{
|
| 1489 |
+
"epoch": 0.0825,
|
| 1490 |
+
"grad_norm": 0.019752507861163327,
|
| 1491 |
+
"learning_rate": 8.248e-06,
|
| 1492 |
+
"loss": 2.4073,
|
| 1493 |
+
"step": 4125
|
| 1494 |
+
},
|
| 1495 |
+
{
|
| 1496 |
+
"epoch": 0.083,
|
| 1497 |
+
"grad_norm": 0.019897433088879975,
|
| 1498 |
+
"learning_rate": 8.298000000000001e-06,
|
| 1499 |
+
"loss": 2.4129,
|
| 1500 |
+
"step": 4150
|
| 1501 |
+
},
|
| 1502 |
+
{
|
| 1503 |
+
"epoch": 0.0835,
|
| 1504 |
+
"grad_norm": 0.02275241957806373,
|
| 1505 |
+
"learning_rate": 8.348e-06,
|
| 1506 |
+
"loss": 2.4243,
|
| 1507 |
+
"step": 4175
|
| 1508 |
+
},
|
| 1509 |
+
{
|
| 1510 |
+
"epoch": 0.084,
|
| 1511 |
+
"grad_norm": 0.02009113389579191,
|
| 1512 |
+
"learning_rate": 8.398e-06,
|
| 1513 |
+
"loss": 2.4138,
|
| 1514 |
+
"step": 4200
|
| 1515 |
+
},
|
| 1516 |
+
{
|
| 1517 |
+
"epoch": 0.084,
|
| 1518 |
+
"eval_loss": 2.4301230907440186,
|
| 1519 |
+
"eval_runtime": 33.0641,
|
| 1520 |
+
"eval_samples_per_second": 3.539,
|
| 1521 |
+
"eval_steps_per_second": 1.784,
|
| 1522 |
+
"step": 4200
|
| 1523 |
+
},
|
| 1524 |
+
{
|
| 1525 |
+
"epoch": 0.0845,
|
| 1526 |
+
"grad_norm": 0.021259070586902896,
|
| 1527 |
+
"learning_rate": 8.448000000000001e-06,
|
| 1528 |
+
"loss": 2.4212,
|
| 1529 |
+
"step": 4225
|
| 1530 |
+
},
|
| 1531 |
+
{
|
| 1532 |
+
"epoch": 0.085,
|
| 1533 |
+
"grad_norm": 0.021461643865178466,
|
| 1534 |
+
"learning_rate": 8.498e-06,
|
| 1535 |
+
"loss": 2.4242,
|
| 1536 |
+
"step": 4250
|
| 1537 |
+
},
|
| 1538 |
+
{
|
| 1539 |
+
"epoch": 0.0855,
|
| 1540 |
+
"grad_norm": 0.02129278617109427,
|
| 1541 |
+
"learning_rate": 8.548e-06,
|
| 1542 |
+
"loss": 2.4153,
|
| 1543 |
+
"step": 4275
|
| 1544 |
+
},
|
| 1545 |
+
{
|
| 1546 |
+
"epoch": 0.086,
|
| 1547 |
+
"grad_norm": 0.019884381961586706,
|
| 1548 |
+
"learning_rate": 8.598000000000001e-06,
|
| 1549 |
+
"loss": 2.4107,
|
| 1550 |
+
"step": 4300
|
| 1551 |
+
},
|
| 1552 |
+
{
|
| 1553 |
+
"epoch": 0.086,
|
| 1554 |
+
"eval_loss": 2.429638385772705,
|
| 1555 |
+
"eval_runtime": 33.1452,
|
| 1556 |
+
"eval_samples_per_second": 3.53,
|
| 1557 |
+
"eval_steps_per_second": 1.78,
|
| 1558 |
+
"step": 4300
|
| 1559 |
+
},
|
| 1560 |
+
{
|
| 1561 |
+
"epoch": 0.0865,
|
| 1562 |
+
"grad_norm": 0.02127578557225418,
|
| 1563 |
+
"learning_rate": 8.648000000000001e-06,
|
| 1564 |
+
"loss": 2.4202,
|
| 1565 |
+
"step": 4325
|
| 1566 |
+
},
|
| 1567 |
+
{
|
| 1568 |
+
"epoch": 0.087,
|
| 1569 |
+
"grad_norm": 0.021749788475476855,
|
| 1570 |
+
"learning_rate": 8.698e-06,
|
| 1571 |
+
"loss": 2.4274,
|
| 1572 |
+
"step": 4350
|
| 1573 |
+
},
|
| 1574 |
+
{
|
| 1575 |
+
"epoch": 0.0875,
|
| 1576 |
+
"grad_norm": 0.021521494708913836,
|
| 1577 |
+
"learning_rate": 8.748000000000002e-06,
|
| 1578 |
+
"loss": 2.4189,
|
| 1579 |
+
"step": 4375
|
| 1580 |
+
},
|
| 1581 |
+
{
|
| 1582 |
+
"epoch": 0.088,
|
| 1583 |
+
"grad_norm": 0.021276426458537334,
|
| 1584 |
+
"learning_rate": 8.798000000000001e-06,
|
| 1585 |
+
"loss": 2.4152,
|
| 1586 |
+
"step": 4400
|
| 1587 |
+
},
|
| 1588 |
+
{
|
| 1589 |
+
"epoch": 0.088,
|
| 1590 |
+
"eval_loss": 2.4292917251586914,
|
| 1591 |
+
"eval_runtime": 33.1057,
|
| 1592 |
+
"eval_samples_per_second": 3.534,
|
| 1593 |
+
"eval_steps_per_second": 1.782,
|
| 1594 |
+
"step": 4400
|
| 1595 |
+
},
|
| 1596 |
+
{
|
| 1597 |
+
"epoch": 0.0885,
|
| 1598 |
+
"grad_norm": 0.019843371943772815,
|
| 1599 |
+
"learning_rate": 8.848e-06,
|
| 1600 |
+
"loss": 2.421,
|
| 1601 |
+
"step": 4425
|
| 1602 |
+
},
|
| 1603 |
+
{
|
| 1604 |
+
"epoch": 0.089,
|
| 1605 |
+
"grad_norm": 0.02031045171970109,
|
| 1606 |
+
"learning_rate": 8.898000000000002e-06,
|
| 1607 |
+
"loss": 2.4201,
|
| 1608 |
+
"step": 4450
|
| 1609 |
+
},
|
| 1610 |
+
{
|
| 1611 |
+
"epoch": 0.0895,
|
| 1612 |
+
"grad_norm": 0.018642717079241176,
|
| 1613 |
+
"learning_rate": 8.948000000000001e-06,
|
| 1614 |
+
"loss": 2.4171,
|
| 1615 |
+
"step": 4475
|
| 1616 |
+
},
|
| 1617 |
+
{
|
| 1618 |
+
"epoch": 0.09,
|
| 1619 |
+
"grad_norm": 0.021016901396559935,
|
| 1620 |
+
"learning_rate": 8.998000000000001e-06,
|
| 1621 |
+
"loss": 2.4257,
|
| 1622 |
+
"step": 4500
|
| 1623 |
+
},
|
| 1624 |
+
{
|
| 1625 |
+
"epoch": 0.09,
|
| 1626 |
+
"eval_loss": 2.4288113117218018,
|
| 1627 |
+
"eval_runtime": 33.1217,
|
| 1628 |
+
"eval_samples_per_second": 3.532,
|
| 1629 |
+
"eval_steps_per_second": 1.781,
|
| 1630 |
+
"step": 4500
|
| 1631 |
+
},
|
| 1632 |
+
{
|
| 1633 |
+
"epoch": 0.0905,
|
| 1634 |
+
"grad_norm": 0.021595090834222327,
|
| 1635 |
+
"learning_rate": 9.048e-06,
|
| 1636 |
+
"loss": 2.4209,
|
| 1637 |
+
"step": 4525
|
| 1638 |
+
},
|
| 1639 |
+
{
|
| 1640 |
+
"epoch": 0.091,
|
| 1641 |
+
"grad_norm": 0.020500341653961213,
|
| 1642 |
+
"learning_rate": 9.098000000000002e-06,
|
| 1643 |
+
"loss": 2.4093,
|
| 1644 |
+
"step": 4550
|
| 1645 |
+
},
|
| 1646 |
+
{
|
| 1647 |
+
"epoch": 0.0915,
|
| 1648 |
+
"grad_norm": 0.021134665935359346,
|
| 1649 |
+
"learning_rate": 9.148e-06,
|
| 1650 |
+
"loss": 2.4238,
|
| 1651 |
+
"step": 4575
|
| 1652 |
+
},
|
| 1653 |
+
{
|
| 1654 |
+
"epoch": 0.092,
|
| 1655 |
+
"grad_norm": 0.018064298488706988,
|
| 1656 |
+
"learning_rate": 9.198e-06,
|
| 1657 |
+
"loss": 2.4163,
|
| 1658 |
+
"step": 4600
|
| 1659 |
+
},
|
| 1660 |
+
{
|
| 1661 |
+
"epoch": 0.092,
|
| 1662 |
+
"eval_loss": 2.428257465362549,
|
| 1663 |
+
"eval_runtime": 33.451,
|
| 1664 |
+
"eval_samples_per_second": 3.498,
|
| 1665 |
+
"eval_steps_per_second": 1.764,
|
| 1666 |
+
"step": 4600
|
| 1667 |
+
},
|
| 1668 |
+
{
|
| 1669 |
+
"epoch": 0.0925,
|
| 1670 |
+
"grad_norm": 0.019704962175624032,
|
| 1671 |
+
"learning_rate": 9.248e-06,
|
| 1672 |
+
"loss": 2.4082,
|
| 1673 |
+
"step": 4625
|
| 1674 |
+
},
|
| 1675 |
+
{
|
| 1676 |
+
"epoch": 0.093,
|
| 1677 |
+
"grad_norm": 0.019712333508134283,
|
| 1678 |
+
"learning_rate": 9.298e-06,
|
| 1679 |
+
"loss": 2.4089,
|
| 1680 |
+
"step": 4650
|
| 1681 |
+
},
|
| 1682 |
+
{
|
| 1683 |
+
"epoch": 0.0935,
|
| 1684 |
+
"grad_norm": 0.021269463834833153,
|
| 1685 |
+
"learning_rate": 9.348000000000001e-06,
|
| 1686 |
+
"loss": 2.408,
|
| 1687 |
+
"step": 4675
|
| 1688 |
+
},
|
| 1689 |
+
{
|
| 1690 |
+
"epoch": 0.094,
|
| 1691 |
+
"grad_norm": 0.021278662940784676,
|
| 1692 |
+
"learning_rate": 9.398e-06,
|
| 1693 |
+
"loss": 2.4189,
|
| 1694 |
+
"step": 4700
|
| 1695 |
+
},
|
| 1696 |
+
{
|
| 1697 |
+
"epoch": 0.094,
|
| 1698 |
+
"eval_loss": 2.4279165267944336,
|
| 1699 |
+
"eval_runtime": 33.1606,
|
| 1700 |
+
"eval_samples_per_second": 3.528,
|
| 1701 |
+
"eval_steps_per_second": 1.779,
|
| 1702 |
+
"step": 4700
|
| 1703 |
+
},
|
| 1704 |
+
{
|
| 1705 |
+
"epoch": 0.0945,
|
| 1706 |
+
"grad_norm": 0.018504564797986272,
|
| 1707 |
+
"learning_rate": 9.448e-06,
|
| 1708 |
+
"loss": 2.4254,
|
| 1709 |
+
"step": 4725
|
| 1710 |
+
},
|
| 1711 |
+
{
|
| 1712 |
+
"epoch": 0.095,
|
| 1713 |
+
"grad_norm": 0.01917099113509997,
|
| 1714 |
+
"learning_rate": 9.498000000000001e-06,
|
| 1715 |
+
"loss": 2.411,
|
| 1716 |
+
"step": 4750
|
| 1717 |
+
},
|
| 1718 |
+
{
|
| 1719 |
+
"epoch": 0.0955,
|
| 1720 |
+
"grad_norm": 0.019097394482211122,
|
| 1721 |
+
"learning_rate": 9.548e-06,
|
| 1722 |
+
"loss": 2.4209,
|
| 1723 |
+
"step": 4775
|
| 1724 |
+
},
|
| 1725 |
+
{
|
| 1726 |
+
"epoch": 0.096,
|
| 1727 |
+
"grad_norm": 0.020220692469392707,
|
| 1728 |
+
"learning_rate": 9.598e-06,
|
| 1729 |
+
"loss": 2.4066,
|
| 1730 |
+
"step": 4800
|
| 1731 |
+
},
|
| 1732 |
+
{
|
| 1733 |
+
"epoch": 0.096,
|
| 1734 |
+
"eval_loss": 2.4273650646209717,
|
| 1735 |
+
"eval_runtime": 33.1079,
|
| 1736 |
+
"eval_samples_per_second": 3.534,
|
| 1737 |
+
"eval_steps_per_second": 1.782,
|
| 1738 |
+
"step": 4800
|
| 1739 |
+
},
|
| 1740 |
+
{
|
| 1741 |
+
"epoch": 0.0965,
|
| 1742 |
+
"grad_norm": 0.019607148490934756,
|
| 1743 |
+
"learning_rate": 9.648000000000001e-06,
|
| 1744 |
+
"loss": 2.4132,
|
| 1745 |
+
"step": 4825
|
| 1746 |
+
},
|
| 1747 |
+
{
|
| 1748 |
+
"epoch": 0.097,
|
| 1749 |
+
"grad_norm": 0.019388710503851023,
|
| 1750 |
+
"learning_rate": 9.698000000000001e-06,
|
| 1751 |
+
"loss": 2.4096,
|
| 1752 |
+
"step": 4850
|
| 1753 |
+
},
|
| 1754 |
+
{
|
| 1755 |
+
"epoch": 0.0975,
|
| 1756 |
+
"grad_norm": 0.019593746411763164,
|
| 1757 |
+
"learning_rate": 9.748e-06,
|
| 1758 |
+
"loss": 2.4064,
|
| 1759 |
+
"step": 4875
|
| 1760 |
+
},
|
| 1761 |
+
{
|
| 1762 |
+
"epoch": 0.098,
|
| 1763 |
+
"grad_norm": 0.018761734791343965,
|
| 1764 |
+
"learning_rate": 9.798e-06,
|
| 1765 |
+
"loss": 2.4033,
|
| 1766 |
+
"step": 4900
|
| 1767 |
+
},
|
| 1768 |
+
{
|
| 1769 |
+
"epoch": 0.098,
|
| 1770 |
+
"eval_loss": 2.4270286560058594,
|
| 1771 |
+
"eval_runtime": 33.0269,
|
| 1772 |
+
"eval_samples_per_second": 3.543,
|
| 1773 |
+
"eval_steps_per_second": 1.786,
|
| 1774 |
+
"step": 4900
|
| 1775 |
+
},
|
| 1776 |
+
{
|
| 1777 |
+
"epoch": 0.0985,
|
| 1778 |
+
"grad_norm": 0.018964507342139367,
|
| 1779 |
+
"learning_rate": 9.848000000000001e-06,
|
| 1780 |
+
"loss": 2.4211,
|
| 1781 |
+
"step": 4925
|
| 1782 |
+
},
|
| 1783 |
+
{
|
| 1784 |
+
"epoch": 0.099,
|
| 1785 |
+
"grad_norm": 0.01858861943184826,
|
| 1786 |
+
"learning_rate": 9.898e-06,
|
| 1787 |
+
"loss": 2.4032,
|
| 1788 |
+
"step": 4950
|
| 1789 |
+
},
|
| 1790 |
+
{
|
| 1791 |
+
"epoch": 0.0995,
|
| 1792 |
+
"grad_norm": 0.01821023564956819,
|
| 1793 |
+
"learning_rate": 9.948e-06,
|
| 1794 |
+
"loss": 2.4031,
|
| 1795 |
+
"step": 4975
|
| 1796 |
+
},
|
| 1797 |
+
{
|
| 1798 |
+
"epoch": 0.1,
|
| 1799 |
+
"grad_norm": 0.018839474555921314,
|
| 1800 |
+
"learning_rate": 9.998000000000002e-06,
|
| 1801 |
+
"loss": 2.4112,
|
| 1802 |
+
"step": 5000
|
| 1803 |
+
},
|
| 1804 |
+
{
|
| 1805 |
+
"epoch": 0.1,
|
| 1806 |
+
"eval_loss": 2.426590919494629,
|
| 1807 |
+
"eval_runtime": 33.0133,
|
| 1808 |
+
"eval_samples_per_second": 3.544,
|
| 1809 |
+
"eval_steps_per_second": 1.787,
|
| 1810 |
+
"step": 5000
|
| 1811 |
}
|
| 1812 |
],
|
| 1813 |
"logging_steps": 25,
|
|
|
|
| 1827 |
"attributes": {}
|
| 1828 |
}
|
| 1829 |
},
|
| 1830 |
+
"total_flos": 1.3926509625684787e+19,
|
| 1831 |
"train_batch_size": 1,
|
| 1832 |
"trial_name": null,
|
| 1833 |
"trial_params": null
|