refactor : metric 日志改为以 optimizer step 为单位,默认每步记录

- log_interval 默认 100 -> 1,语义从 batch iteration 改为 optimizer step
- step 指标从 on_batch_end 移到 on_optimizer_step,不受梯度累积影响
- JSONL 条目新增 step 字段,保留 iter
- flush 落盘仍在 on_batch_end
This commit is contained in:
2026-06-30 15:12:31 +08:00
parent 0f1fcb079f
commit 44579ea6dc
3 changed files with 13 additions and 9 deletions
+2 -2
View File
@@ -162,8 +162,8 @@ def parse_args() -> argparse.Namespace:
parser.add_argument(
"--log_interval",
type=int,
default=100,
help="Number of batch iterations between metric logs.",
default=1,
help="Number of optimizer steps between metric logs.",
)
parser.add_argument(
"--grpo_sync_interval",