王路宝
66ad1c4b7f
fix: add find_unused_parameters=True to DDP for mixed text-image batches
...
When a batch contains only text samples (pixel_values is None),
vision_proj parameters do not participate in the forward pass,
causing DDP to raise:
'Expected to have finished reduction in the prior iteration'
Adding find_unused_parameters=True resolves this.
2026-04-01 14:20:30 +08:00
jingyaogong
e51dcf0fda
[update] minimind-3v
2026-03-26 00:09:02 +08:00
jingyaogong
33db76fe10
[update] dataset name
2026-02-05 00:22:58 +08:00
jingyaogong
7e094afa59
[update] shuffle data
2026-01-18 17:02:24 +08:00
jingyaogong
326b24727e
[update] sync minimind
2026-01-15 21:51:58 +08:00
jingyaogong
3684285a91
[feat] support parquet data
2025-12-27 10:49:50 +08:00
jingyaogong
32cf4c5c01
[fix] add gc
2025-12-27 01:09:33 +08:00
jingyaogong
13c3a2c093
[fix] cuda memory
2025-12-01 16:35:05 +08:00
jingyaogong
a157b66ace
[feat] del cuda memory
2025-11-27 19:37:39 +08:00
jingyaogong
7605850e1d
[feat] update args
2025-10-30 10:50:54 +08:00
jingyaogong
778aeb2946
[feat] update trainer
2025-10-28 23:34:16 +08:00
jingyaogong
2ce5c31170
[feat] update sft
2025-10-26 21:24:19 +08:00
jingyaogong
aebc6fb191
[feat] pause-training
2025-10-26 20:54:13 +08:00
jingyaogong
4c3b5bacbf
[fix] 1024-bugs
2025-10-24 00:35:56 +08:00
jingyaogong
623a130c34
250427
2025-04-27 18:13:22 +08:00