 Kai Wu
|
de3e32c8a7
fix typo in auto wrap policy (#793)
|
hai 10 meses |
 Guanghui Qin
|
a62aff3876
fix typo in auto wrap policy
|
hai 10 meses |
 JimChienTW
|
d31ee18e4f
Print model's frozen status after freezing
|
hai 10 meses |
|
JimChienTW
|
d1195a6fd8
Fix model parameter mismatch by printing parameters before FSDP
|
hai 10 meses |
|
JimChienTW
|
21e8368c7e
add freeze_LLM_only option for mllama finetuning
|
hai 10 meses |
|
Kai Wu
|
bd31680eb9
fixed the missing processor after conversion
|
hai 11 meses |
 Sanyam Bhutani
|
b554b24b6e
Fix/unit test 3.2 (#726)
|
hai 11 meses |
 Patrik Lambert
|
f521c93d56
Fix numpy seed in finetuning.py
|
hai 11 meses |
 Matthias Reso
|
1090ccf455
Fix test_finetuning
|
hai 11 meses |
 Huang Zhihong
|
a620831762
Fix the bug when continue the peft. (#717)
|
hai 11 meses |
|
Kai Wu
|
9c7a5b421f
fix AutoModel and bump transformers version to 4.45 (#686)
|
hai 11 meses |
|
Kai Wu
|
57afa0b51e
use AutoModel
|
hai 11 meses |
|
Kai Wu
|
2730bcaab7
fix readme and fsdp logic
|
hai 11 meses |
|
Kai Wu
|
50dff0b78e
gradient_checkpointing_enable()
|
hai 11 meses |
|
Kai Wu
|
c18a0d277f
changed dataset to ocrvqa
|
hai 11 meses |
|
Kai Wu
|
bd22f407d5
changed to aid2 dataset
|
hai 11 meses |
|
Kai Wu
|
1a76080807
lora+fsdp working
|
hai 1 ano |
|
Kai Wu
|
8a11b48022
lora+fsdp not working
|
hai 1 ano |
|
Kai Wu
|
79dbe05a94
batch fine-tuning lmm working
|
hai 1 ano |
|
Kai Wu
|
ce299b3439
add get_custom_data_collator feature
|
hai 1 ano |
|
Kai Wu
|
12da109823
Merge branch 'main' into lmm_finetune
|
hai 1 ano |
|
Kai Wu
|
bb990be967
not working, need create dataloader function
|
hai 1 ano |
|
Matthias Reso
|
778e31e35c
Fix checkpoint saving (#650)
|
hai 1 ano |
|
Kai Wu
|
ee204ccb98
working now
|
hai 1 ano |
|
Kai Wu
|
b566582a86
finetune not working with fsdp
|
hai 1 ano |
 Hamid Shojanazeri
|
808a3f7a0c
Adding support for FSDP+Qlora. (#572)
|
hai 1 ano |
 haozhx23
|
e6b0f97199
Fix hsdp_device_mesh=None when enable HSDP and HYBRID_SHARD (#402)
|
hai 1 ano |
|
Kai Wu
|
41a46d811d
fix alpaca dataset by using 5% of the data as eval and make sure len((eval_loader)>0
|
hai 1 ano |
|
Kai Wu
|
f1d90d0ff0
fix wandb config update
|
hai 1 ano |
|
Kai Wu
|
480c4f2b5e
resume the finetuning given the path of the previous peft checkpoint folder
|
hai 1 ano |
