423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-14 19:06:26 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor mm training

Browse Source 
Former-commit-id: 3382317e32
This commit is contained in:
hiyouga
2024-08-30 02:14:31 +08:00
parent 98b0c7530c
commit a83756b5e9
32 changed files with 505 additions and 472 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
examples/train_lora/qwen2vl_lora_sft.yaml
View File

@@ -1,5 +1,5 @@
### model
model_name_or_path: qwen2-vl-hf/qwen2-vl-7b-hf
model_name_or_path: Qwen/Qwen2-VL-7B-Instruct
visual_inputs: true
### method
@@ -9,23 +9,23 @@ finetuning_type: lora
lora_target: all
### dataset
dataset: qwen2vl_demo
template: qwen2vl
dataset: mllm_demo
template: qwen2_vl
cutoff_len: 1024
max_samples: 1000
overwrite_cache: true
preprocessing_num_workers: 16
### output
output_dir: saves/qwen2-vl-7b/lora/sft
output_dir: saves/qwen2_vl-7b/lora/sft
logging_steps: 10
save_steps: 500
plot_loss: true
overwrite_output_dir: true
### train
per_device_train_batch_size: 2
gradient_accumulation_steps: 1
per_device_train_batch_size: 1
gradient_accumulation_steps: 8
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine