423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-14 10:56:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor mllm param logic

Browse Source 
Former-commit-id: f6f630a1c9
This commit is contained in:
hiyouga
2025-01-10 15:41:54 +00:00
parent b3561ae552
commit c89d17ab63
10 changed files with 198 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
examples/train_full/qwen2vl_full_sft.yaml
View File

@@ -7,6 +7,7 @@ stage: sft
do_train: true
finetuning_type: full
freeze_vision_tower: true # choices: [true, false]
freeze_multi_modal_projector: true # choices: [true, false]
train_mm_proj_only: false # choices: [true, false]
deepspeed: examples/deepspeed/ds_z3_config.json # choices: [ds_z0_config.json, ds_z2_config.json, ds_z3_config.json]
@@ -29,7 +30,7 @@ overwrite_output_dir: true
per_device_train_batch_size: 1
gradient_accumulation_steps: 2
learning_rate: 1.0e-5
num_train_epochs: 30.0
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true