Upload qwen2_5_vl_full_sft.yaml

sft_related/qwen2_5_vl_full_sft.yaml

@@ -1,6 +1,6 @@
 ### model
 model_name_or_path: Qwen/Qwen2.5-VL-3B-Instruct
-image_max_pixels: 262144
+image_max_pixels: 12845056
 video_max_pixels: 16384
 trust_remote_code: true
 
@@ -8,16 +8,16 @@ trust_remote_code: true
 stage: sft
 do_train: true
 finetuning_type: full
-freeze_vision_tower: true  # choices: [true, false]
-freeze_multi_modal_projector: true  # choices: [true, false]
+freeze_vision_tower: false  # choices: [true, false]
+freeze_multi_modal_projector: false  # choices: [true, false]
 train_mm_proj_only: false  # choices: [true, false]
 deepspeed: examples/deepspeed/ds_z2_config.json  # choices: [ds_z0_config.json, ds_z2_config.json, ds_z3_config.json]
 
 ### dataset
 dataset: mllm_rec_json
 template: qwen2_vl
-cutoff_len: 2048
-max_samples: 1000000
+cutoff_len: 131072
+max_samples: 10000000
 overwrite_cache: true
 preprocessing_num_workers: 16
 
@@ -31,8 +31,8 @@ overwrite_output_dir: true
 ### train
 per_device_train_batch_size: 1
 gradient_accumulation_steps: 2
-learning_rate: 2.0e-5
-num_train_epochs: 3.0
+learning_rate: 1.0e-6
+num_train_epochs: 2.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
 bf16: true