Qwen2.5-VL-3B-SFT
Qwen2.5-VL-3B-SFT-wo_single_turn-7k
Training Data: PyVision-SFT
Filtered single turn trajactory
learning_rate: 1.0e-5
lr_scheduler_type: cosine
Qwen2.5-VL-3B-SFT-wo_single_turn-7k-train-1epoch-0.1warm
num_train_epochs: 1.0
warmup_ratio: 0.1
equivalent_batchsize: 16
Qwen2.5-VL-3B-SFT-wo_single_turn-7k-train-1epoch-0.2warm
num_train_epochs: 1
warmup_ratio: 0.2
equivalent_batchsize: 16
Qwen2.5-VL-3B-SFT-wo_single_turn-7k-train-1epoch-8bs-0.1warm
num_train_epochs: 1
warmup_ratio: 0.1
equivalent_batchsize: 8
Qwen2.5-VL-3B-SFT-wo_single_turn-7k-train-5epoch-0.1warm
num_train_epochs: 5
warmup_ratio: 0.1
equivalent_batchsize: 16