source_code/sam3/medsam3_brats/run_train_lora.sh

#!/bin/bash
# SAM3 LoRA 微调训练脚本 - BraTS2023

# 配置路径
DATA_ROOT="/data/yty/brats2023/ASNR-MICCAI-BraTS2023-GLI-Challenge-TrainingData"
SAM3_CHECKPOINT="/data/yty/sam3/sam3.pt"
OUTPUT_DIR="/data/yty/brats23_sam3_lora_output"

# 训练参数
MODALITY=0          # 0=t1c, 1=t1n, 2=t2f, 3=t2w
TARGET_SIZE="512 512"
BATCH_SIZE=4
GRAD_ACCUM=4        # 有效batch size = BATCH_SIZE * GRAD_ACCUM = 16
EPOCHS=50
LR=1e-4

# LoRA参数
LORA_RANK=8
LORA_ALPHA=16
LORA_DROPOUT=0.1

# 其他
NUM_WORKERS=4
VAL_FREQ=5
SEED=42

echo "============================================"
echo "SAM3 LoRA Fine-tuning for BraTS2023"
echo "============================================"
echo ""
echo "Configuration:"
echo "  Data root: $DATA_ROOT"
echo "  Checkpoint: $SAM3_CHECKPOINT"
echo "  Output: $OUTPUT_DIR"
echo "  LoRA rank: $LORA_RANK"
echo "  LoRA alpha: $LORA_ALPHA"
echo "  Batch size: $BATCH_SIZE x $GRAD_ACCUM = $((BATCH_SIZE * GRAD_ACCUM))"
echo "  Learning rate: $LR"
echo "  Epochs: $EPOCHS"
echo ""

# 创建输出目录
mkdir -p "$OUTPUT_DIR"

# 运行训练
cd /root/githubs/sam3/medsam3_brats

python train_lora.py \
    --data_root "$DATA_ROOT" \
    --modality $MODALITY \
    --target_size $TARGET_SIZE \
    --dataset_type image \
    --checkpoint "$SAM3_CHECKPOINT" \
    --lora_rank $LORA_RANK \
    --lora_alpha $LORA_ALPHA \
    --lora_dropout $LORA_DROPOUT \
    --epochs $EPOCHS \
    --batch_size $BATCH_SIZE \
    --lr $LR \
    --grad_accum $GRAD_ACCUM \
    --num_workers $NUM_WORKERS \
    --output_dir "$OUTPUT_DIR" \
    --val_freq $VAL_FREQ \
    --seed $SEED \
    2>&1 | tee "$OUTPUT_DIR/train.log"

echo ""
echo "============================================"
echo "Training completed!"
echo "Output saved to: $OUTPUT_DIR"
echo "============================================"