File size: 6,564 Bytes
ee6da62 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 | #!/bin/bash
# Multi-Target TD3B Training Launch Script
# Trains TD3B on multiple protein targets with random sampling strategy
# ============================================================================
# Configuration
# ============================================================================
# Paths — update these to your local paths
BASE_PATH="/path/to/TD3B"
PRETRAINED_CHECKPOINT="${BASE_PATH}/checkpoints/pretrained.ckpt"
TRAIN_CSV="${BASE_PATH}/data/train.csv"
VAL_CSV="${BASE_PATH}/data/test.csv" # Optional: create validation split
# Run configuration
RUN_NAME="multi_target_td3b" # Timestamp will be added automatically
DEVICE="cuda:0"
# Multi-target sampling
TARGETS_PER_MCTS=2 # Number of targets sampled per MCTS round (K)
RESAMPLE_TARGETS_EVERY=1 # Resample targets every N epochs
# Training hyperparameters
NUM_EPOCHS=200
LEARNING_RATE=3e-4
TRAIN_BATCH_SIZE=1 # Small batch size to prevent OOM
GRADIENT_ACCUMULATION_STEPS=32 # Effective batch size = 16 * 4 = 64
RESAMPLE_EVERY=10 # Run MCTS every N epochs
SAVE_EVERY=20
VALIDATE_EVERY=20
RESET_TREE_EVERY=50
# MCTS hyperparameters (aligned with v1, but can reduce for multi-target)
NUM_ITER=20 # MCTS iterations per resample (v1 default: 50, reduced for multi-target)
NUM_CHILDREN=16 # Children per MCTS expansion
BUFFER_SIZE=50 # Pareto buffer size (v1 default: 50)
REPLAY_BUFFER_SIZE=1000 # Recommended range: 500-5000 (0 disables replay)
REPLAY_BUFFER_STRATEGY="fifo" # fifo or random
ALPHA=0.1 # Temperature for importance weighting
EXPLORATION=1.0 # UCB exploration constant
# TD3B hyperparameters (aligned with v1 defaults)
CONTRASTIVE_WEIGHT=0.1 # v1 default: 0.1
CONTRASTIVE_MARGIN=1.0
KL_BETA=0.1 # v1 default: 0.1
MIN_AFFINITY_THRESHOLD=0.0 # CRITICAL: minimum affinity for allosteric control
SIGMOID_TEMPERATURE=0.1
# Validation
VAL_SAMPLES_PER_TARGET=20 # Number of sequences per target during validation
# Directional oracle (GPCR classifier)
ORACLE_CKPT="${BASE_PATH}/checkpoints/direction_oracle.pt"
ORACLE_TR2D2_CHECKPOINT="${BASE_PATH}/checkpoints/pretrained.ckpt"
ORACLE_TOKENIZER_VOCAB="${BASE_PATH}/tokenizer/new_vocab.txt"
ORACLE_TOKENIZER_SPLITS="${BASE_PATH}/tokenizer/new_splits.txt"
ORACLE_ESM_NAME="facebook/esm2_t33_650M_UR50D"
ORACLE_ESM_CACHE_DIR="" # Optional: set to a cache dir path
ORACLE_ESM_LOCAL_FILES_ONLY=0 # Set to 1 to avoid network access
ORACLE_MAX_LIGAND_LENGTH=768
ORACLE_MAX_PROTEIN_LENGTH=1024
ORACLE_D_MODEL=256
ORACLE_N_HEADS=4
ORACLE_N_SELF_ATTN_LAYERS=1
ORACLE_N_BMCA_LAYERS=2
ORACLE_DROPOUT=0.3
EXTRA_ORACLE_ARGS=""
if [ -n "$ORACLE_ESM_CACHE_DIR" ]; then
EXTRA_ORACLE_ARGS="$EXTRA_ORACLE_ARGS --direction_oracle_esm_cache_dir $ORACLE_ESM_CACHE_DIR"
fi
if [ "$ORACLE_ESM_LOCAL_FILES_ONLY" -eq 1 ]; then
EXTRA_ORACLE_ARGS="$EXTRA_ORACLE_ARGS --direction_oracle_esm_local_files_only"
fi
# W&B (optional)
WANDB_PROJECT="tr2d2-multi-target"
WANDB_ENTITY="phos_zj"
# ============================================================================
# Launch Training
# ============================================================================
cd ${BASE_PATH}
echo "============================================================================"
echo "Multi-Target TD3B Training"
echo "============================================================================"
echo "Configuration:"
echo " - Targets per MCTS: ${TARGETS_PER_MCTS}"
echo " - Training batch size: ${TRAIN_BATCH_SIZE}"
echo " - Gradient accumulation: ${GRADIENT_ACCUMULATION_STEPS}"
echo " - Effective batch size: $((TRAIN_BATCH_SIZE * GRADIENT_ACCUMULATION_STEPS))"
echo " - Epochs: ${NUM_EPOCHS}"
echo " - MCTS iterations: ${NUM_ITER}"
echo " - MCTS children: ${NUM_CHILDREN}"
echo " - Buffer size: ${BUFFER_SIZE}"
echo " - Replay buffer size: ${REPLAY_BUFFER_SIZE} (${REPLAY_BUFFER_STRATEGY})"
echo "============================================================================"
echo ""
# Build command
CMD="python finetune_multi_target.py \
--base_path ${BASE_PATH} \
--train_csv ${TRAIN_CSV} \
--pretrained_checkpoint ${PRETRAINED_CHECKPOINT} \
--run_name ${RUN_NAME} \
--device ${DEVICE} \
\
--targets_per_mcts ${TARGETS_PER_MCTS} \
--resample_targets_every ${RESAMPLE_TARGETS_EVERY} \
\
--num_epochs ${NUM_EPOCHS} \
--learning_rate ${LEARNING_RATE} \
--train_batch_size ${TRAIN_BATCH_SIZE} \
--gradient_accumulation_steps ${GRADIENT_ACCUMULATION_STEPS} \
--resample_every_n_step ${RESAMPLE_EVERY} \
--save_every_n_epochs ${SAVE_EVERY} \
--validate_every_n_epochs ${VALIDATE_EVERY} \
--reset_every_n_step ${RESET_TREE_EVERY} \
\
--num_iter ${NUM_ITER} \
--num_children ${NUM_CHILDREN} \
--buffer_size ${BUFFER_SIZE} \
--replay_buffer_size ${REPLAY_BUFFER_SIZE} \
--replay_buffer_strategy ${REPLAY_BUFFER_STRATEGY} \
--alpha ${ALPHA} \
--exploration ${EXPLORATION} \
\
--contrastive_weight ${CONTRASTIVE_WEIGHT} \
--contrastive_margin ${CONTRASTIVE_MARGIN} \
--kl_beta ${KL_BETA} \
--min_affinity_threshold ${MIN_AFFINITY_THRESHOLD} \
--sigmoid_temperature ${SIGMOID_TEMPERATURE} \
\
--direction_oracle_ckpt ${ORACLE_CKPT} \
--direction_oracle_tr2d2_checkpoint ${ORACLE_TR2D2_CHECKPOINT} \
--direction_oracle_tokenizer_vocab ${ORACLE_TOKENIZER_VOCAB} \
--direction_oracle_tokenizer_splits ${ORACLE_TOKENIZER_SPLITS} \
--direction_oracle_esm_name ${ORACLE_ESM_NAME} \
--direction_oracle_max_ligand_length ${ORACLE_MAX_LIGAND_LENGTH} \
--direction_oracle_max_protein_length ${ORACLE_MAX_PROTEIN_LENGTH} \
--direction_oracle_d_model ${ORACLE_D_MODEL} \
--direction_oracle_n_heads ${ORACLE_N_HEADS} \
--direction_oracle_n_self_attn_layers ${ORACLE_N_SELF_ATTN_LAYERS} \
--direction_oracle_n_bmca_layers ${ORACLE_N_BMCA_LAYERS} \
--direction_oracle_dropout ${ORACLE_DROPOUT} \
${EXTRA_ORACLE_ARGS} \
\
--val_samples_per_target ${VAL_SAMPLES_PER_TARGET} \
\
--grad_clip \
--gradnorm_clip 1.0 \
--wandb_project ${WANDB_PROJECT}"
# Add validation CSV if it exists
if [ -f "${VAL_CSV}" ]; then
CMD="${CMD} --val_csv ${VAL_CSV}"
echo "Validation CSV: ${VAL_CSV}"
else
echo "No validation CSV found (${VAL_CSV})"
echo "Skipping validation during training"
fi
# Add W&B entity if specified
if [ -n "${WANDB_ENTITY}" ]; then
CMD="${CMD} --wandb_entity ${WANDB_ENTITY}"
fi
echo ""
echo "Launching training..."
echo ""
# Execute
eval $CMD
|