transfer/code/CCFM/eval_joint_generate.sh

#!/bin/bash
#PJM -L rscgrp=b-batch
#PJM -L gpu=1
#PJM -L elapse=3:00:00
#PJM -j
#PJM -S
#PJM -o /home/hp250092/ku50001222/qian/aivc/lfj/transfer/logs/ccfm_eval_joint_%j.out

# Evaluate CCFM with joint-generation inference (LatentForcing style)
# using the checkpoint from two-stage ODE training.

source /home/hp250092/ku50001222/qian/aivc/lfj/stack_env/bin/activate
cd /home/hp250092/ku50001222/qian/aivc/lfj/transfer/code/CCFM

export PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:256

echo "=========================================="
echo "Job ID:    $PJM_JOBID"
echo "Start:     $(date)"
echo "Node:      $(hostname)"
echo "Eval mode: joint-generate (LatentForcing style)"
echo "=========================================="

CHECKPOINT="/home/hp250092/ku50001222/qian/aivc/lfj/transfer/code/CCFM/result/ccfm-fusion_differential_perceiver-norman-cascaded-predict_y-gamma_0.5-perturbation_function_crisper-lr_5e-05-dim_model_128-infer_top_gene_1000-split_method_additive-use_mmd_loss_True-fold_1-latent_weight_1.0-choose_latent_p_0.4/iteration_110000/checkpoint.pt"

python scripts/run_cascaded.py \
    --data-name norman \
    --d-model 128 \
    --nhead 8 \
    --nlayers 4 \
    --batch-size 128 \
    --lr 5e-5 \
    --steps 200000 \
    --fusion-method differential_perceiver \
    --perturbation-function crisper \
    --noise-type Gaussian \
    --infer-top-gene 1000 \
    --n-top-genes 5000 \
    --use-mmd-loss \
    --gamma 0.5 \
    --split-method additive \
    --fold 1 \
    --scgpt-dim 512 \
    --bottleneck-dim 128 \
    --latent-weight 1.0 \
    --choose-latent-p 0.4 \
    --dh-depth 2 \
    --latent-steps 20 \
    --expr-steps 20 \
    --result-path ./result_joint_generate \
    --checkpoint-path "$CHECKPOINT" \
    --test-only

echo "=========================================="
echo "Finished:  $(date)"
echo "=========================================="