scripts/slurm/run_inference_single.sh

#!/usr/bin/env bash
set -euo pipefail

# Single deterministic inference for one protein sequence.
# Run this inside an allocated/interactive GPU session. No SLURM resources are requested here.

PROJECT_DIR="${PROJECT_DIR:-/public/home/scnb9biwet/jiangqq/CodonTransformer-main}"
HF_HOME="${HF_HOME:-/public/home/scnb9biwet/.cache/huggingface}"
CONDA_ENV="${CONDA_ENV:-struct-evo}"

PROTEIN="${PROTEIN:-MFWY}"
ORGANISM="${ORGANISM:-Escherichia coli general}"
OFFLINE="${OFFLINE:-1}"

cd "${PROJECT_DIR}"

export HF_HOME
export PYTHONPATH="${PROJECT_DIR}/model:${PYTHONPATH:-}"
export PROTEIN
export ORGANISM
export OFFLINE
export PYTHONFAULTHANDLER=1

if [[ "${OFFLINE}" == "1" ]]; then
    export HF_HUB_OFFLINE=1
    export TRANSFORMERS_OFFLINE=1
fi

if [[ -n "${CONDA_ENV}" ]] && command -v conda >/dev/null 2>&1; then
    # shellcheck disable=SC1091
    source "$(conda info --base)/etc/profile.d/conda.sh"
    conda activate "${CONDA_ENV}"
fi

python - <<'PY'
import os

import torch
from transformers import AutoTokenizer, BigBirdForMaskedLM

from CodonTransformer.CodonJupyter import format_model_output
from CodonTransformer.CodonPrediction import predict_dna_sequence

protein = os.environ["PROTEIN"]
organism = os.environ["ORGANISM"]
local_files_only = os.environ.get("OFFLINE", "1") == "1"

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"HF_HOME: {os.environ.get('HF_HOME')}")
print(f"Device: {device}")
print(f"Local files only: {local_files_only}")

tokenizer = AutoTokenizer.from_pretrained(
    "adibvafa/CodonTransformer",
    local_files_only=local_files_only,
)
model = BigBirdForMaskedLM.from_pretrained(
    "adibvafa/CodonTransformer",
    local_files_only=local_files_only,
).to(device)

output = predict_dna_sequence(
    protein=protein,
    organism=organism,
    device=device,
    tokenizer=tokenizer,
    model=model,
    attention_type="original_full",
    deterministic=True,
)

print(format_model_output(output))
PY