examples/load_model_test.py

import os
import torch
from transformers import BertConfig, BertModel, BertForMaskedLM, DNATokenizer
import argparse

# Define MODEL_CLASSES as it's required by your loadmodel function
MODEL_CLASSES = {
    "dna": (BertConfig, BertForMaskedLM, DNATokenizer),
    # ... (other classes omitted for brevity)
}

def loadmodel(model_dir):
    config_class, model_class, tokenizer_class = MODEL_CLASSES['dna'] # Changed 'DNA' to 'dna' for Python keys
    print(f"Loading using: {config_class.__name__}, {model_class.__name__}, {tokenizer_class.__name__}")
    
    # 1. Load Configuration
    config = config_class.from_pretrained(
            model_dir,
            cache_dir = None,
        )
        
    # 2. Load Model Weights
    # NOTE: Since you are extracting embeddings, we should use BertModel, not BertForMaskedLM
    # BertModel is the base transformer without the MLM head.
    base_model_class = BertModel if model_class == BertForMaskedLM else model_class
    
    model = base_model_class.from_pretrained(
            model_dir,
            from_tf=bool(".ckpt" in model_dir),
            config=config,
            cache_dir= None,
        )    
        
    # 3. Set Device
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model.to(device)
    model.eval() # Set model to evaluation mode
    print(f"Model loaded onto device: {device}")

    # 4. Load Tokenizer (using custom environment variables)
    #tokenizer_class.vocab_files_names = {"vocab_file": os.getenv("VOCAB_NAME")}
    #tokenizer_class.pretrained_vocab_files_map = {"vocab_file": {'dna': os.getenv("VOCAB_PATH")}} # Use 'dna' key
    tokenizer = tokenizer_class.from_pretrained(model_dir)
    print(f"Tokenizer vocabulary size: {len(tokenizer)}")
    
    return config, model, tokenizer

# --- Main Call ---
# Use the environment variable set in the shell as the model directory
parser = argparse.ArgumentParser()
parser.add_argument("--MODEL_DIR", type=str, required=True)
args = parser.parse_args()

model_dir = args.MODEL_DIR

if model_dir != "/path/to/default":
    config, model, tokenizer = loadmodel(model_dir)
    print("Model and Tokenizer loaded successfully.")
    
    embedding_layer = model.get_input_embeddings()
    print(embedding_layer.weight.shape)
    

    seq = "ACGTACGTACGT"
    tokens = tokenizer.tokenize(" ".join([seq[i:i+6] for i in range(len(seq)-5)]))
    print(tokens[:10])
else:
    print("Error: MODEL_DIR environment variable was not set.")