Update app.py

app.py

@@ -1,3 +1,4 @@
+import spaces  # 🚨 MUST BE LINE 1. Fixes the "CUDA Initialized" error!
 import gradio as gr
 import os
 import re
@@ -11,25 +12,24 @@ from Bio.Blast import NCBIWWW, NCBIXML
 # 🚨 Colab Native Imports for the Phenotype model
 from transformers import AutoTokenizer, AutoModelForSequenceClassification, AutoModel, BertTokenizer, BertForSequenceClassification, AutoConfig
 from huggingface_hub import hf_hub_download
-import spaces 
 
 # ===================================
-# 1. LOAD AI MODELS (GLOBALLY CACHED ON CPU)
+# 1. LOAD AI MODELS (STREAMLIT RAW METHOD)
 # ===================================
-print("Waking up the Genomic Oracle... Loading models safely into CPU RAM.\n")
+print("Waking up the Genomic Oracle... Loading raw models to CPU (Streamlit Style).\n")
 
 # A. Kadir's Gatekeeper 
 clf_coding = joblib.load("coding_classifier_universal.joblib")
 
 # B. Base DNABERT 
 tokenizer_base = AutoTokenizer.from_pretrained("DNABERT_Local", trust_remote_code=True)
-# 🚨 _fast_init=False prevents the Meta Device crash during UI Build
-model_base = AutoModel.from_pretrained("DNABERT_Local", trust_remote_code=True, _fast_init=False)
+# 🚨 device_map="cpu" blocks the meta device and perfectly mimics Streamlit's raw loading!
+model_base = AutoModel.from_pretrained("DNABERT_Local", trust_remote_code=True, device_map="cpu")
 model_base.eval()
 
 # C. DNABERT-2 Promoter Model 
 tokenizer_promoter = AutoTokenizer.from_pretrained("llm_promoter_classifier_v2", trust_remote_code=True)
-model_promoter = AutoModelForSequenceClassification.from_pretrained("llm_promoter_classifier_v2", trust_remote_code=True, _fast_init=False)
+model_promoter = AutoModelForSequenceClassification.from_pretrained("llm_promoter_classifier_v2", trust_remote_code=True, device_map="cpu")
 model_promoter.eval()
 
 # D. Multi-Feature LightGBM 
@@ -39,7 +39,7 @@ lightgbm_model = joblib.load(lgbm_path)
 # E. Custom Lean/Obese Phenotype BERT (🚨 Forced Native Architecture via Colab Fix!)
 tokenizer_pheno = BertTokenizer.from_pretrained("Geonomic/Genomic-Oracle-Weights", do_lower_case=False)
 config_pheno = AutoConfig.from_pretrained("Geonomic/Genomic-Oracle-Weights", trust_remote_code=True)
-model_pheno = BertForSequenceClassification.from_pretrained("Geonomic/Genomic-Oracle-Weights", config=config_pheno, _fast_init=False)
+model_pheno = BertForSequenceClassification.from_pretrained("Geonomic/Genomic-Oracle-Weights", config=config_pheno, device_map="cpu")
 model_pheno.eval()
 
 FEATURE_DICT = {