Upload src/veris_classifier/classifier.py with huggingface_hub

src/veris_classifier/classifier.py

@@ -54,6 +54,15 @@ def load_hf_model():
     from peft import PeftModel
     from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
+    # This model path expects GPU execution (ZeroGPU on Spaces). On CPU-only
+    # runtimes, transformers can fail with opaque disk offload errors.
+    if not torch.cuda.is_available():
+        raise RuntimeError(
+            "Fine-tuned model requires GPU. This Space appears to be on CPU-only "
+            "(no CUDA device available). Request ZeroGPU (A10G) or provide an "
+            "OpenAI API key to use fallback inference."
+        )
+
     logger.info(f"Loading base model: {BASE_MODEL_ID}")
     _hf_tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL_ID, trust_remote_code=True)
     if _hf_tokenizer.pad_token is None: