fix: remove zerogpu-offload before prediction to avoid OOM on disk

prediction.py

@@ -22,7 +22,10 @@ from transformers import pipeline as hf_pipeline
 import litellm
 
 from tqdm import tqdm
+import subprocess
 
+# https://huggingface.co/spaces/zero-gpu-explorers/README/discussions/132
+subprocess.run("rm -rf /data-nvme/zerogpu-offload/*", env={}, shell=True)
 
 pipeline = hf_pipeline(
     "text-generation",
@@ -120,6 +123,7 @@ class ModelPrediction:
         # https://huggingface.co/docs/inference-endpoints/en/pricing?utm_source=chatgpt.com
         cost_per_second=0.001
         response = outputs[0]["generated_text"][-1]['content']
+        print(response)
         return {
             "response": response, 
             "cost": elapsed_time * cost_per_second