Upload app.py

app.py

@@ -1,18 +1,26 @@
 
+import os
 import gradio as gr
+import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from peft import PeftModel
 
-BASE = "openlm-research/open_llama_3b"   # base of your LoRA
-LORA = "GilbertAkham/openlm-llama-lora-codetrans"   # your LoRA on HF
+BASE = "openlm-research/open_llama_3b"
+LORA = "GilbertAkham/openlm-llama-lora-codetrans"
+
+# ---- FIX HERE ----
+tokenizer = AutoTokenizer.from_pretrained(
+    BASE,
+    use_fast=False    # MUST be here, not on model
+)
 
-tokenizer = AutoTokenizer.from_pretrained(BASE)
 model = AutoModelForCausalLM.from_pretrained(
     BASE,
     load_in_8bit=True,
-    use_fast=False,          # fits on T4/low VRAM
     device_map="auto"
 )
+# ------------------
+
 model = PeftModel.from_pretrained(model, LORA)
 model.eval()
 
@@ -35,5 +43,5 @@ demo = gr.Interface(
     title="openlm-llama-LoRA codetrans",
 )
 
-demo.launch(share=True)   # share=True gives you a free public link
+demo.launch(share=True)