Update app.py

app.py

@@ -3,8 +3,8 @@ import shutil
 import json
 
 # Set custom cache directories to avoid permission issues
-os.environ["TRANSFORMERS_CACHE"] = "/tmp/transformers_cache"
-os.makedirs("/tmp/transformers_cache", exist_ok=True)
+os.environ["HF_HOME"] = "/tmp/huggingface"
+os.makedirs("/tmp/huggingface", exist_ok=True)
 
 os.environ["XDG_CACHE_HOME"] = "/tmp/.cache"
 os.makedirs("/tmp/.cache", exist_ok=True)
@@ -24,7 +24,7 @@ HF_TOKEN = os.getenv("HF_TOKEN")  # Load from Render environment variable
 
 # Optional: Cleanup if corrupted config is detected
 def check_and_cleanup_corrupt_cache(repo_name: str):
-    cache_dir = os.environ["TRANSFORMERS_CACHE"]
+    cache_dir = os.environ["HF_HOME"]
     model_dir = os.path.join(cache_dir, f"models--{repo_name.replace('/', '--')}")
     if os.path.exists(model_dir):
         for root, dirs, files in os.walk(model_dir):
@@ -57,12 +57,12 @@ app.add_middleware(
 # BASE_MODEL = "microsoft/Phi-3-mini-4k-instruct"
 # HF_TOKEN = os.getenv("HF_TOKEN")  # Load from Render environment variable
 
-config = AutoConfig.from_pretrained(REPO_NAME, use_auth_token=HF_TOKEN)
+config = AutoConfig.from_pretrained(REPO_NAME, token=HF_TOKEN)
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
 #tokenizer = LlamaTokenizerFast.from_pretrained(REPO_NAME, token=HF_TOKEN)
-#tokenizer = LlamaTokenizer.from_pretrained(REPO_NAME, use_auth_token=HF_TOKEN)
+#tokenizer = LlamaTokenizer.from_pretrained(REPO_NAME, token=HF_TOKEN)
 
 #tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct")
 # tokenizer = AutoTokenizer.from_pretrained(
@@ -76,7 +76,7 @@ tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
     REPO_NAME,
     config = config,
-    use_auth_token=HF_TOKEN,
+    token=HF_TOKEN,
     torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
     device_map="auto"
 )