Use snapshot_download+allow_patterns without subfolder arg

app.py

@@ -4,24 +4,22 @@ from huggingface_hub import snapshot_download
 from transformers import AutoTokenizer, AutoModelForCausalLM
 import gradio as gr
 
-# ——— CONFIG ———  
-REPO_ID   = "CodCodingCode/llama-3.1-8b-clinical"  
-SUBFOLDER = "checkpoint-45000"  
-HF_TOKEN  = os.environ["HUGGINGFACE_HUB_TOKEN"]  # set in Settings→Secrets  
-
-# ——— SNAPSHOT & LOAD ———  
-# This will grab all .json and .safetensors under checkpoint-45000:
-local_dir = snapshot_download(
+# ——— CONFIG ———
+REPO_ID = "CodCodingCode/llama-3.1-8b-clinical"
+SUBFOLDER = "checkpoint-45000"
+HF_TOKEN = os.getenv("HUGGINGFACE_HUB_TOKEN")
+
+# ——— DOWNLOAD ONLY THE CHECKPOINT FILES ———
+local_cache = snapshot_download(
     repo_id=REPO_ID,
-    subfolder=SUBFOLDER,
     token=HF_TOKEN,
-    allow_patterns=["*.json", "*.safetensors"],
+    allow_patterns=[f"{SUBFOLDER}/*.json", f"{SUBFOLDER}/*.safetensors"],
 )
 
-# Now point at that folder:
-MODEL_DIR = local_dir  # e.g. ~/.cache/huggingface/…/checkpoint-45000
+# ——— POINT AT THE REAL FILES ———
+MODEL_DIR = os.path.join(local_cache, SUBFOLDER)
 
-# Load tokenizer & model from the real files you just pulled:
+# ——— LOAD MODEL & TOKENIZER ———
 tokenizer = AutoTokenizer.from_pretrained(
     MODEL_DIR,
     use_fast=False,
@@ -36,7 +34,6 @@ model = AutoModelForCausalLM.from_pretrained(
 model.eval()
 
 
-
 # === Role Agent with instruction/input/output format ===
 class RoleAgent:
     def __init__(self, role_instruction):

requirements.txt

@@ -1,4 +1,4 @@
-huggingface_hub==0.25.2
+huggingface-hub==0.31.4
 transformers>=4.38.0
 torch>=2.1.0
 peft>=0.9.0