Trying another fix to build

app.py

@@ -1,28 +1,35 @@
 import gradio as gr
 import os
+from huggingface_hub import hf_hub_download
 from ctransformers import AutoModelForCausalLM
 
 # Configuration
 HF_REPO = "Zedel17/llama_1b_gguf"
-GGUF_FILE = "llama_1b_finetuned-Q4_K_M.gguf"
+GGUF_FILE = "llama_1b_finetuned-Q4_K_M.gguf"  # make sure this EXACT name exists in the repo
 
+# Get HF token from Space secrets (optional if repo is public)
 hf_token = os.environ.get("HF_TOKEN")
 
-print("Loading GGUF model with ctransformers...\n")
+print("Downloading GGUF model from HuggingFace...\n")
+model_path = hf_hub_download(
+    repo_id=HF_REPO,
+    filename=GGUF_FILE,
+    token=hf_token,   # ok here – this is huggingface_hub, not ctransformers
+)
+print(f"Model downloaded to: {model_path}")
 
+print("Loading GGUF model with ctransformers...")
 llm = AutoModelForCausalLM.from_pretrained(
-    HF_REPO,               # repo on HF Hub
-    model_file=GGUF_FILE,  # exact GGUF filename
-    model_type="llama",    # Llama-family GGUF
-    gpu_layers=0,          # CPU only on free Spaces
-    hf=True,               # tell ctransformers it's an HF repo
-    hf_token=hf_token      # may be None if public
+    model_path,           # local path to the .gguf file
+    model_type="llama",   # LLaMA / LLaMA 3.x family
+    context_length=2048,
+    gpu_layers=0,         # CPU only (free HF Space)
+    threads=2,            # free Space usually has 2 vCPUs
 )
-
 print("Model loaded successfully!")
 
 def chat(message, history):
-    # Simple chat-style prompt
+    # Simple conversation formatting
     parts = []
     for user_msg, bot_msg in history:
         parts.append(f"User: {user_msg}\nAssistant: {bot_msg}")

requirements.txt

@@ -1,3 +1,4 @@
 gradio
 huggingface_hub
-ctransformers>=0.2.24
+ctransformers
+