fixing issue with the requirements

src/find_cached.py

@@ -1,57 +1,69 @@
 """
-Finds the full LLM GGUF path from the Hugging Face cache.
+Finds the full GGUF path from the Hugging Face cache using huggingface_hub.
 """
 
-import os
 import argparse
-
+import os
-CACHE_DIR = "/runpod-volume/huggingface-cache/hub"
+from huggingface_hub import snapshot_download, scan_cache_dir
 
 
-def find_model_path(model_name, gguf_in_repo="model.gguf"):
+def find_model_path(model_name: str, gguf_in_repo: str) -> str | None:
     """
-    Find the path to a cached model.
+    Resolve the GGUF file path from the Hugging Face cache.
 
     Args:
-        model_name: The model name from Hugging Face
+        model_name: Hugging Face repo id (e.g. TheBloke/Mistral-7B-GGUF)
+        gguf_in_repo: Relative path to the GGUF file inside the repo
 
     Returns:
-        The full path to the cached model, or None if not found
+        Full filesystem path to the GGUF file, or None if not found
     """
 
-    cache_name = model_name.replace("/", "--")
+    # First try the official way: resolve snapshot path from cache
-    snapshots_dir = os.path.join(
+    try:
-        CACHE_DIR, f"models--{cache_name}", "snapshots"
+        snapshot_path = snapshot_download(
+            repo_id=model_name,
+            local_files_only=True
         )
+        candidate = os.path.join(snapshot_path, gguf_in_repo)
+        if os.path.isfile(candidate):
+            return candidate
+    except Exception:
+        pass
 
-    if os.path.exists(snapshots_dir):
+    # Fallback: scan cache metadata explicitly (no downloads)
-        snapshots = os.listdir(snapshots_dir)
+    cache = scan_cache_dir()
-
+    for repo in cache.repos:
-        if snapshots:
+        if repo.repo_id == model_name:
-            return os.path.join(snapshots_dir, snapshots[0], gguf_in_repo)
+            for revision in repo.revisions:
+                candidate = os.path.join(revision.snapshot_path, gguf_in_repo)
+                if os.path.isfile(candidate):
+                    return candidate
 
     return None
 
 
 def main():
-    """
-    Main function to find and print the model path.
-    """
-
     parser = argparse.ArgumentParser(
         description="Find the full GGUF path from the Hugging Face cache."
     )
     parser.add_argument(
-        "model", type=str, help="The model name from Hugging Face"
+        "model",
+        type=str,
+        help="Hugging Face model repo id (e.g. TheBloke/Mistral-7B-GGUF)",
     )
     parser.add_argument(
         "path",
         type=str,
-        help="The path to the GGUF file within the model repository",
+        help="Relative path to the GGUF file inside the repo",
     )
     args = parser.parse_args()
 
     model_path = find_model_path(args.model, args.path)
+
+    if model_path is None:
+        raise SystemExit("GGUF file not found in Hugging Face cache")
+
     print(model_path, end="")
 
 

src/requirements.txt

@@ -1,3 +1,5 @@
 runpod
 python-dotenv
 openai
+huggingface_hub
+hf-transfer

src/start.sh

@@ -56,7 +56,7 @@ echo "start.sh: Stopping existing llama-server instances (if any)..."
 # we have a string with all the command line arguments in the env var LLAMA_SERVER_CMD_ARGS;
 # it contains a.e. "-hf modelname --ctx-size 4096 -ngl 999".
 
-echo "start.sh: Running llama-server $LLAMA_SERVER_CMD_ARGS --port 3098"
+echo "start.sh: Running /app/llama-server $CACHED_LLAMA_ARGS $LLAMA_SERVER_CMD_ARGS --port 3098"
 
 touch llama.server.log
 
@@ -65,6 +65,8 @@ LD_LIBRARY_PATH=/app /app/llama-server $CACHED_LLAMA_ARGS $LLAMA_SERVER_CMD_ARGS
 
 LLAMA_SERVER_PID=$! # store the process ID (PID) of the background command
 
+tries_so_far=0
+
 check_server_is_running() {
     echo "start.sh: Checking if llama-server is done initializing..."
 
@@ -74,6 +76,13 @@ check_server_is_running() {
         return 1 # failure
     fi
 
+    tries_so_far=$((tries_so_far + 1))
+
+    if [ $tries_so_far -ge 120 ]; then
+        echo "start.sh: Error: llama-server did not start within 60 seconds."
+        exit 1
+    fi
+
     # check if the process is still running
     if ! kill -0 $LLAMA_SERVER_PID 2>/dev/null; then
         echo "start.sh: Error: llama-server process has exited unexpectedly."