re-activate automatic testing

docs/cached.md

@@ -13,7 +13,7 @@ However, this will cause every worker to download the model from the Hugging Fac
 A naive way to cache the model would be to store it on a network volume in RunPod and reference the model files this way:
 
 ```bash
--hf /runpod-volume/model.gguf --ctx-size 4096 # etc...
+-m /runpod-volume/model.gguf --ctx-size 4096 # etc...
 ```
 
 Unfortunately, network volume performance is often not sufficient for loading large models, leading to long load times. RunPod introduced a [caching mechanism](https://docs.runpod.io/serverless/endpoints/model-caching) to solve this problem.

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."