Add .env, update readme

ai/llama_cpp/.env.example

@@ -0,0 +1,10 @@
+# Host directory path where GGUF models are stored
+MODELS_PATH=/pwspool/software/ai/llama_cpp
+
+# Filename of the default model to load
+MODEL_FILE=Llama-3.1-8B-Instruct-Q4_K_M.gguf
+
+# Performance and hardware tuning
+GPU_LAYERS=999
+CTX_SIZE=4096
+THREADS=6

ai/llama_cpp/.gitignore

@@ -0,0 +1 @@
+.env

ai/llama_cpp/README.md

@@ -3,3 +3,16 @@ A general purpose llm server that projects can connect to.
 
 ## Network
 Create the `ai` docker network, applications can access the OpenAI spec server through `http://llm-server:8080/v1`.
+
+## Setup
+1. Start service
+
+```bash
+docker compose up
+```
+
+2. Download model(s)
+```bash
+cd /pwspool/software/ai/llama_cpp
+sudo wget https://huggingface.co/bartowski/Meta-Llama-3.1-8B-Instruct-GGUF/resolve/main/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf -O Llama-3.1-8B-Instruct-Q4_K_M.gguf
+```

ai/llama_cpp/docker-compose.yml

@@ -1,5 +1,6 @@
 networks:
   ai:
+    name: ai
     driver: bridge
 
 services:
@@ -10,14 +11,14 @@ services:
     networks:
       - ai
     volumes:
-      - /pwspool/software/ai/llama_cpp:/models
+      - ${MODELS_PATH:-/pwspool/software/ai/llama_cpp}:/models
-    command: >
+    environment:
-      -m /models/Llama-3.1-8B-Instruct-Q4_K_M.gguf
+      - LLAMA_ARG_MODEL=/models/${MODEL_FILE:-Llama-3.1-8B-Instruct-Q4_K_M.gguf}
-      --host 0.0.0.0
+      - LLAMA_ARG_HOST=0.0.0.0
-      --port 8080
+      - LLAMA_ARG_PORT=8080
-      -ngl 999
+      - LLAMA_ARG_N_GPU_LAYERS=${GPU_LAYERS:-999}
-      -c 4096
+      - LLAMA_ARG_CTX_SIZE=${CTX_SIZE:-4096}
-      --threads 6
+      - LLAMA_ARG_THREADS=${THREADS:-6}
     deploy:
       resources:
         reservations: