examples/llm: Reintroduce dataDir and models, with explanation in README.md

example/llm/README.md

@@ -0,0 +1,15 @@
+# Running local LLM using ollama and open-webui
+
+While `services-flake` is generally used for running services in a *development* project, typically under a source code checkout, you can also write flakes to derive an end-user app which runs a group of services, which then can be run using `nix run` (or installed using `nix profile install`):
+
+```sh
+# You can also use `nix profile install` on this URL, and run `services-flake-llm`
+nix run github:juspay/services-flake?dir=example/llm
+```
+
+## Default configuration & models
+
+`example/llm` runs two processes ollama and open-webui
+
+- The ollama data is stored under `$HOME/.services-flake/ollama`. You can change this path in `flake.nix` by setting the `dataDir` option.
+- A single model (`llama2-uncensored`) is automatically downloaded. You can modify this in `flake.nix` as well by setting the `models` option. You can also download models in the open-webui UI.

example/llm/flake.nix

@@ -13,23 +13,34 @@
       inputs.process-compose-flake.flakeModule
     ];
     perSystem = { self', pkgs, lib, ... }: {
-      process-compose."default" = pc: {
+      packages.default = self'.packages.services-flake-llm;
+
+      process-compose."services-flake-llm" = pc: {
         imports = [
           inputs.services-flake.processComposeModules.default
         ];
-        services = {
+        services = let dataDirBase = "$HOME/.services-flake/llm"; in {
           # Backend service to perform inference on LLM models
           ollama."ollama1" = {
             enable = true;
+
             # The models are usually huge, downloading them in every project
             # directory can lead to a lot of duplication. Change here to a
             # directory where the Ollama models can be stored and shared across
             # projects.
-            # dataDir = "$HOME/.services-flake/ollama1";
+            dataDir = "${dataDirBase}/ollama1";
+
+            # Define the models to download when our app starts
+            # 
+            # You can also initialize this to empty list, and download the
+            # models manually in the UI.
+            models = [ "llama2-uncensored" ];
           };
+
           # Get ChatGPT like UI, but open-source, with Open WebUI
           open-webui."open-webui1" = {
             enable = true;
+            dataDir = "${dataDirBase}/open-webui";
             environment =
               let
                 inherit (pc.config.services.ollama.ollama1) host port;
@@ -43,7 +54,8 @@
                 RAG_EMBEDDING_ENGINE = "ollama";
                 RAG_EMBEDDING_MODEL = "mxbai-embed-large:latest";
                 RAG_EMBEDDING_MODEL_AUTO_UPDATE = "True";
-                RAG_RERANKING_MODEL_AUTO_UPDATE = "True";              };
+                RAG_RERANKING_MODEL_AUTO_UPDATE = "True";
+              };
           };
         };