Dual model templates (#263)

* Dual-mode templates

* Fixed generate-all so that works

* Fix ability to specify Claude model

templates/components/cohere-rag.jsonnet

@@ -0,0 +1,56 @@
+local base = import "base/base.jsonnet";
+local images = import "values/images.jsonnet";
+local url = import "values/url.jsonnet";
+local prompts = import "prompts/mixtral.jsonnet";
+
+{
+
+    with:: function(key, value)
+        self + {
+            ["cohere-rag-" + key]:: value,
+        },
+
+    "cohere-rag-temperature":: 0.0,
+
+    "text-completion-rag" +: {
+    
+        create:: function(engine)
+
+            local envSecrets = engine.envSecrets("cohere-credentials")
+                .with_env_var("COHERE_KEY", "cohere-key");
+
+            local containerRag =
+                engine.container("text-completion-rag")
+                    .with_image(images.trustgraph)
+                    .with_command([
+                        "text-completion-cohere",
+                        "-p",
+                        url.pulsar,
+                        "-t",
+                        "%0.3f" % $["cohere-rag-temperature"],
+                        "-i",
+                        "non-persistent://tg/request/text-completion-rag",
+                        "-o",
+                        "non-persistent://tg/response/text-completion-rag",
+                    ])
+                    .with_limits("0.5", "128M")
+                    .with_reservations("0.1", "128M");
+
+            local containerSetRag = engine.containers(
+                "text-completion-rag", [ containerRag ]
+            );
+
+            local serviceRag =
+                engine.internalService(containerSetRag)
+                .with_port(8000, 8000, "metrics");
+
+            engine.resources([
+                envSecrets,
+                containerSetRag,
+                serviceRag,
+            ])
+
+    },
+
+} + prompts
+