standard docs for templates

templates/dummy/README.md

@@ -6,6 +6,10 @@ This is an example implementation of a Kalavai template. Use it to bootstrap the
 
 None
 
+## Key template variables
+
+- `deployment_name`: Name of the deployment job
+
 ## How to use
 
 ```bash

templates/vllm/README.md

@@ -6,8 +6,22 @@ Deploy LLM models across multiple worker nodes using the vLLM library.
 
 This template makes heavy use of the [vLLM library](https://docs.vllm.ai/en/latest/index.html).
 
+## Key template variables
+
+- `num_workers`: Number of workers per deployment (for tensor parallelism, i.e. how many pieces to divide the model into)
+- `model_id`: Huggingface model id to load from [Huggingface](https://huggingface.co/models)
+- `hf_token`: Huggingface token, required to load licensed model weights
+
+
 ## How to use
 
+Get default values, edit them and deploy:
+```bash
+kalavai job defaults vllm > values.yaml
+# edit values.yaml as required
+kalavai job run vllm --values-path values.yaml
+```
+
 Find out the url endpoint of the model with:
 
 ```bash
@@ -46,4 +60,4 @@ client = OpenAI(
 completion = client.completions.create(model="facebook/opt-350m",
                                       prompt="San Francisco is a")
 print("Completion result:", completion)
-```
+```