docs(operate): relayminer exposing deepseek service + datanode deployment on Kubernetes (using ollama)

docusaurus/docs/1_operate/3_configs/6_example_configs.md

@@ -54,3 +54,157 @@ config:
         websocket:
           backend_url: http://<YOUR_INTERNALIP>:48546
 ```
+
+### RelayMiner exposing a `deepseek` service
+
+![ollama-deepseek](../../../static/img/ollama-deepseek-relayminer-pocketd.png)
+
+#### What?
+
+This section explains how to deploy a `deepseek` datanode and expose a `deepseek` endpoint on the Pocket Network using the `deepSeek-r1:1.5b` model, a distilled (smaller) variant in the DeepSeek R1 family with 1.5 billion parameters. Once you complete this short tutorial, your large language model instance will be able to respond to relay requests sent from an Application and/or a Gateway through the Pocket Network.
+
+#### How?
+
+##### 1. Datanode
+
+A datanode represents the process entity responsible for performing inference computations. In this example, we will use Ollama to run the DeepSeek R1 large language model (LLM) on Kubernetes, and then expose a flexible [API](https://docs.ollama.com/api) that allows the RelayMiner to forward relay requests and return responses across the network.
+
+First, create a Kubernetes manifest YAML file named `deepseek.yaml` with this content:
+
+```yaml
+---
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: datanodes
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: ollama-deepseek
+  namespace: datanodes
+spec:
+  replicas: 1
+  selector:
+    matchLabels:
+      app: ollama-deepseek
+  template:
+    spec:  
+      containers:
+      - image: ollama/ollama:0.12.3
+        name: ollama
+        env:
+        - name: OLLAMA_HOST
+          value: 0.0.0.0
+        ports:
+        - containerPort: 11434
+        resources:
+          requests:
+            cpu: 1000m
+            memory: 2Gi
+          limits: 
+            cpu: 2000m
+            memory: 3Gi
+      - image: curlimages/curl
+        name: load-model
+        command: ["/bin/sh", "-c"]
+        args:
+          - "sleep infinity"
+        lifecycle:
+          postStart:
+            exec:
+              command:
+                - "/bin/sh"
+                - "-c"
+                - "curl -X POST http://localhost:11434/api/pull -H 'Content-Type: application/json' -d '{\"name\": \"deepseek-r1:1.5b\"}'"
+        resources:
+          limits:
+            memory: 50Mi
+            cpu: 25m
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: ollama-deepseek
+  namespace: datanodes
+spec:
+  selector:
+    app: ollama-deepseek
+  ports:
+    - protocol: TCP
+      name: http
+      port: 11434
+      targetPort: 11434
+```
+
+This Kubernetes manifest will deploy three resources into your Kubernetes cluster:
+1. **Namespace**: This will create a dedicated namespace for the datanode.
+2. **Deployment**: This deploys an Application workload to run the Ollama docker image.
+3. **Service**: This defines a stable (long-lived) IP address (and hostname) and port ([Ollama API](https://docs.ollama.com/api) reachable at `ollama-deepseek.datanodes:11434` within your Kubernetes cluster).
+
+##### 2. Service
+
+There should already be a `deepseek` service staked on-chain on `mainnet`. You can make sure of that by running this command below:
+```shell
+?> pocketd query service show-service deepseek --node=https://shannon-grove-rpc.mainnet.poktroll.com
+service:
+  compute_units_per_relay: "5000"
+  id: deepseek
+  name: Deepseek V3 0324
+  owner_address: pokt1lf0kekv9zcv9v3wy4v6jx2wh7v4665s8e0sl9s
+```
+
+If the service is not declared, please follow the [Service Creation](../1_cheat_sheets/1_service_cheatsheet.md) documentation to declare a `deepseek` service.
+
+##### 3. Supplier
+
+For your RelayMiner to forward relays to your datanode (llm), you must declare one on-chain Supplier entity that reference your Relayminer public RPC endpoint for the `deepseek` service.
+
+For more information, you can follow this documentation to:
+1. [Create a supplier account](http://localhost:3000/operate/cheat_sheets/supplier_cheatsheet#account-setup)
+2. [Stake a supplier](http://localhost:3000/operate/cheat_sheets/supplier_cheatsheet#supplier-configuration)
+
+Here is a example of the `pocketd tx supplier stake-supplier`
+```yaml
+cat <<🚀 > /tmp/stake_supplier_config.yaml
+owner_address: <SUPPLIER_ADDR>
+operator_address: <SUPPLIER_ADDR>
+stake_amount: 1000069upokt
+default_rev_share_percent:
+  $SUPPLIER_ADDR: 100
+services:
+  - service_id: "deepseek" # change if not using Anvil
+    endpoints:
+      - publicly_exposed_url: http://<EXTERNAL_IP>:8545 # must be public
+        rpc_type: JSON_RPC
+🚀
+```
+
+##### 4. Relayminer
+
+Once you got your Datanodes running, and the on-chain Service and Supplier entity declared, you got everything to start running your Relayminer with the appropriate configuration.
+This configuration below will start the RelayMiner in the appropriate state to.
+
+```yaml
+config:
+  default_signing_key_names: [mainnet-prod-relayminer]
+  smt_store_path: ":memory:" # /home/pocket/.pocket
+  pocket_node:
+    query_node_rpc_url: http://<YOUR_POKT_NODE>.<YOUR_TLD>:443
+    query_node_grpc_url: tcp://<YOUR_POKT_NODE>.<YOUR_TLD>:443
+    tx_node_rpc_url: http://<YOUR_POKT_NODE>.<YOUR_TLD>:443
+  suppliers:
+    - service_id: deepseek
+      listen_url: http://0.0.0.0:8545
+      service_config:
+        backend_url: http://ollama-deepseek.datanodes:11434/api/generate
+```
+
+> Note: The `/api/generate` is mandatory in this example because, this is the Ollama API endpoint that will generate a response for a given prompt with a provided model. 
+
+To make sure your RelayMiner instance is able to process relays correct, you can use this command that simulate a relay:
+```
+ pocketd relayminer relay --app=<app_address> --supplier=<supplier_address> --node=tcp://127.0.0.1:26657 --grpc-addr=localhost:9090 --grpc-insecure=true --payload="{\"model\": \"deepseek-r1:1.5b\", \"prompt\": \"How are you today?\", \"stream\": false}"
+```
+
+> Note: Be careful, you must make sure the `stream` field is equal to `false` as your are sending an RPC relay requests and not streaming data.