minor updates to pics, nav, and text

f5-rahm · f5-rahm · commit 0b47277a53e9 · 2025-10-20T14:46:18.000-05:00
diff --git a/docs/class1/class1.rst b/docs/class1/class1.rst
@@ -25,9 +25,11 @@ build out and customize in your own lab and extend for your specific needs.
 Resources
 ---------
 
-- :doc:`Docker Cheatsheet <resources/docker-cheatsheet>`
-- :doc:`Ollama Cheatsheet <resources/ollama-cheatsheet>`
+- `Docker Cheatsheet <resources/docker-cheatsheet>`_
+- `Ollama Cheatsheet <resources/ollama-cheatsheet>`_
 
+Labs
+----
 
 .. toctree::
    :maxdepth: 1
diff --git a/docs/class1/module1/lab1.rst b/docs/class1/module1/lab1.rst
@@ -25,17 +25,17 @@ small models on any system with modest specifications.
 **Perform these steps from the LLM Server (Web Shell recommended for ease of use)**
 
 In your deployment, click on the **Components** tab, and under **Systems**, click **Access** on the
-LLM Server and select **WEB SHELL** as shown in the image below. This will launch the shell which
+**LLM Server** and select **WEB SHELL** as shown in the image below. This will launch the shell which
 you will use for the remainder of the labs in this module.
 
 .. image:: images/00_llmserver_webshell_interface.png
 
 Install Ollama
 --------------
 
-The benefit of using Docker is the install aspect is a bit of a misnomer. Docker is the engine that
-is going to simply run the pre-configured install of Ollama in a container. That said, our Ollama
-server has an NVIDIA T4 GPU so we need to configure docker to use it before proceeding.
+Using Docker simplifies the installation process—or rather, eliminates it entirely. Docker runs a
+pre-configured Ollama container, so there's no traditional installation required. However, since our
+Ollama server has an NVIDIA T4 GPU, we need to configure Docker to access it before proceeding.
 
 1. Configure Docker for GPU use
 
@@ -94,7 +94,9 @@ The output should resemble this:
       llmserver-labnet:
         external: true
 
-Note the key details on what image, container name, ports, persistent data volumen, and networks are associated.
+Note the key details on what image, container name, ports, persistent data volume, and networks are associated.
+The environment variables in this case are there to keep the models loaded in memory and allow concurrent
+requests. This should improve the wait times in some of the later labs.
 
 4. Run the Ollama compose service.
 
@@ -146,7 +148,7 @@ The output should resemble this:
 
 .. code-block:: console
 
-    root@ip-10-1-1-5:/root# curl http://lcoalhost:11434
+    root@ip-10-1-1-5:/root# curl http://localhost:11434
     Ollama is running
 
 Recap
@@ -157,4 +159,4 @@ You now have the following:
 - A Docker container running the Ollama server
 - A Docker volume which is used as the file repository to store the models we'll install so they survive container restarts
 
-Next we'll install a couple models.
+Next we'll install some models.
diff --git a/docs/class1/module1/lab2.rst b/docs/class1/module1/lab2.rst
@@ -105,7 +105,7 @@ The output should resemble this:
 
     root@ip-10-1-1-5:/root/ollama# docker exec ollama ollama ps
     NAME                ID              SIZE      PROCESSOR    CONTEXT    UNTIL
-    tinyllama:latest    2644915ede35    827 MB    100% CPU     4096       About a minute from now
+    tinyllama:latest    2644915ede35    827 MB    100% CPU     4096       Forever
 
 4. Let's run a quick test against the model! We'll do this three different ways, a one-shot prompt,
 an interactive shell, and with curl via the API.
@@ -176,7 +176,7 @@ We'll use curl in this lab to run a prompt against the Ollama API.
             "model": "tinyllama",
             "prompt": "Why is grass green?",
             "stream": false
-          }'
+          }' | jq .
 
 The output should resemble this (cleaned up for readability):
 
@@ -220,19 +220,19 @@ Field                      Description
 
 .. code-block:: console
 
-    docker exec ollama ollama pull codellama
-    docker exec ollama ollama pull deepseek-r1:1.5b
-    docker exec ollama ollama pull deepseek-r1:7b
-    docker exec ollama ollama pull llama3.2:3b
+    docker exec ollama ollama run codellama
+    docker exec ollama ollama run deepseek-r1:1.5b
+    docker exec ollama ollama run deepseek-r1:7b
+    docker exec ollama ollama run llama3.2:3b
 
 The output should resemble this:
 
 .. code-block:: console
 
-    root@ip-10-1-1-5:/root/ollama# docker exec ollama ollama pull codellama
-     docker exec ollama ollama pull deepseek-r1:1.5b
-     docker exec ollama ollama pull deepseek-r1:7b
-     docker exec ollama ollama pull llama3.2:3b
+    root@ip-10-1-1-5:/root/ollama# docker exec ollama ollama run codellama
+     docker exec ollama ollama run deepseek-r1:1.5b
+     docker exec ollama ollama run deepseek-r1:7b
+     docker exec ollama ollama run llama3.2:3b
     pulling manifest
     pulling 3a43f93b78ec: 100% ▕██████████████████▏ 3.8 GB
     pulling 8c17c2ebb0ea: 100% ▕██████████████████▏ 7.0 KB
diff --git a/docs/class1/module1/lab3.rst b/docs/class1/module1/lab3.rst
@@ -136,11 +136,25 @@ The output should resemble:
       -v /tmp/custom-models:/tmp/custom-models \
       alpine sh -c "cp -r /tmp/custom-models/* /root/.ollama/models/"
 
+The output should resemble the following:
+
+.. code-block:: console
+
+    root@ip-10-1-1-5:/root/ollama# docker run --rm \
+      -v ollama_model_data:/root/.ollama \
+      -v /tmp/custom-models:/tmp/custom-models \
+      alpine sh -c "cp -r /tmp/custom-models/* /root/.ollama/models/"
+    Unable to find image 'alpine:latest' locally
+    latest: Pulling from library/alpine
+    2d35ebdb57d9: Pull complete
+    Digest: sha256:4b7ce07002c69e8f3d704a9c5d6fd3053be500b7f1c69fc0d80990c2ad8dd412
+    Status: Downloaded newer image for alpine:latest
+
 4. Verify the modelfiles are now in your model_data docker volume
 
 .. code-block:: console
 
-    ls -als /var/lib/docker/volumes/model_data/_data/models
+    ls -als /var/lib/docker/volumes/ollama_model_data/_data/models
 
 The output should resemble:
 
diff --git a/docs/class1/module1/module1.rst b/docs/class1/module1/module1.rst
@@ -1,7 +1,7 @@
 Module 1: Large Language Models
 ===============================
 
-Here we'll start with the basics of large language models (LLMs) by installing and performing some preliminary work with the Ollama model management platform.
+We'll start with the basics of large language models (LLMs) by installing and performing some preliminary work with the Ollama model management platform.
 
 .. toctree::
    :maxdepth: 1
diff --git a/docs/class1/module2/images/06_openwebui_chatbot_models.png b/docs/class1/module2/images/06_openwebui_chatbot_models.png
diff --git a/docs/class1/module2/lab1.rst b/docs/class1/module2/lab1.rst
@@ -153,11 +153,12 @@ should resemble this one:
 Now take a step back and see what you've just built. You have your own working generative AI environment!
 And your prompt session history in the left-hand menu, no less. Not too shabby, right?!?
 
-You might have noticed that your initial prompt took a hot minute to get a response. This is due to the
-way Ollama is set up in docker by default. When you ran a model in Module 1 via a ``docker exec``
-command within the container, it loaded that model into memory, but only for a short while. You can see
-when I drop the model list down that there is a green dot next to tinylama, indicating that the model is
-loaded, and hovering over the green dot shows the tool tip that it will unload in 4 minutes.
+Depending on the model you chose, you might have noticed that your initial prompt took a hot minute to get
+a response. This is due to the way Ollama is set up in docker by default. When you ran a model in Module 1
+via a ``docker exec`` command within the container, it loaded that model into memory, but only for a short
+while after the first five models, which are set to stay loaded forever. You can see when I drop the model
+list down that there is a green dot next to the loaded models. Hovering over the model's green dot shows the
+tool tip that it will unload in 292 years. I think you'll make it through the lab!
 
 .. image:: images/06_openwebui_chatbot_models.png
 
diff --git a/docs/class1/module2/lab2.rst b/docs/class1/module2/lab2.rst
@@ -399,20 +399,16 @@ The output should resemble this:
     root@ip-10-1-1-4:/root/open-webui# docker ps | grep open-webui$
     2fb45a840c35   ghcr.io/open-webui/open-webui:main   "bash start.sh"          4 hours ago    Up 19 minutes (healthy)   0.0.0.0:3000->8080/tcp                        open-webui
 
-.. note::
-
-    If you want to pre-load the llama3.2:3b model into memory on the LLM server while you wait for open WebUI to get
-    healthy, you can save a little wait time in prompting step below. Login to the webshell of the LLM server and type
-    **docker exec ollama ollama run llama3.2:3b** and just leave that tab open.
-
-4. Launch your Open WebUI tab again and login. Select the llama3.2:3b model, then click the diamond pattern
-in your chat block to reveal your tools and select the f5mcp tool.
+4. Launch your Open WebUI tab again and login. Select the **llama3.2:3b** model, then click the diamond pattern
+in your chat block to reveal your tools and select the **F5 MCP Server** tool.
 
 .. image:: images/f5mcp_tools_display.png
 
 .. note::
 
-    For each model you select to work with, you'll need to reattach your tools.
+    For each model you select to work with, you'll need to reattach your tools. Also, if you don't see your tools
+    yet, in the **web shell** in the **/root/open-webui** directory, do a **docker compose down** and then a
+    **docker compse up**, wait for it to be healthy by running **docker ps**, and then check the model tools again.
 
 5. Now prompt for the list of BIG-IP pools. I find on these smaller models I need to be more explicit and
 nudged as much as possible. Here's your chance to experiment on how you can get the model to a) actually use
diff --git a/docs/class1/module2/lab3.rst b/docs/class1/module2/lab3.rst
@@ -64,8 +64,6 @@ This should take a few minutes and resemble the output below.
 
 You'll notice you are now at the container shell prompt instead of at the App Server prompt.
 
-CONTINUE HERE AFTER LUNCH!!
-
 Configure Fabric
 ----------------
 
@@ -141,7 +139,7 @@ for the other two. Your output should resemble the following before being presen
 
     Specify HTTP timeout duration for Ollama requests (e.g. 30s, 5m, 1h) (leave empty for '20m' or type 'reset' to remove the value):
 
-3. Select the **Default AI Vendor and Model** Tool by number and hit **Enter**.  For now, choose llama3.2:3b.
+3. Select the **Default AI Vendor and Model** Tool by number and hit **Enter**.  For now, choose **llama3.2:3b**.
 In my instance, that is number 6 but that might be different for you. Skip the model context length. Your output should
 resemble the following before being presented again with the main screen:
 
@@ -376,6 +374,8 @@ You now have the following:
 
 - A powerful command-line AI Assistant to optimize your data ingestion and note taking experiences.
 
+In the **web shell** exit the container by typing **exit**
+
 
 This completes Module 2. Click Next to move on to Module 3.
 
diff --git a/docs/class1/module3/images/n8n_classifier_parameters.png b/docs/class1/module3/images/n8n_classifier_parameters.png
diff --git a/docs/class1/module3/images/n8n_generalist_model.png b/docs/class1/module3/images/n8n_generalist_model.png
diff --git a/docs/class1/module3/images/n8n_generalist_result.png b/docs/class1/module3/images/n8n_generalist_result.png
diff --git a/docs/class1/module3/lab1.rst b/docs/class1/module3/lab1.rst
@@ -9,11 +9,11 @@ verify we can access it via web UI. You will want to open the **WEB SHELL**.
 
 Installing n8n
 --------------
-1. Change directory into /root/n8n and start up your compose service.
+1. Change directory into /root/n8n (and review the compose file, right? RIGHT?!?) and start up your compose service.
 
 .. code-block:: console
 
-	cd /root/n8n
+    cd /root/n8n
     docker compose up -d
 
 You should see output similar to the following:
diff --git a/docs/class1/module3/lab2.rst b/docs/class1/module3/lab2.rst
@@ -22,18 +22,18 @@ You should see an output similar to this:
     and type **docker compose up -d** and then check **docker ps** again.
 
 Goals
-======
+-----
 
 By the end of this short lab, you will have created your first AI agent. It doesn't do a whole lot, except to proxy the conversation between n8n's native chat
 interface and your LLM served by Ollama, but it will clearly display the power that an agent provides and it will set you up for easy AI-powered automation.
 
 Steps
-=====
+-----
 
 #. Now, it's time to open your deployment's n8n Interface Access Method on the **App Server** and create an owner
 account for the instance:
 
-   .. image:: images/00_appserver_n8n_Interface.png
+   .. image:: images/00_appserver_n8n_interface.png
 
 .. note::
 
diff --git a/docs/class1/module3/lab3.rst b/docs/class1/module3/lab3.rst
@@ -49,16 +49,37 @@ understanding which model they should use for which tasks.
             Category: Coding
             Deescription: If the chat message indicates a need to code or asks for help with computer languages and scripting languages like iRules, JSON or node.js, assign this category.
 
-    #. Click the **Add Option** button, select the **Allow Multiple Classes To Be True** option, and then click the toggle to enable it.
     #. Click the **Add Option** button, select the **When No Clear Match** option, then select the **Output on Extra, 'Other' Branch** in the drop down list.
     #. Click the **Add Option** button, select the **System Prompt Template** option, then paste the system prompt below:
 
         .. code-block:: console
 
-            Please classify the text provided by the user into one of the following categories: {categories}, and use
-            the provided formatting instructions below: If they explicitly ask for coding help, do not fail and
-            classify the message as 'Coding'. If they explicitly ask for reasoning help, do not fail and classify
-            the message as 'Reasoning'. Otherwise, send the {{ $json.chatInput }} on to the next agent.
+            Please classify the text provided by the user into one of the following categories, and use the
+            provided formatting instructions below. Don't explain, and only output the json.
+
+            Categories:
+
+            REASONING: Tasks requiring logical analysis, problem-solving, decision-making, or explanations that
+            do not involve writing or modifying code. This includes answering questions, providing advice, analyzing
+            situations, explaining concepts, solving math problems, making comparisons, planning strategies, and
+            general knowledge queries.
+
+            Examples: "What's the best way to approach learning a new language?", "Explain the differences between
+            socialism and capitalism", "I have $5000 to invest. Should I put it in stocks or bonds?", "How can I
+            improve my time management skills?"
+
+            CODING: Tasks that involve writing, debugging, modifying, reviewing, or generating source code in any
+            programming language. This includes creating functions, scripts, or applications, fixing bugs,
+            refactoring code, implementing APIs, database queries, configuration files, regex patterns, and any
+            request where the expected output is executable or structured code.
+
+            Examples: "Write a Python function that calculates the Fibonacci sequence", "Debug this JavaScript code",
+            "Create a SQL query to find all users who registered last month", "Generate a regex pattern to validate
+            email addresses"
+
+            Classification rule: If the user asks to "explain how code/algorithm works" without requesting code
+            output, classify as REASONING. If they ask to "write", "create", "implement", or "generate" code,
+            classify as CODING.
 
 #. Next we'll tackle the classifier's settings. Click the **settings** tab, toggle **Retry on Fail** to enabled, then click **Back to canvas**
 
@@ -92,11 +113,23 @@ understanding which model they should use for which tasks.
 
     .. image:: images/n8n_coding_model.png
 
-#. Now you'll again repeat the previous steps to add the Generalist Agent. Everything should be the same except the model, which you can just drag a line over to the existing deepseek-r1:1.5b. After you add the new Generalist Agent node, your canvas should look like this:
+#. Now you'll again repeat the previous steps to add the Generalist Agent. Everything should be the same except the node names and the model, which should be **llama3.2:3b** in this case. After you add the two nodes, your canvas should look like this:
 
     .. image:: images/n8n_generalist_model.png
 
-#. Test time!! Need to populate a few tests that will work to show off the workflow...TBD!!
+Test time!!
+-----------
+
+Given the prompts below, I was able to successfully transit each of the three classification path,
+but please feel free to test your own and see how explicit you need to be to find success.
+
+* What is the capital of France?
+* If all Bloops are Razzies and all Razzies are Lazzies, are all Bloops definitely Lazzies?
+* Can you write a python function for me that will add two numbers and return the sum?
+
+You can see the workflow actions from the first geography-related prompt below.
+
+    .. image:: images/n8n_generalist_result.png
 
 Challenges
 ----------
