Merge pull request #1283 from sicoyle/feat-convo-api-new-fields

feat: use ollama to show new conversation api fields

Author: alicejgibbons

.github/workflows/validate_csharp_quickstarts.yaml

@@ -84,6 +84,15 @@ jobs:
           export GITHUB_TOKEN=${{ secrets.GITHUB_TOKEN }}
           dapr init --runtime-version=${{ env.DAPR_RUNTIME_VERSION }}
           dapr --version 
+      - name: Cache Ollama models
+        uses: actions/cache@v4
+        with:
+          path: ~/.ollama
+          key: ${{ runner.os }}-ollama-llama3.2
+      - name: Setup Ollama
+        uses: ai-action/setup-ollama@v2
+      - name: Pull Ollama model
+        run: ollama pull llama3.2:latest
       - name: Install utilities dependencies
         run: |
           echo "PATH=$PATH:$HOME/.local/bin" >> $GITHUB_ENV

.github/workflows/validate_go_quickstarts.yaml

@@ -82,6 +82,15 @@ jobs:
           export GITHUB_TOKEN=${{ secrets.GITHUB_TOKEN }}
           dapr init --runtime-version=${{ env.DAPR_RUNTIME_VERSION }}
           dapr --version
+      - name: Cache Ollama models
+        uses: actions/cache@v4
+        with:
+          path: ~/.ollama
+          key: ${{ runner.os }}-ollama-llama3.2
+      - name: Setup Ollama
+        uses: ai-action/setup-ollama@v2
+      - name: Pull Ollama model
+        run: ollama pull llama3.2:latest
       - name: Install utilities dependencies
         run: |
           echo "PATH=$PATH:$HOME/.local/bin" >> $GITHUB_ENV

.github/workflows/validate_java_quickstarts.yaml

@@ -83,6 +83,15 @@ jobs:
           export GITHUB_TOKEN=${{ secrets.GITHUB_TOKEN }}
           dapr init --runtime-version=${{ env.DAPR_RUNTIME_VERSION }}
           dapr --version
+      - name: Cache Ollama models
+        uses: actions/cache@v4
+        with:
+          path: ~/.ollama
+          key: ${{ runner.os }}-ollama-llama3.2
+      - name: Setup Ollama
+        uses: ai-action/setup-ollama@v2
+      - name: Pull Ollama model
+        run: ollama pull llama3.2:latest
       - name: Install utilities dependencies
         run: |
           echo "PATH=$PATH:$HOME/.local/bin" >> $GITHUB_ENV

.github/workflows/validate_javascript_quickstarts.yaml

@@ -78,6 +78,15 @@ jobs:
           export GITHUB_TOKEN=${{ secrets.GITHUB_TOKEN }}
           dapr init --runtime-version=${{ env.DAPR_RUNTIME_VERSION }}
           dapr --version
+      - name: Cache Ollama models
+        uses: actions/cache@v4
+        with:
+          path: ~/.ollama
+          key: ${{ runner.os }}-ollama-llama3.2
+      - name: Setup Ollama
+        uses: ai-action/setup-ollama@v2
+      - name: Pull Ollama model
+        run: ollama pull llama3.2:latest
       - name: Install utilities dependencies
         run: |
           echo "PATH=$PATH:$HOME/.local/bin" >> $GITHUB_ENV

.github/workflows/validate_python_quickstarts.yaml

@@ -86,6 +86,15 @@ jobs:
           export GITHUB_TOKEN=${{ secrets.GITHUB_TOKEN }}
           dapr init --runtime-version=${{ env.DAPR_RUNTIME_VERSION }}
           dapr --version
+      - name: Cache Ollama models
+        uses: actions/cache@v4
+        with:
+          path: ~/.ollama
+          key: ${{ runner.os }}-ollama-llama3.2
+      - name: Setup Ollama
+        uses: ai-action/setup-ollama@v2
+      - name: Pull Ollama model
+        run: ollama pull llama3.2:latest
       - name: Install utilities dependencies
         run: |
           echo "PATH=$PATH:$HOME/.local/bin" >> $GITHUB_ENV

conversation/components/ollama.yaml

@@ -0,0 +1,10 @@
+apiVersion: dapr.io/v1alpha1
+kind: Component
+metadata:
+  name: ollama
+spec:
+  type: conversation.ollama
+  version: v1
+  metadata:
+    - name: model
+      value: llama3.2:latest

conversation/csharp/http/README.md

@@ -14,27 +14,24 @@ This quickstart includes one app:
 
 This section shows how to run the application using the [multi-app run template file](https://docs.dapr.io/developing-applications/local-development/multi-app-dapr-run/multi-app-overview/) and Dapr CLI with `dapr run -f .`.  
 
-This example uses the default LLM component (Echo) which simply returns the input for testing purposes. You can switch to the OpenAI component by adding your API token in the provided OpenAI component file and changing the component name from `echo` to `openai`. For other available integrations, see the other [supported conversation components](https://docs.dapr.io/reference/components-reference/supported-conversation/).
+This example uses the Ollama LLM component for local inference. You can switch to the OpenAI component by adding your API token in the provided OpenAI component file and changing the component name from `ollama` to `openai`. For other available integrations, see the other [supported conversation components](https://docs.dapr.io/reference/components-reference/supported-conversation/).
 
 Open a new terminal window and run the multi app run template:
 
 <!-- STEP
 name: Run multi app run template
 expected_stdout_lines:
   - '== APP - conversation == Conversation input sent: What is dapr?'
-  - '== APP - conversation == Output response: What is dapr?'
+  - '== APP - conversation == Usage:'
+  - '== APP - conversation == Output response:'
   - '== APP - conversation == Tool calling input sent: What is the weather like in San Francisco in celsius?'
-  - '== APP - conversation == Output message: What is the weather like in San Francisco in celsius?'
-  - '== APP - conversation == Tool calls detected:'
-  - "== APP - conversation == Tool call: {\"id\":0,\"function\":{\"name\":\"get_weather\",\"arguments\":"
-  - '== APP - conversation == Function name: get_weather'
-  - '== APP - conversation == Function arguments: '
+  - '== APP - conversation == Tool calls'
 expected_stderr_lines:
 output_match_mode: substring
 match_order: none
-background: true
-sleep: 15
-timeout_seconds: 30
+background: false
+sleep: 30
+timeout_seconds: 120
 -->
 
 ```bash
@@ -43,25 +40,19 @@ dapr run -f .
 
 The terminal console output should look similar to this, where:
 
-- The app first sends an input `What is dapr?` to the `echo` Component mock LLM.
-- The mock LLM echoes `What is dapr?`.
+- The app first sends an input `What is dapr?` to the Ollama LLM component with a structured JSON response format.
+- The LLM returns a JSON object with an `answer` field describing Dapr.
+- The app then sends a weather request with a `get_weather` tool available; the LLM calls the tool.
 
 ```text
 == APP - conversation == Conversation input sent: What is dapr?
-== APP - conversation == Output response: What is dapr?
-```
-
-- The app then sends an input `What is the weather like in San Francisco in celsius?` to the `echo` Component mock LLM.
-- The mock LLM echoes `What is the weather like in San Francisco in celsius?` and calls the `get_weather` tool.
-- The echo Component calls the `get_weather` tool and returns the requested weather information.
-
-```text
+== APP - conversation == Usage: prompt_tokens=30 completion_tokens=64 total_tokens=94
+== APP - conversation == Output response: { "answer": "Dapr is an open-source, cross-platform microservices framework..." }
 == APP - conversation == Tool calling input sent: What is the weather like in San Francisco in celsius?
-== APP - conversation == Output message: What is the weather like in San Francisco in celsius?
 == APP - conversation == Tool calls detected:
-== APP - conversation == Tool call: {"id":0,"function":{"name":"get_weather","arguments":"location,unit"}}
+== APP - conversation == Tool call: {"id":0,"function":{"name":"get_weather","arguments":"{\"location\":\"San Francisco, CA\",\"unit\":\"celsius\"}"}}
 == APP - conversation == Function name: get_weather
-== APP - conversation == Function arguments: location,unit
+== APP - conversation == Function arguments: {"location":"San Francisco, CA","unit":"celsius"}
 ```
 
 <!-- END_STEP -->
@@ -95,19 +86,17 @@ dapr run --app-id conversation --resources-path ../../../components/ -- dotnet r
 
 The terminal console output should look similar to this, where:
 
-- The app sends an input `What is dapr?` to the `echo` Component mock LLM.
-- The mock LLM echoes `What is dapr?`.
-- The app then sends an input `What is the weather like in San Francisco in celsius?` to the `echo` Component mock LLM.
-- The mock LLM echoes `What is the weather like in San Francisco in celsius?` and calls the `get_weather` tool.
-- The echo Component calls the `get_weather` tool and returns the requested weather information.
+- The app sends an input `What is dapr?` to the Ollama LLM component with a structured JSON response format.
+- The LLM returns a JSON object with an `answer` field describing Dapr.
+- The app then sends a weather request with a `get_weather` tool available; the LLM calls the tool.
 
 ```text
 == APP - conversation == Conversation input sent: What is dapr?
-== APP - conversation == Output response: What is dapr?
+== APP - conversation == Usage: prompt_tokens=30 completion_tokens=64 total_tokens=94
+== APP - conversation == Output response: { "answer": "Dapr is an open-source, cross-platform microservices framework..." }
 == APP - conversation == Tool calling input sent: What is the weather like in San Francisco in celsius?
-== APP - conversation == Output message: What is the weather like in San Francisco in celsius?
 == APP - conversation == Tool calls detected:
-== APP - conversation == Tool call: {"id":0,"function":{"name":"get_weather","arguments":"location,unit"}}
+== APP - conversation == Tool call: {"id":0,"function":{"name":"get_weather","arguments":"{\"location\":\"San Francisco, CA\",\"unit\":\"celsius\"}"}}
 == APP - conversation == Function name: get_weather
-== APP - conversation == Function arguments: location,unit
+== APP - conversation == Function arguments: {"location":"San Francisco, CA","unit":"celsius"}
 ```

conversation/csharp/http/conversation/Program.cs

@@ -1,4 +1,4 @@
-/*
+/*
 Copyright 2024 The Dapr Authors
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -14,9 +14,9 @@ limitations under the License.
 */
 
 using System.Net.Http.Json;
+using System.Text.Encodings.Web;
 using System.Text.Json;
 
-// const string conversationComponentName = "echo";
 const string conversationText = "What is dapr?";
 const string toolCallInput = "What is the weather like in San Francisco in celsius?";
 
@@ -40,25 +40,47 @@ limitations under the License.
       }]
     }],
     "parameters": {},
-    "metadata": {}
+    "metadata": {},
+    "response_format": {
+      "type": "object",
+      "properties": {"answer": {"type": "string"}},
+      "required": ["answer"]
+    },
+    "prompt_cache_retention": "86400s"
   }
 """);
 
-var conversationResponse = await httpClient.PostAsJsonAsync("http://localhost:3500/v1.0-alpha2/conversation/echo/converse", conversationRequestBody);
-var conversationResult = await conversationResponse.Content.ReadFromJsonAsync<JsonElement>();
+var conversationResponse = await httpClient.PostAsJsonAsync("http://localhost:3500/v1.0-alpha2/conversation/ollama/converse", conversationRequestBody);
+conversationResponse.EnsureSuccessStatusCode();
+var responseText = await conversationResponse.Content.ReadAsStringAsync();
+var conversationResult = JsonSerializer.Deserialize<JsonElement>(responseText);
 
-var conversationContent = conversationResult
-  .GetProperty("outputs")
-  .EnumerateArray()
-  .First()
+if (conversationResult.ValueKind == JsonValueKind.Null || conversationResult.ValueKind == JsonValueKind.Undefined)
+{
+    throw new InvalidOperationException($"Failed to parse response as JSON. Response: {responseText}");
+}
+
+if (!conversationResult.TryGetProperty("outputs", out var outputsElement))
+{
+    throw new InvalidOperationException($"Response does not contain 'outputs' property. Response: {responseText}");
+}
+
+var firstOutput = outputsElement.EnumerateArray().First();
+
+Console.WriteLine($"Conversation input sent: {conversationText}");
+if (firstOutput.TryGetProperty("model", out var modelElement) && modelElement.GetString() is { Length: > 0 } model)
+    Console.WriteLine($"Model: {model}");
+if (firstOutput.TryGetProperty("usage", out var usageElement))
+    Console.WriteLine($"Usage: prompt_tokens={usageElement.GetProperty("promptTokens").GetString()} completion_tokens={usageElement.GetProperty("completionTokens").GetString()} total_tokens={usageElement.GetProperty("totalTokens").GetString()}");
+
+var conversationContent = firstOutput
   .GetProperty("choices")
   .EnumerateArray()
   .First()
   .GetProperty("message")
   .GetProperty("content")
   .GetString();
 
-Console.WriteLine($"Conversation input sent: {conversationText}");
 Console.WriteLine($"Output response: {conversationContent}");
 
 //
@@ -82,20 +104,8 @@ limitations under the License.
         "scrubPii": false
       }
     ],
-    "parameters": {
-      "max_tokens": {
-        "@type": "type.googleapis.com/google.protobuf.Int64Value",
-        "value": "100"
-      },
-      "model": {
-        "@type": "type.googleapis.com/google.protobuf.StringValue",
-        "value": "claude-3-5-sonnet-20240620"
-      }
-    },
-    "metadata": {
-      "api_key": "test-key",
-      "version": "1.0"
-    },
+    "parameters": {},
+    "metadata": {},
     "scrubPii": false,
     "temperature": 0.7,
     "tools": [
@@ -126,48 +136,71 @@ limitations under the License.
         }
       }
     ],
-    "toolChoice": "auto"
+    "toolChoice": "required"
   }
 """);
 
-var toolCallingResponse = await httpClient.PostAsJsonAsync("http://localhost:3500/v1.0-alpha2/conversation/echo/converse", toolCallRequestBody);
-var toolCallingResult = await toolCallingResponse.Content.ReadFromJsonAsync<JsonElement>();
+var toolCallingResponse = await httpClient.PostAsJsonAsync("http://localhost:3500/v1.0-alpha2/conversation/ollama/converse", toolCallRequestBody);
+toolCallingResponse.EnsureSuccessStatusCode();
+var toolCallingResponseText = await toolCallingResponse.Content.ReadAsStringAsync();
+var toolCallingResult = JsonSerializer.Deserialize<JsonElement>(toolCallingResponseText);
 
-var messageElement = toolCallingResult
-  .GetProperty("outputs")
+if (toolCallingResult.ValueKind == JsonValueKind.Null || toolCallingResult.ValueKind == JsonValueKind.Undefined)
+{
+    throw new InvalidOperationException($"Failed to parse response as JSON. Response: {toolCallingResponseText}");
+}
+
+if (!toolCallingResult.TryGetProperty("outputs", out var toolCallingOutputsElement))
+{
+    throw new InvalidOperationException($"Response does not contain 'outputs' property. Response: {toolCallingResponseText}");
+}
+
+var messageElement = toolCallingOutputsElement
   .EnumerateArray()
   .First()
   .GetProperty("choices")
   .EnumerateArray()
   .First()
   .GetProperty("message");
 
-var toolCallingContent = messageElement.TryGetProperty("content", out var contentElement) 
-    ? contentElement.GetString() 
+var toolCallingContent = messageElement.TryGetProperty("content", out var contentElement)
+    ? contentElement.GetString()
     : null;
 
-var functionCalled = messageElement
-  .GetProperty("toolCalls")
-  .EnumerateArray()
-  .First()
-  .GetProperty("function");
-
-var functionName = functionCalled.GetProperty("name").GetString();
-var functionArguments = functionCalled.GetProperty("arguments").GetString();
+Console.WriteLine($"Tool calling input sent: {toolCallInput}");
+Console.WriteLine($"Output message: {toolCallingContent}");
 
-var toolCallJson = JsonSerializer.Serialize(new
+if (messageElement.TryGetProperty("toolCalls", out var toolCallsElement) && toolCallsElement.GetArrayLength() > 0)
 {
-  id = 0,
-  function = new
-  {
-    name = functionName,
-    arguments = functionArguments,
-  },
-});
+    var functionCalled = toolCallsElement
+        .EnumerateArray()
+        .First()
+        .GetProperty("function");
+
+    var functionName = functionCalled.GetProperty("name").GetString();
+    var functionArguments = functionCalled.GetProperty("arguments").GetString();
+
+    var toolCallJson = JsonSerializer.Serialize(new
+    {
+        id = 0,
+        function = new
+        {
+            name = functionName,
+            arguments = functionArguments,
+        },
+    }, s_jsonOptions);
+
+    Console.WriteLine("Tool calls detected:");
+    Console.WriteLine($"Tool call: {toolCallJson}");
+    Console.WriteLine($"Function name: {functionName}");
+    Console.WriteLine($"Function arguments: {functionArguments}");
+}
+else
+{
+    Console.WriteLine("Tool calls not found");
+}
 
-Console.WriteLine($"Tool calling input sent: {toolCallInput}");
-Console.WriteLine($"Output message: {toolCallingContent}");
-Console.WriteLine("Tool calls detected:");
-Console.WriteLine($"Tool call: {toolCallJson}");
-Console.WriteLine($"Function name: {functionName}");
-Console.WriteLine($"Function arguments: {functionArguments}");
+static partial class Program
+{
+    static readonly JsonSerializerOptions s_jsonOptions = new() { Encoder = JavaScriptEncoder.UnsafeRelaxedJsonEscaping };
+}