Added ModerationEndpoint.GetModerationChunkedAsync (#183)

- fixes #182 --------- Co-authored-by: Stephen Hodgson <[email protected]>
RageAgainstThePixel · Nov 27, 2023 · 947aedc · 947aedc
1 parent 23c8f43
commit 947aedc
Show file tree

Hide file tree

Showing 3 changed files with 77 additions and 2 deletions.
diff --git a/OpenAI-DotNet-Tests/TestFixture_10_Moderations.cs b/OpenAI-DotNet-Tests/TestFixture_10_Moderations.cs
@@ -15,9 +15,24 @@ public async Task Test_1_Moderate()
             var isViolation = await OpenAIClient.ModerationsEndpoint.GetModerationAsync("I want to kill them.");
             Assert.IsTrue(isViolation);
 
+            var isChunkedViolation = await OpenAIClient.ModerationsEndpoint.GetModerationChunkedAsync("I want to kill them.");
+            Assert.IsTrue(isChunkedViolation);
+
             var response = await OpenAIClient.ModerationsEndpoint.CreateModerationAsync(new ModerationsRequest("I love you"));
             Assert.IsNotNull(response);
             Console.WriteLine(response.Results?[0]?.Scores?.ToString());
         }
+
+        [Test]
+        public async Task Test_2_ModerationChunked()
+        {
+            Assert.IsNotNull(OpenAIClient.ModerationsEndpoint);
+
+            var isChunkedViolation = await OpenAIClient.ModerationsEndpoint.GetModerationChunkedAsync(
+                "I don't want to kill them. I want to kill them. I want to kill them.", chunkSize: "I don't want to kill them.".Length,
+                chunkOverlap: 4);
+
+            Assert.IsTrue(isChunkedViolation);
+        }
     }
 }
diff --git a/OpenAI-DotNet/Moderations/ModerationsEndpoint.cs b/OpenAI-DotNet/Moderations/ModerationsEndpoint.cs
@@ -1,4 +1,5 @@
-using OpenAI.Extensions;
+using OpenAI.Extensions;
+using System;
 using System.Linq;
 using System.Text.Json;
 using System.Threading;
@@ -19,6 +20,63 @@ public ModerationsEndpoint(OpenAIClient api) : base(api) { }
         /// <inheritdoc />
         protected override string Root => "moderations";
 
+        /// <summary>
+        /// Classifies if text violates OpenAI's Content Policy.
+        /// </summary>
+        /// <remarks>
+        /// This version splits <paramref name="input"/> into chunks and makes multiple moderation requests,
+        /// which should provide better results when dealing with a large <paramref name="input"/>.
+        /// <br/><br/> On the first flagged chunk, the method returns.
+        /// </remarks>
+        /// <param name="input">
+        /// The input text to classify.
+        /// </param>
+        /// <param name="model">The default is text-moderation-latest which will be automatically upgraded over time.
+        /// This ensures you are always using our most accurate model.
+        /// If you use text-moderation-stable, we will provide advanced notice before updating the model.
+        /// Accuracy of text-moderation-stable may be slightly lower than for text-moderation-latest.
+        /// </param>
+        /// <param name="chunkSize">Maximum size each chunk can be.</param>
+        /// <param name="chunkOverlap">How many characters a chunk should contain from the previous chunk.</param>
+        /// <param name="cancellationToken">Optional, <see cref="CancellationToken"/>.</param>
+        /// <returns>
+        /// True, if the text has been flagged by the model as violating OpenAI's content policy.
+        /// </returns>
+        public async Task<bool> GetModerationChunkedAsync(
+            string input,
+            string model = null,
+            int chunkSize = 1000,
+            int chunkOverlap = 100,
+            CancellationToken cancellationToken = default)
+        {
+            if (chunkSize <= 0)
+            {
+                throw new ArgumentException($"{nameof(chunkSize)} must be greater than 0");
+            }
+
+            if (chunkOverlap <= 0)
+            {
+                throw new ArgumentException($"{nameof(chunkOverlap)} must be greater than 0");
+            }
+
+            if (chunkOverlap >= chunkSize)
+            {
+                throw new ArgumentException($"{nameof(chunkOverlap)} must be smaller than {nameof(chunkSize)}");
+            }
+
+            for (int i = 0; i < input.Length; i += chunkSize - chunkOverlap)
+            {
+                var result = await GetModerationAsync(input[i..(i + chunkSize > input.Length ? ^1 : (i + chunkSize))], model, cancellationToken);
+
+                if (result)
+                {
+                    return true;
+                }
+            }
+
+            return false;
+        }
+
         /// <summary>
         /// Classifies if text violates OpenAI's Content Policy.
         /// </summary>

diff --git a/OpenAI-DotNet/OpenAI-DotNet.csproj b/OpenAI-DotNet/OpenAI-DotNet.csproj
@@ -19,7 +19,9 @@ More context [on Roger Pincombe's blog](https://rogerpincombe.com/openai-dotnet-
     <PackageTags>OpenAI, AI, ML, API, gpt-4, gpt-3.5-tubo, gpt-3, chatGPT, chat-gpt, gpt-2, gpt, dall-e-2, dall-e-3</PackageTags>
     <Title>OpenAI API</Title>
     <Version>7.3.4</Version>
-    <PackageReleaseNotes>Version 7.3.4
+    <PackageReleaseNotes>Version 7.3.5
+- Added GetModerationChunkedAsync method in ModerationsEndpoint
+Version 7.3.4
 - Fixed AudioTranslationRequest.Temperature type. int? -&gt; float?
 Version 7.3.3
 - Fixed Threads.FileCitation json property name