Rebuilt how AI Studio knows what a model can do (#960)

app/MindWork AI Studio/Provider/HuggingFace/HFModel.cs

@@ -5,4 +5,30 @@ namespace AIStudio.Provider.HuggingFace;
 /// </summary>
 /// <param name="Id">The ID of the model, written as "org/model".</param>
 /// <param name="Providers">The inference providers serving this model.</param>
-public readonly record struct HFModel(string Id, IList<HFModelProvider>? Providers);
+public readonly record struct HFModel(string Id, IList<HFModelProvider>? Providers)
+{
+    /// <summary>
+    /// The window this model has when it is reached the way this user set things up.
+    /// </summary>
+    /// <remarks>
+    /// A window belongs to an inference provider here, not to the model: the same weights run
+    /// behind several of them, each configured by somebody else. Where the user named one, its
+    /// number is the answer. Where they let the router choose, the smallest window among the
+    /// providers currently serving the model is -- nobody knows which one the router will take, and
+    /// a number promising more than the chosen provider delivers would walk a conversation into an
+    /// error the user could not see coming.
+    /// </remarks>
+    /// <param name="providerSlug">The inference provider the user chose, or empty when the router chooses.</param>
+    /// <returns>The window in tokens, or null where nobody stated one.</returns>
+    public int? ContextWindowTokens(string providerSlug)
+    {
+        if (this.Providers is null)
+            return null;
+
+        var serving = this.Providers.Where(provider => provider.IsLive);
+        if (!string.IsNullOrEmpty(providerSlug))
+            serving = serving.Where(provider => string.Equals(provider.Provider, providerSlug, StringComparison.OrdinalIgnoreCase));
+
+        return serving.Where(provider => provider.ContextWindowTokens is > 0).Min(provider => provider.ContextWindowTokens);
+    }
+}

app/MindWork AI Studio/Provider/HuggingFace/HFModelProvider.cs

@@ -1,3 +1,5 @@
+using System.Text.Json.Serialization;
+
 namespace AIStudio.Provider.HuggingFace;
 
 /// <summary>
@@ -5,4 +7,13 @@ namespace AIStudio.Provider.HuggingFace;
 /// </summary>
 /// <param name="Provider">The slug of the inference provider, e.g. "novita".</param>
 /// <param name="Status">Whether the provider currently serves the model. Known value: "live".</param>
-public readonly record struct HFModelProvider(string Provider, string Status);
+/// <param name="ContextWindowTokens">How much this provider reads and writes in one conversation, in tokens.</param>
+public readonly record struct HFModelProvider(string Provider, string Status, [property: JsonPropertyName("context_length")] int? ContextWindowTokens)
+{
+    private const string LIVE = "live";
+
+    /// <summary>
+    /// Whether this provider serves the model right now.
+    /// </summary>
+    public bool IsLive => string.Equals(this.Status, LIVE, StringComparison.OrdinalIgnoreCase);
+}

app/MindWork AI Studio/Provider/HuggingFace/ProviderHuggingFace.cs

@@ -2,6 +2,8 @@
 using System.Runtime.CompilerServices;
 
 using AIStudio.Chat;
+using AIStudio.Models;
+using AIStudio.Models.Live;
 using AIStudio.Provider.OpenAI;
 using AIStudio.Settings;
 using AIStudio.Tools.PluginSystem;
@@ -127,10 +129,10 @@ public sealed class ProviderHuggingFace : BaseProvider
     }
 
     /// <inheritdoc />
-    protected override string GetProviderRequestFailureUserMessage(ProviderRequestFailureReason failureReason)
+    protected override string GetProviderRequestFailureUserMessage(ProviderRequestFailureReason failureReason, ContextWindow contextWindow = default)
     {
         if (failureReason is not ProviderRequestFailureReason.MODEL_NOT_SUPPORTED_BY_PROVIDER)
-            return base.GetProviderRequestFailureUserMessage(failureReason);
+            return base.GetProviderRequestFailureUserMessage(failureReason, contextWindow);
 
         //
         // When Hugging Face chose the provider itself, naming it back to the user would help
@@ -166,7 +168,7 @@ public sealed class ProviderHuggingFace : BaseProvider
                            async (systemPrompt, apiParameters, tools) =>
                            {
                                // Build the list of messages:
-                               var messages = await chatThread.Blocks.BuildMessagesUsingNestedImageUrlAsync(this.Provider, chatModel);
+                               var messages = await chatThread.Blocks.BuildMessagesUsingNestedImageUrlAsync(this.CreateSettingsProvider(chatModel));
 
                                return new ChatCompletionAPIRequest
                                {
@@ -221,7 +223,23 @@ public sealed class ProviderHuggingFace : BaseProvider
     /// <inheritdoc />
     public override Task<ModelLoadResult> GetTextModels(string? apiKeyProvisional = null, CancellationToken token = default)
     {
-        return this.LoadModelsResponse<ModelsResponse>(SecretStoreType.LLM_PROVIDER, "models", this.SelectChatModels, apiKeyProvisional, token: token);
+        return this.LoadModelsResponse<ModelsResponse>(SecretStoreType.LLM_PROVIDER, "models", this.SelectChatModels, apiKeyProvisional, listingFactory: this.ListingsOf, token: token);
+    }
+
+    /// <summary>
+    /// What the router stated about the models it knows.
+    /// </summary>
+    /// <remarks>
+    /// Every model the router reports, not only the ones offered for chatting below: which models
+    /// are offered depends on the chosen inference provider, while a window belongs to whoever is
+    /// configured here, and both questions are asked of the same list.
+    /// </remarks>
+    /// <param name="response">The response of the model endpoint.</param>
+    /// <returns>One listing per model, which says nothing for the models nobody stated a window for.</returns>
+    private IEnumerable<ModelListing> ListingsOf(ModelsResponse response)
+    {
+        var providerSlug = this.hfProvider.EndpointsId();
+        return response.Data.Select(hfModel => ModelListing.For(hfModel.Id, hfModel.ContextWindowTokens(providerSlug)));
     }
 
     /// <summary>
@@ -237,7 +255,7 @@ public sealed class ProviderHuggingFace : BaseProvider
     /// <returns>The models to offer.</returns>
     private IEnumerable<Model> SelectChatModels(ModelsResponse response)
     {
-        var chatModels = response.Data.Where(hfModel => new Model(hfModel.Id, null).IsChatModel());
+        var chatModels = response.Data.Where(hfModel => new Model(hfModel.Id, null).IsChatModel(this.Provider));
         var providerSlug = this.hfProvider.EndpointsId();
         if (string.IsNullOrEmpty(providerSlug))
             return ToModels(chatModels);
@@ -253,8 +271,8 @@ public sealed class ProviderHuggingFace : BaseProvider
             return false;
 
         return hfModel.Providers.Any(provider =>
-            string.Equals(provider.Provider, providerSlug, StringComparison.OrdinalIgnoreCase) &&
-            string.Equals(provider.Status, "live", StringComparison.OrdinalIgnoreCase));
+            provider.IsLive &&
+            string.Equals(provider.Provider, providerSlug, StringComparison.OrdinalIgnoreCase));
     }
 
     /// <inheritdoc />