From 2ee757b64a9e5c73da5cc2a373bcabf60b228567 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=CE=A7=CE=B3=CF=86=CF=84=20Kompanion?=
 <chigammaphitau@chelydra.at>
Date: Sun, 19 Oct 2025 08:33:27 +0200
Subject: [PATCH] ki: implement multi-text embeddings in OllamaProvider::embed
 by batching one request per text and aggregating results

---
 src/KI/Provider/OllamaProvider.cpp | 68 +++++++++++++++---------------
 1 file changed, 34 insertions(+), 34 deletions(-)

diff --git a/src/KI/Provider/OllamaProvider.cpp b/src/KI/Provider/OllamaProvider.cpp
index 721ba0a..ce0ebfd 100644
--- a/src/KI/Provider/OllamaProvider.cpp
+++ b/src/KI/Provider/OllamaProvider.cpp
@@ -108,43 +108,43 @@ QFuture<KIReply*> OllamaProvider::chat(const KIThread& thread, const KIChatOptio
 
 QFuture<KIEmbeddingResult> OllamaProvider::embed(const QStringList& texts, const KIEmbedOptions& opts)
 {
-    QNetworkRequest req{QUrl(QStringLiteral("http://localhost:11434/api/embeddings"))};
-    req.setHeader(QNetworkRequest::ContentTypeHeader, QStringLiteral("application/json"));
+    // Execute one request per input text; aggregate outputs.
+    QFutureInterface<KIEmbeddingResult> fi;
+    fi.reportStarted();
+    if (texts.isEmpty()) { KIEmbeddingResult r; r.model = opts.model; fi.reportResult(r); fi.reportFinished(); return fi.future(); }
 
-    QJsonObject data;
-    data["model"] = opts.model;
-    data["prompt"] = texts.join("\n"); // Join all texts into a single prompt
+    struct Accum { QVector<QVector<float>> vectors; int remaining = 0; QString model; };
+    auto acc = new Accum();
+    acc->vectors.resize(texts.size());
+    acc->remaining = texts.size();
 
-    auto netReply = m_manager->post(req, QJsonDocument(data).toJson());
+    const QUrl url(QStringLiteral("http://localhost:11434/api/embeddings"));
+    for (int i = 0; i < texts.size(); ++i) {
+        QNetworkRequest req{url};
+        req.setHeader(QNetworkRequest::ContentTypeHeader, QStringLiteral("application/json"));
+        const QJsonObject body{ {QStringLiteral("model"), opts.model}, {QStringLiteral("prompt"), texts[i]} };
+        auto rep = m_manager->post(req, QJsonDocument(body).toJson());
+        connect(rep, &QNetworkReply::finished, this, [rep, i, acc, fi]() mutable {
+            if (rep->error() == QNetworkReply::NoError) {
+                const auto obj = QJsonDocument::fromJson(rep->readAll()).object();
+                if (acc->model.isEmpty()) acc->model = obj.value(QStringLiteral("model")).toString();
+                const auto arr = obj.value(QStringLiteral("embedding")).toArray();
+                QVector<float> vec; vec.reserve(arr.size());
+                for (const auto &v : arr) vec.push_back(static_cast<float>(v.toDouble()));
+                acc->vectors[i] = std::move(vec);
+            }
+            rep->deleteLater();
+            acc->remaining -= 1;
+            if (acc->remaining == 0) {
+                KIEmbeddingResult res; res.vectors = std::move(acc->vectors); res.model = acc->model;
+                fi.reportResult(res);
+                fi.reportFinished();
+                delete acc;
+            }
+        });
+    }
 
-    QFutureInterface<KIEmbeddingResult> interface;
-    interface.reportStarted();
-
-    connect(netReply, &QNetworkReply::finished, this, [netReply, interface]() mutable {
-        if (netReply->error() != QNetworkReply::NoError) {
-            // TODO: Handle error
-            interface.reportFinished();
-            netReply->deleteLater();
-            return;
-        }
-
-        const auto json = QJsonDocument::fromJson(netReply->readAll());
-        const auto embeddingArray = json["embedding"].toArray();
-
-        KIEmbeddingResult result;
-        QVector<float> embedding;
-        for (const QJsonValue &value : embeddingArray) {
-            embedding.push_back(value.toDouble());
-        }
-        result.vectors.push_back(embedding);
-        result.model = json["model"].toString();
-
-        interface.reportResult(result);
-        interface.reportFinished();
-        netReply->deleteLater();
-    });
-
-    return interface.future();
+    return fi.future();
 }
 
 void OllamaProvider::cancel(quint64 requestId)