Quick wins

Compact recipes for the tasks you are most likely to try first. Each recipe is 10-20 lines and links to a full use case with the running end-to-end example.

Prerequisites

Baseline pattern

Every recipe below assumes this outer pattern:

using Aspose.LLM;
using Aspose.LLM.Abstractions.Parameters.Presets;

var license = new Aspose.LLM.License();
license.SetLicense("Aspose.LLM.lic");

var preset = new Qwen25Preset();
using var api = AsposeLLMApi.Create(preset);

// ... recipe body goes here.

Recipes show only the body. For full runnable examples, follow the links at the end of each recipe.

1. Send your first message

Ask the model a single question and print the reply.

string reply = await api.SendMessageAsync("What is 2 + 2?");
Console.WriteLine(reply);
// Output: 2 + 2 equals 4.

SendMessageAsync creates a session if none exists and returns the complete response.

→ Full example: Simple chat.

2. Attach an image to a message

Use a vision preset and pass image bytes via the media parameter.

// Replace the preset in the baseline pattern:
var preset = new Qwen25VL3BPreset();
using var api = AsposeLLMApi.Create(preset);

byte[] imageBytes = File.ReadAllBytes("cat.jpg");

string reply = await api.SendMessageAsync(
    "Describe what you see in one sentence.",
    media: new[] { imageBytes });

Console.WriteLine(reply);
// Output: A gray tabby cat sits on a wooden floor near a window.

Supported image formats: JPEG, PNG, BMP, GIF, WebP. Per-attachment limit: 50 MB.

→ Full example: Use cases (vision-qa — in preparation).

3. Save and resume a chat session

Persist a conversation to disk and continue it later.

// Start and exchange a few messages.
string sessionId = await api.StartNewChatAsync();
await api.SendMessageToSessionAsync(sessionId, "Remember: my project is called Helios.");

// Save it.
api.SaveChatSession(sessionId, "helios-chat.json");

// ... later, in the same or another process:
string resumedId = await api.LoadChatSession("helios-chat.json");
string reply = await api.SendMessageToSessionAsync(resumedId, "What is my project called?");
Console.WriteLine(reply);
// Output: Your project is called Helios.

The saved file contains the message history and the KV cache positions. The format is internal and may change between major SDK versions.

→ Full example: Save and restore session.

4. Run on CPU only

Force all computation to the CPU by setting GpuLayers = 0 on the preset before creating the API.

// Between creating the preset and creating the API:
var preset = new Qwen25Preset();
preset.BaseModelInferenceParameters.GpuLayers = 0;

using var api = AsposeLLMApi.Create(preset);

string reply = await api.SendMessageAsync("Summarize CPU inference in one sentence.");
Console.WriteLine(reply);

On first use, BinaryManager still selects the best CPU variant available (AVX2, AVX512, or a no-AVX fallback). Expect lower throughput than GPU inference; a 7B Q4 model typically produces 5-15 tokens/second on a modern CPU.

→ Related: Use cases (cpu-only-deployment — in preparation).

5. Run on a CUDA GPU

Request CUDA acceleration and offload all layers to the GPU. Make sure an NVIDIA driver and a compatible CUDA runtime are installed.

using Aspose.LLM.Abstractions.Acceleration;

var preset = new Qwen25Preset();
preset.BinaryManagerParameters.PreferredAcceleration = AccelerationType.CUDA;
preset.BaseModelInferenceParameters.GpuLayers = 999; // offload all layers

using var api = AsposeLLMApi.Create(preset);

string reply = await api.SendMessageAsync("Summarize GPU inference in one sentence.");
Console.WriteLine(reply);

GpuLayers = 999 is an idiomatic way to request full offload. The engine caps the value at the model’s actual layer count.

On first use, BinaryManager downloads the CUDA variant of the native binaries — larger than the CPU variant (typically 400-800 MB). Subsequent runs use the local cache.

→ Related: Use cases (gpu-deployment-cuda — in preparation).

What’s next

  • Hello, world! — the minimum-possible example, explained step by step.
  • Use cases — full running examples for every scenario.
  • Supported presets — pick a preset that fits your model and hardware.
  • Architecture — what happens behind the scenes on first Create.
How to run examples