AuthorBuddy/Services/LlamaCppBackend.cs

using System.Text.Json;
using AuthorBuddy.Web.Models;

namespace AuthorBuddy.Web.Services;

public class LlamaCppBackend : ILLMBackend
{
    private readonly HttpClient _http;
    private readonly OllamaSettings _settings;

    public string Name => "llama.cpp";
    public bool SupportsModelManagement => false;

    public LlamaCppBackend(OllamaSettings settings)
    {
        _settings = settings;
        _http = new HttpClient { Timeout = TimeSpan.FromSeconds(300) };
    }

    private string Url => _settings.LlamaCppUrl;

    public async Task<bool> TestConnectionAsync()
    {
        try
        {
            var response = await _http.GetAsync($"{Url}/health");
            return response.IsSuccessStatusCode;
        }
        catch { return false; }
    }

    public async Task<IEnumerable<LLM_Model>> GetModelsAsync()
    {
        try
        {
            var response = await _http.GetAsync($"{Url}/v1/models");
            if (!response.IsSuccessStatusCode)
                return new List<LLM_Model> { new LLM_Model("llama.cpp") };

            var content = await response.Content.ReadAsStringAsync();
            using var doc = JsonDocument.Parse(content);
            if (!doc.RootElement.TryGetProperty("data", out var data))
                return new List<LLM_Model> { new LLM_Model("llama.cpp") };

            var list = new List<LLM_Model>();
            foreach (var m in data.EnumerateArray())
            {
                var id = m.TryGetProperty("id", out var idProp) ? idProp.GetString() ?? "llama.cpp" : "llama.cpp";
                list.Add(new LLM_Model(id));
            }
            return list;
        }
        catch { return new List<LLM_Model> { new LLM_Model("llama.cpp") }; }
    }

    public async Task<string> ChatAsync(string model, string prompt, string? systemPrompt = null, double? temperature = null)
    {
        try
        {
            var messages = new List<object>();
            if (!string.IsNullOrEmpty(systemPrompt))
                messages.Add(new { role = "system", content = systemPrompt });
            messages.Add(new { role = "user", content = prompt });

            var request = new Dictionary<string, object?>
            {
                ["model"] = model,
                ["messages"] = messages,
                ["stream"] = false
            };
            if (temperature.HasValue) request["temperature"] = Math.Clamp(temperature.Value, 0.0, 2.0);

            var response = await _http.PostAsync(
                $"{Url}/v1/chat/completions",
                new StringContent(JsonSerializer.Serialize(request), System.Text.Encoding.UTF8, "application/json")
            );
            if (!response.IsSuccessStatusCode) return "Fehler bei der Verarbeitung.";

            var content = await response.Content.ReadAsStringAsync();
            using var doc = JsonDocument.Parse(content);
            var root = doc.RootElement;
            if (root.TryGetProperty("choices", out var choices) && choices.GetArrayLength() > 0)
            {
                var first = choices[0];
                if (first.TryGetProperty("message", out var msg) && msg.TryGetProperty("content", out var text))
                    return text.GetString() ?? "";
            }
            return "Fehler bei der Verarbeitung.";
        }
        catch (Exception ex) { return $"Error: {ex.Message}"; }
    }

    public async Task ChatStreamAsync(string model, string prompt, Func<string, Task> onToken, string? systemPrompt = null, double? temperature = null)
    {
        var messages = new List<object>();
        if (!string.IsNullOrEmpty(systemPrompt))
            messages.Add(new { role = "system", content = systemPrompt });
        messages.Add(new { role = "user", content = prompt });

        var request = new Dictionary<string, object?>
        {
            ["model"] = model,
            ["messages"] = messages,
            ["stream"] = true
        };
        if (temperature.HasValue) request["temperature"] = Math.Clamp(temperature.Value, 0.0, 2.0);

        using var response = await _http.PostAsync(
            $"{Url}/v1/chat/completions",
            new StringContent(JsonSerializer.Serialize(request), System.Text.Encoding.UTF8, "application/json")
        );
        if (!response.IsSuccessStatusCode) return;

        using var stream = await response.Content.ReadAsStreamAsync();
        using var reader = new StreamReader(stream);

        string? line;
        while ((line = await reader.ReadLineAsync()) != null)
        {
            if (string.IsNullOrWhiteSpace(line)) continue;
            if (!line.StartsWith("data: ")) continue;

            var json = line[6..];
            if (json == "[DONE]") break;

            try
            {
                using var doc = JsonDocument.Parse(json);
                var root = doc.RootElement;
                if (root.TryGetProperty("choices", out var choices) && choices.GetArrayLength() > 0)
                {
                    var delta = choices[0].TryGetProperty("delta", out var d) ? d : default;
                    if (delta.ValueKind != JsonValueKind.Undefined && delta.TryGetProperty("content", out var content))
                    {
                        var text = content.GetString();
                        if (!string.IsNullOrEmpty(text))
                            await onToken(text);
                    }
                }
            }
            catch { }
        }
    }

    public async Task<float[]> GetEmbeddingAsync(string text, string? model = null)
    {
        try
        {
            var request = new { content = text };
            var response = await _http.PostAsync(
                $"{Url}/embedding",
                new StringContent(JsonSerializer.Serialize(request), System.Text.Encoding.UTF8, "application/json")
            );
            if (!response.IsSuccessStatusCode) return Array.Empty<float>();

            var content = await response.Content.ReadAsStringAsync();
            using var doc = JsonDocument.Parse(content);
            if (!doc.RootElement.TryGetProperty("embedding", out var emb)) return Array.Empty<float>();
            var list = new List<float>();
            foreach (var v in emb.EnumerateArray()) list.Add(v.GetSingle());
            return list.ToArray();
        }
        catch { return Array.Empty<float>(); }
    }

    public Task<bool> PullModelAsync(string name, IProgress<string>? progress = null)
        => Task.FromResult(false);

    public Task<bool> DeleteModelAsync(string name)
        => Task.FromResult(false);

    public async Task<int?> GetContextSizeAsync(string model)
    {
        try
        {
            var response = await _http.GetAsync($"{Url}/slots");
            if (!response.IsSuccessStatusCode) return 2048;
            var content = await response.Content.ReadAsStringAsync();
            using var doc = JsonDocument.Parse(content);
            if (doc.RootElement.GetArrayLength() > 0)
            {
                var first = doc.RootElement[0];
                if (first.TryGetProperty("n_ctx", out var ctx))
                    return ctx.GetInt32();
            }
            return 2048;
        }
        catch { return 2048; }
    }

    public async Task<long> MeasureLatencyAsync(string model)
    {
        try
        {
            var sw = System.Diagnostics.Stopwatch.StartNew();
            var response = await _http.GetAsync($"{Url}/health");
            sw.Stop();
            return response.IsSuccessStatusCode ? sw.ElapsedMilliseconds : -1;
        }
        catch { return -1; }
    }

    public async Task<bool> ModelExistsAsync(string model)
    {
        try
        {
            var models = await GetModelsAsync();
            return models.Any();
        }
        catch { return false; }
    }

    public async Task<(bool Success, int ContextWindow)> GetModelContextWindowAsync(string model)
    {
        var ctx = await GetContextSizeAsync(model);
        return (ctx.HasValue, ctx ?? 2048);
    }

    public bool TestEmbedding(string model)
    {
        try
        {
            var response = _http.PostAsync(
                $"{Url}/embedding",
                new StringContent(JsonSerializer.Serialize(new { content = "test" }), System.Text.Encoding.UTF8, "application/json")
            ).GetAwaiter().GetResult();
            if (!response.IsSuccessStatusCode) return false;
            var content = response.Content.ReadAsStringAsync().GetAwaiter().GetResult();
            using var doc = JsonDocument.Parse(content);
            if (doc.RootElement.TryGetProperty("embedding", out var emb))
                return emb.EnumerateArray().Any();
            return false;
        }
        catch { return false; }
    }
}