AuthorBuddy/Services/OllamaService.cs

342 lines
11 KiB
C#
Raw Normal View History

2026-06-21 07:52:03 +02:00
using System.Diagnostics;
2026-05-14 12:31:35 +02:00
using AuthorBuddy.Web.Models;
namespace AuthorBuddy.Web.Services;
2026-05-16 16:13:35 +02:00
public class OllamaService : IOllamaService
2026-05-14 12:31:35 +02:00
{
2026-05-31 19:31:27 +02:00
private readonly IEnumerable<ILLMBackend> _backends;
2026-05-17 07:27:18 +02:00
private readonly OllamaSettings _settings;
2026-05-31 19:31:27 +02:00
private readonly OperationState _opState;
2026-06-21 07:52:03 +02:00
private readonly object _startLock = new();
private bool _startAttempted;
2026-05-14 12:31:35 +02:00
2026-05-31 19:31:27 +02:00
public OllamaService(IEnumerable<ILLMBackend> backends, OllamaSettings settings, OperationState opState)
2026-05-14 12:31:35 +02:00
{
2026-05-31 19:31:27 +02:00
_backends = backends;
2026-05-17 07:27:18 +02:00
_settings = settings;
2026-05-31 19:31:27 +02:00
_opState = opState;
2026-05-14 12:31:35 +02:00
}
2026-07-04 10:32:35 +02:00
private readonly object _warmLock = new();
private bool _warmupAttempted;
2026-08-30 08:11:01 +02:00
private ILLMBackend GetBackend(string? name) =>
_backends.FirstOrDefault(b => b.Name == (name ?? _settings.Backend)) ?? _backends.First();
private async Task EnsureRunningAsync(string? backendName = null)
2026-06-21 07:52:03 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(backendName);
if (backend.Name != "Ollama")
2026-06-21 07:52:03 +02:00
return;
2026-08-30 08:11:01 +02:00
var reachable = await backend.TestConnectionAsync();
2026-06-21 07:52:03 +02:00
if (reachable)
2026-07-04 10:32:35 +02:00
{
await WarmupModelAsync();
2026-06-21 07:52:03 +02:00
return;
2026-07-04 10:32:35 +02:00
}
2026-06-21 07:52:03 +02:00
lock (_startLock)
{
if (_startAttempted)
return;
_startAttempted = true;
}
_opState.SetThought("Ollama läuft nicht – starte ollama serve...");
try
{
var psi = new ProcessStartInfo("ollama", "serve")
{
CreateNoWindow = true,
UseShellExecute = false
};
Process.Start(psi);
}
catch (Exception ex)
{
_opState.AppendThought($" Fehler: {ex.Message}");
return;
}
for (int i = 0; i < 30; i++)
{
await Task.Delay(1000);
2026-08-30 08:11:01 +02:00
reachable = await backend.TestConnectionAsync();
2026-06-21 07:52:03 +02:00
if (reachable)
{
_opState.AppendThought(" Ollama ist bereit");
2026-07-04 10:32:35 +02:00
await WarmupModelAsync();
2026-06-21 07:52:03 +02:00
return;
}
_opState.AppendThought(".");
}
_opState.AppendThought(" Zeitüberschreitung");
}
2026-07-04 10:32:35 +02:00
private async Task WarmupModelAsync()
{
lock (_warmLock)
{
if (_warmupAttempted) return;
_warmupAttempted = true;
}
2026-08-30 08:11:01 +02:00
var modelsToWarm = new[]
{
(Model: _settings.StyleModel, Backend: _settings.StyleBackend),
(Model: _settings.EmbedModel, Backend: _settings.RagBackend)
}
.Where(m => !string.IsNullOrWhiteSpace(m.Model))
.DistinctBy(m => m.Model)
2026-07-04 10:32:35 +02:00
.ToList();
if (modelsToWarm.Count == 0) return;
_opState.SetThought("Lade Modelle in den Arbeitsspeicher...");
2026-08-30 08:11:01 +02:00
foreach (var (model, backendName) in modelsToWarm)
2026-07-04 10:32:35 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(backendName);
if (backend.Name != "Ollama")
continue;
_opState.AppendThought($" {model} ({backendName})...");
2026-07-04 10:32:35 +02:00
try
{
2026-08-30 08:11:01 +02:00
var dummy = await backend.ChatAsync(model, " ", null, 0.0);
2026-07-04 10:32:35 +02:00
_opState.AppendThought(" ✓");
}
catch (Exception ex)
{
_opState.AppendThought($" ✗ ({ex.Message})");
}
}
_opState.AppendThought(" Fertig");
}
2026-05-16 16:13:35 +02:00
public void Reconfigure(string url)
2026-05-14 12:31:35 +02:00
{
2026-05-31 19:31:27 +02:00
_settings.OllamaUrl = url;
2026-05-14 12:31:35 +02:00
}
2026-05-31 19:31:27 +02:00
public async Task<(bool Success, long LatencyMs)> TestConnectionAsync()
2026-05-20 06:12:54 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(null);
2026-06-21 07:52:03 +02:00
await EnsureRunningAsync();
2026-08-30 08:11:01 +02:00
_opState.SetThought($"Teste Verbindung zu {backend.Name}...");
2026-06-21 07:52:03 +02:00
var sw = Stopwatch.StartNew();
2026-08-30 08:11:01 +02:00
var result = await backend.TestConnectionAsync();
2026-05-31 19:31:27 +02:00
sw.Stop();
_opState.AppendThought(result ? " OK" : " Fehlgeschlagen");
return (result, result ? sw.ElapsedMilliseconds : 0);
2026-05-20 06:12:54 +02:00
}
2026-08-30 08:11:01 +02:00
public bool SupportsModelManagement => GetBackend(null).SupportsModelManagement;
2026-05-31 19:31:27 +02:00
public async Task<IEnumerable<LLM_Model>> GetLocalModelsAsync()
2026-05-20 06:12:54 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(null);
2026-06-21 07:52:03 +02:00
await EnsureRunningAsync();
2026-08-30 08:11:01 +02:00
_opState.SetThought($"Lade Modelle von {backend.Name}...");
var models = await backend.GetModelsAsync();
2026-05-31 19:31:27 +02:00
_opState.AppendThought($" {models.Count()} gefunden");
return models;
2026-05-20 06:12:54 +02:00
}
2026-08-30 08:11:01 +02:00
public async Task<IEnumerable<LLM_Model>> GetModelsForBackendAsync(string backendName)
2026-05-14 12:31:35 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(backendName);
if (backend.Name == "Ollama")
await EnsureRunningAsync();
_opState.SetThought($"Lade Modelle von {backend.Name}...");
var models = await backend.GetModelsAsync();
_opState.AppendThought($" {models.Count()} gefunden");
return models;
2026-05-31 19:31:27 +02:00
}
2026-05-16 16:13:35 +02:00
2026-08-30 08:11:01 +02:00
public async Task<bool> TestEmbeddingAsync(string model, string? backendName = null)
2026-05-31 19:31:27 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(backendName);
await EnsureRunningAsync(backendName);
return backend.TestEmbedding(model);
}
public async Task<float[]> GetEmbeddingAsync(string text, string? backendName = null)
{
var backend = GetBackend(backendName);
_opState.SetBusy(true);
_opState.SetThought($"{backend.Name}: Erzeuge Embeddings...");
try
{
await EnsureRunningAsync(backendName);
return await backend.GetEmbeddingAsync(text, _settings.EmbedModel);
}
catch
{
_opState.SetThought("Fehler bei Embeddings");
throw;
}
finally
{
_opState.SetBusy(false);
}
2026-05-14 12:31:35 +02:00
}
2026-05-17 07:27:18 +02:00
public async Task<string> QuickSpellCheckAsync(string text)
{
2026-08-30 08:11:01 +02:00
_opState.SetBusy(true);
_opState.SetThought("Rechtschreibprüfung läuft...");
try
{
await EnsureRunningAsync(_settings.SpellingBackend);
return await Chat(_settings.SpellingModel, text, _settings.SpellingPrompt, _settings.SpellingTemp, _settings.SpellingBackend);
}
catch
{
_opState.SetThought("Fehler bei Rechtschreibprüfung");
throw;
}
finally
{
_opState.SetBusy(false);
}
2026-05-17 07:27:18 +02:00
}
public async Task<string> StreamStyleCheckAsync(string text)
{
2026-08-30 08:11:01 +02:00
_opState.SetBusy(true);
_opState.SetThought("Stilprüfung läuft...");
try
{
await EnsureRunningAsync(_settings.StyleBackend);
return await Chat(_settings.StyleModel, text, _settings.LektorSystemPrompt, _settings.LektorTemp, _settings.StyleBackend);
}
catch
{
_opState.SetThought("Fehler bei Stilprüfung");
throw;
}
finally
{
_opState.SetBusy(false);
}
2026-05-17 07:27:18 +02:00
}
public async Task<string> AnswerBasedOnFactsAsync(string question, string facts)
{
2026-08-30 08:11:01 +02:00
_opState.SetBusy(true);
_opState.SetThought("Faktencheck läuft...");
try
{
await EnsureRunningAsync(_settings.RagBackend);
var prompt = _settings.GetFactCheckPrompt(question, facts);
return await Chat(_settings.EmbedModel, prompt, temperature: _settings.FactCheckTemp, backendName: _settings.RagBackend);
}
catch
{
_opState.SetThought("Fehler bei Faktencheck");
throw;
}
finally
{
_opState.SetBusy(false);
}
2026-05-17 07:27:18 +02:00
}
2026-08-30 08:11:01 +02:00
public async Task<bool> ModelExistsAsync(string model, string? backendName = null)
2026-05-14 12:31:35 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(backendName);
await EnsureRunningAsync(backendName);
return await backend.ModelExistsAsync(model);
2026-05-14 12:31:35 +02:00
}
2026-08-30 08:11:01 +02:00
public async Task<(bool Success, int ContextWindow)> GetModelContextWindowAsync(string model, string? backendName = null)
2026-05-14 12:31:35 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(backendName);
await EnsureRunningAsync(backendName);
return await backend.GetModelContextWindowAsync(model);
2026-05-14 12:31:35 +02:00
}
2026-08-30 08:11:01 +02:00
public async Task<long> MeasureLatencyAsync(string model, string? backendName = null)
2026-05-14 12:31:35 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(backendName);
await EnsureRunningAsync(backendName);
return await backend.MeasureLatencyAsync(model);
2026-05-31 19:31:27 +02:00
}
2026-05-16 16:13:35 +02:00
2026-08-30 08:11:01 +02:00
public async Task<string> Chat(string model, string userMessage, string? systemPrompt = null, double? temperature = null, string? backendName = null)
2026-05-31 19:31:27 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(backendName);
_opState.SetBusy(true);
_opState.SetThought($"{backend.Name}: Sende Anfrage an {model}...");
try
{
await EnsureRunningAsync(backendName);
var result = await backend.ChatAsync(model, userMessage, systemPrompt, temperature);
_opState.AppendThought(" Fertig");
return result;
}
catch
{
_opState.SetThought($"Fehler bei {model}");
throw;
}
finally
{
_opState.SetBusy(false);
}
2026-05-14 12:31:35 +02:00
}
2026-08-30 08:11:01 +02:00
public async Task ChatStream(string model, string userMessage, Func<string, Task> onToken, string? systemPrompt = null, double? temperature = null, string? backendName = null)
2026-05-14 12:31:35 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(backendName);
_opState.SetBusy(true);
_opState.SetThought($"{backend.Name}: Verbinde mit Modell {model}...");
try
2026-05-16 16:13:35 +02:00
{
2026-08-30 08:11:01 +02:00
await EnsureRunningAsync(backendName);
int tokenCount = 0;
await backend.ChatStreamAsync(model, userMessage, async chunk =>
{
tokenCount++;
_opState.AppendThought(chunk);
await onToken(chunk);
}, systemPrompt, temperature);
2026-05-31 19:31:27 +02:00
2026-08-30 08:11:01 +02:00
_opState.AppendThought($" Fertig ({tokenCount} Token)");
}
catch
{
_opState.SetThought($"Fehler bei {model}");
throw;
}
finally
{
_opState.SetBusy(false);
}
2026-05-14 12:31:35 +02:00
}
2026-05-16 16:13:35 +02:00
public async Task<bool> PullModelAsync(string modelName, IProgress<string> progress)
2026-05-14 12:31:35 +02:00
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(null);
2026-06-21 07:52:03 +02:00
await EnsureRunningAsync();
2026-05-31 19:31:27 +02:00
_opState.SetThought($"Lade Modell {modelName}...");
2026-08-30 08:11:01 +02:00
var result = await backend.PullModelAsync(modelName, progress);
2026-05-31 19:31:27 +02:00
_opState.AppendThought(result ? " Erfolgreich" : " Fehlgeschlagen");
return result;
2026-05-14 12:31:35 +02:00
}
2026-07-04 10:32:35 +02:00
public async Task<bool> DeleteModelAsync(string modelName)
{
2026-08-30 08:11:01 +02:00
var backend = GetBackend(null);
2026-07-04 10:32:35 +02:00
await EnsureRunningAsync();
_opState.SetThought($"Lösche Modell {modelName}...");
2026-08-30 08:11:01 +02:00
var result = await backend.DeleteModelAsync(modelName);
2026-07-04 10:32:35 +02:00
_opState.AppendThought(result ? " Erfolgreich" : " Fehlgeschlagen");
return result;
}
2026-05-14 12:31:35 +02:00
}