53 lines
2.1 KiB
C#
53 lines
2.1 KiB
C#
using Microsoft.Extensions.Options;
|
|
using Api.Services.Contracts;
|
|
using Api.Settings;
|
|
|
|
namespace Api.Services;
|
|
|
|
public sealed class CachedAiClient : IAiClient
|
|
{
|
|
private readonly RawAiClient _raw;
|
|
private readonly IRagRepository _repository;
|
|
private readonly AiSettings _settings;
|
|
|
|
public CachedAiClient(RawAiClient raw, IRagRepository repository, IOptions<AiSettings> options)
|
|
{
|
|
_raw = raw;
|
|
_repository = repository;
|
|
_settings = options.Value;
|
|
}
|
|
|
|
public async Task<float[]> CreateEmbeddingAsync(string input, CancellationToken ct)
|
|
{
|
|
var model = GetEmbeddingModel();
|
|
var textHash = HashHelper.Compute(input);
|
|
var cacheKey = HashHelper.Compute($"embedding:{_settings.Provider}:{model}:{textHash}");
|
|
var cached = await _repository.GetEmbeddingAsync(cacheKey, ct);
|
|
if (cached is not null) return cached;
|
|
|
|
var vector = await _raw.CreateEmbeddingAsync(input, ct);
|
|
await _repository.SaveEmbeddingAsync(cacheKey, model, textHash, vector, ct);
|
|
return vector;
|
|
}
|
|
|
|
public async Task<string> CreateChatCompletionAsync(string systemPrompt, string userPrompt, decimal temperature, CancellationToken ct)
|
|
{
|
|
var model = GetChatModel();
|
|
var cacheKey = HashHelper.Compute($"chat:{_settings.Provider}:{model}:{temperature:0.00}:{systemPrompt}:{userPrompt}");
|
|
var cached = await _repository.GetChatCompletionAsync(cacheKey, ct);
|
|
if (cached is not null) return cached;
|
|
|
|
var response = await _raw.CreateChatCompletionAsync(systemPrompt, userPrompt, temperature, ct);
|
|
await _repository.SaveChatCompletionAsync(cacheKey, model, temperature, response, ct);
|
|
return response;
|
|
}
|
|
|
|
private string GetEmbeddingModel() => string.Equals(_settings.Provider, "Ollama", StringComparison.OrdinalIgnoreCase)
|
|
? _settings.Ollama.EmbeddingModel
|
|
: _settings.OpenAI.EmbeddingModel;
|
|
|
|
private string GetChatModel() => string.Equals(_settings.Provider, "Ollama", StringComparison.OrdinalIgnoreCase)
|
|
? _settings.Ollama.ChatModel
|
|
: _settings.OpenAI.ChatModel;
|
|
}
|