Files
embeddingsearch/src/Server/AIProvider.cs

144 lines
5.2 KiB
C#

using System.Text;
using Newtonsoft.Json;
using Newtonsoft.Json.Linq;
using Server.Exceptions;
namespace Server;
public class AIProvider
{
private readonly ILogger<AIProvider> _logger;
private readonly IConfiguration _configuration;
public AIProvidersConfiguration aIProvidersConfiguration;
public AIProvider(ILogger<AIProvider> logger, IConfiguration configuration)
{
_logger = logger;
_configuration = configuration;
AIProvidersConfiguration? retrievedAiProvidersConfiguration = _configuration
.GetSection("Embeddingsearch")
.Get<AIProvidersConfiguration>();
if (retrievedAiProvidersConfiguration is null)
{
_logger.LogCritical("Unable to build AIProvidersConfiguration. Please check your configuration.");
throw new ServerConfigurationException("Unable to build AIProvidersConfiguration. Please check your configuration.");
}
else
{
aIProvidersConfiguration = retrievedAiProvidersConfiguration;
}
}
public float[] GenerateEmbeddings(string modelUri, string[] input)
{
Uri uri = new(modelUri);
string provider = uri.Scheme;
string model = uri.AbsolutePath;
AIProviderConfiguration? aIProvider = aIProvidersConfiguration.AiProviders
.FirstOrDefault(x => String.Equals(x.Key.ToLower(), provider.ToLower()))
.Value;
if (aIProvider is null)
{
_logger.LogError("Model provider {provider} not found in configuration. Requested model: {modelUri}", [provider, modelUri]);
throw new ServerConfigurationException($"Model provider {provider} not found in configuration. Requested model: {modelUri}");
}
using var httpClient = new HttpClient();
string embeddingsJsonPath = "";
Uri baseUri = new(aIProvider.BaseURL);
Uri requestUri;
IEmbedRequestBody embedRequest;
string[][] requestHeaders = [];
switch (aIProvider.Handler)
{
case "ollama":
embeddingsJsonPath = "$.embeddings[*]";
requestUri = new Uri(baseUri, "/api/embed");
embedRequest = new OllamaEmbedRequestBody()
{
input = input,
model = model
};
break;
case "openai":
embeddingsJsonPath = "$.data[*].embedding";
requestUri = new Uri(baseUri, "/v1/embeddings");
embedRequest = new OpenAIEmbedRequestBody()
{
input = input,
model = model
};
if (aIProvider.ApiKey is not null)
{
requestHeaders = [
["Authorization", $"Bearer {aIProvider.ApiKey}"]
];
}
break;
default:
_logger.LogError("Unknown handler {aIProvider.Handler} in AiProvider {provider}.", [aIProvider.Handler, provider]);
throw new ServerConfigurationException($"Unknown handler {aIProvider.Handler} in AiProvider {provider}.");
}
var requestContent = new StringContent(
JsonConvert.SerializeObject(embedRequest),
UnicodeEncoding.UTF8,
"application/json"
);
var request = new HttpRequestMessage()
{
RequestUri = requestUri,
Method = HttpMethod.Post,
Content = requestContent
};
foreach (var header in requestHeaders)
{
request.Headers.Add(header[0], header[1]);
}
HttpResponseMessage response = httpClient.PostAsync(requestUri, requestContent).Result;
string responseContent = response.Content.ReadAsStringAsync().Result;
try
{
JObject responseContentJson = JObject.Parse(responseContent);
JToken? responseContentTokens = responseContentJson.SelectToken(embeddingsJsonPath);
if (responseContentTokens is null)
{
_logger.LogError("Unable to select tokens using JSONPath {embeddingsJsonPath} for string: {responseContent}.", [embeddingsJsonPath, responseContent]);
throw new JSONPathSelectionException(embeddingsJsonPath, responseContent);
}
return [.. responseContentTokens.Values<float>()];
}
catch (Exception ex)
{
_logger.LogError("Unable to parse the response to valid embeddings. {ex.Message}", [ex.Message]);
throw;
}
}
}
public class AIProvidersConfiguration
{
public required Dictionary<string, AIProviderConfiguration> AiProviders { get; set; }
}
public class AIProviderConfiguration
{
public required string Handler { get; set; }
public required string BaseURL { get; set; }
public string? ApiKey { get; set; }
}
public interface IEmbedRequestBody { }
public class OllamaEmbedRequestBody : IEmbedRequestBody
{
public required string model { get; set; }
public required string[] input { get; set; }
}
public class OpenAIEmbedRequestBody : IEmbedRequestBody
{
public required string model { get; set; }
public required string[] input { get; set; }
}