Adding support to Azure API

+llms/+internal/callAzureChatAPI.m

@@ -0,0 +1,142 @@
+function [text, message, response] = callAzureChatAPI(endpoint, deploymentID, messages, functions, nvp)
+%callAzureChatAPI Calls the openAI chat completions API on Azure.
+%
+%   MESSAGES and FUNCTIONS should be structs matching the json format
+%   required by the OpenAI Chat Completions API.
+%   Ref: https://platform.openai.com/docs/guides/gpt/chat-completions-api
+%
+%   Currently, the supported NVP are, including the equivalent name in the API:
+%    - ToolChoice (tool_choice)
+%    - Temperature (temperature)
+%    - TopProbabilityMass (top_p)
+%    - NumCompletions (n)
+%    - StopSequences (stop)
+%    - MaxNumTokens (max_tokens)
+%    - PresencePenalty (presence_penalty)
+%    - FrequencyPenalty (frequence_penalty)
+%    - ResponseFormat (response_format)
+%    - Seed (seed)
+%    - ApiKey
+%    - TimeOut
+%    - StreamFun
+%   More details on the parameters: https://platform.openai.com/docs/api-reference/chat/create
+%
+%   Example
+%
+%   % Create messages struct
+%   messages = {struct("role", "system",...
+%       "content", "You are a helpful assistant");
+%       struct("role", "user", ...
+%       "content", "What is the edit distance between hi and hello?")};
+%
+%   % Create functions struct
+%   functions = {struct("name", "editDistance", ...
+%       "description", "Find edit distance between two strings or documents.", ...
+%       "parameters", struct( ...
+%       "type", "object", ...
+%       "properties", struct(...
+%           "str1", struct(...
+%               "description", "Source string.", ...
+%               "type", "string"),...
+%           "str2", struct(...
+%               "description", "Target string.", ...
+%               "type", "string")),...
+%       "required", ["str1", "str2"]))};
+%
+%   % Define your API key
+%   apiKey = "your-api-key-here"
+%
+%   % Send a request
+%   [text, message] = llms.internal.callOpenAIChatAPI(messages, functions, ApiKey=apiKey)
+
+%   Copyright 2023-2024 The MathWorks, Inc.
+
+arguments
+    endpoint
+    deploymentID
+    messages
+    functions
+    nvp.ToolChoice = []
+    nvp.APIVersion = "2023-05-15"
+    nvp.Temperature = 1
+    nvp.TopProbabilityMass = 1
+    nvp.NumCompletions = 1
+    nvp.StopSequences = []
+    nvp.MaxNumTokens = inf
+    nvp.PresencePenalty = 0
+    nvp.FrequencyPenalty = 0
+    nvp.ResponseFormat = "text"
+    nvp.Seed = []
+    nvp.ApiKey = ""
+    nvp.TimeOut = 10
+    nvp.StreamFun = []
+end
+
+URL = endpoint + "openai/deployments/" + deploymentID + "/chat/completions?api-version=" + nvp.APIVersion;
+
+parameters = buildParametersCall(messages, functions, nvp);
+
+[response, streamedText] = llms.internal.sendRequest(parameters,nvp.ApiKey, URL, nvp.TimeOut, nvp.StreamFun);
+
+% If call errors, "choices" will not be part of response.Body.Data, instead
+% we get response.Body.Data.error
+if response.StatusCode=="OK"
+    % Outputs the first generation
+    if isempty(nvp.StreamFun)
+        message = response.Body.Data.choices(1).message;
+    else
+        message = struct("role", "assistant", ...
+            "content", streamedText);
+    end
+    if isfield(message, "tool_choice")
+        text = "";
+    else
+        text = string(message.content);
+    end
+else
+    text = "";
+    message = struct();
+end
+end
+
+function parameters = buildParametersCall(messages, functions, nvp)
+% Builds a struct in the format that is expected by the API, combining
+% MESSAGES, FUNCTIONS and parameters in NVP.
+
+parameters = struct();
+parameters.messages = messages;
+
+parameters.stream = ~isempty(nvp.StreamFun);
+
+if ~isempty(functions)
+    parameters.tools = functions;
+end
+
+if ~isempty(nvp.ToolChoice)
+    parameters.tool_choice = nvp.ToolChoice;
+end
+
+if ~isempty(nvp.Seed)
+    parameters.seed = nvp.Seed;
+end
+
+dict = mapNVPToParameters;
+
+nvpOptions = keys(dict);
+for opt = nvpOptions.'
+    if isfield(nvp, opt)
+        parameters.(dict(opt)) = nvp.(opt);
+    end
+end
+end
+
+function dict = mapNVPToParameters()
+dict = dictionary();
+dict("Temperature") = "temperature";
+dict("TopProbabilityMass") = "top_p";
+dict("NumCompletions") = "n";
+dict("StopSequences") = "stop";
+dict("MaxNumTokens") = "max_tokens";
+dict("PresencePenalty") = "presence_penalty";
+dict("FrequencyPenalty") = "frequency_penalty";
+end

+llms/+internal/callOllamaChatAPI.m

@@ -0,0 +1,127 @@
+function [text, message, response] = callOllamaChatAPI(model, messages, nvp)
+%callOllamaChatAPI Calls the ollama chat completions API.
+%
+%   MESSAGES and FUNCTIONS should be structs matching the json format
+%   required by the ollama Chat Completions API.
+%   Ref: https://github.com/ollama/ollama/blob/main/docs/api.md
+%
+%   Currently, the supported NVP are, including the equivalent name in the API:
+%  TODO TODO TODO
+%    - Temperature (temperature)
+%    - TopProbabilityMass (top_p)
+%    - NumCompletions (n)
+%    - StopSequences (stop)
+%    - MaxNumTokens (max_tokens)
+%    - PresencePenalty (presence_penalty)
+%    - FrequencyPenalty (frequence_penalty)
+%    - ResponseFormat (response_format)
+%    - Seed (seed)
+%    - ApiKey
+%    - TimeOut
+%    - StreamFun
+%   More details on the parameters: https://platform.openai.com/docs/api-reference/chat/create
+%
+%   Example
+%
+%   % Create messages struct
+%   messages = {struct("role", "system",...
+%       "content", "You are a helpful assistant");
+%       struct("role", "user", ...
+%       "content", "What is the edit distance between hi and hello?")};
+%
+%   % Create functions struct
+%   functions = {struct("name", "editDistance", ...
+%       "description", "Find edit distance between two strings or documents.", ...
+%       "parameters", struct( ...
+%       "type", "object", ...
+%       "properties", struct(...
+%           "str1", struct(...
+%               "description", "Source string.", ...
+%               "type", "string"),...
+%           "str2", struct(...
+%               "description", "Target string.", ...
+%               "type", "string")),...
+%       "required", ["str1", "str2"]))};
+%
+%   % Define your API key
+%   apiKey = "your-api-key-here"
+%
+%   % Send a request
+%   [text, message] = llms.internal.callOpenAIChatAPI(messages, functions, ApiKey=apiKey)
+
+%   Copyright 2023-2024 The MathWorks, Inc.
+
+arguments
+    model
+    messages
+    nvp.Temperature = 1
+    nvp.TopProbabilityMass = 1
+    nvp.NumCompletions = 1
+    nvp.StopSequences = []
+    nvp.MaxNumTokens = inf
+    nvp.PresencePenalty = 0
+    nvp.FrequencyPenalty = 0
+    nvp.ResponseFormat = "text"
+    nvp.Seed = []
+    nvp.TimeOut = 10
+    nvp.StreamFun = []
+end
+
+URL = "http://localhost:11434/api/chat"; % TODO: model parameter
+
+parameters = buildParametersCall(model, messages, nvp);
+
+[response, streamedText] = llms.internal.sendRequest(parameters,[],URL,nvp.TimeOut,nvp.StreamFun);
+
+% If call errors, "choices" will not be part of response.Body.Data, instead
+% we get response.Body.Data.error
+if response.StatusCode=="OK"
+    % Outputs the first generation
+    if isempty(nvp.StreamFun)
+        message = response.Body.Data.message;
+    else
+        message = struct("role", "assistant", ...
+            "content", streamedText);
+    end
+    text = string(message.content);
+else
+    text = "";
+    message = struct();
+end
+end
+
+function parameters = buildParametersCall(model, messages, nvp)
+% Builds a struct in the format that is expected by the API, combining
+% MESSAGES, FUNCTIONS and parameters in NVP.
+
+parameters = struct();
+parameters.model = model;
+parameters.messages = messages;
+
+parameters.stream = ~isempty(nvp.StreamFun);
+
+options = struct;
+if ~isempty(nvp.Seed)
+    options.seed = nvp.Seed;
+end
+
+dict = mapNVPToParameters;
+
+nvpOptions = keys(dict);
+for opt = nvpOptions.'
+    if isfield(nvp, opt)
+        options.(dict(opt)) = nvp.(opt);
+    end
+end
+
+parameters.options = options;
+end
+
+function dict = mapNVPToParameters()
+dict = dictionary();
+dict("Temperature") = "temperature";
+dict("TopProbabilityMass") = "top_p";
+dict("NumCompletions") = "n";
+dict("StopSequences") = "stop";
+dict("MaxNumTokens") = "num_predict";
+end

+llms/+internal/getApiKeyFromNvpOrEnv.m

@@ -1,23 +1,23 @@
-function key = getApiKeyFromNvpOrEnv(nvp)
+function key = getApiKeyFromNvpOrEnv(nvp,envVarName)
 % This function is undocumented and will change in a future release
 
 %getApiKeyFromNvpOrEnv Retrieves an API key from a Name-Value Pair struct or environment variable.
 %
-%   This function takes a struct nvp containing name-value pairs and checks
-%   if it contains a field called "ApiKey". If the field is not found, 
-%   the function attempts to retrieve the API key from an environment
-%   variable called "OPENAI_API_KEY". If both methods fail, the function 
-%   throws an error.
+%   This function takes a struct nvp containing name-value pairs and checks if
+%   it contains a field called "ApiKey". If the field is not found, the
+%   function attempts to retrieve the API key from an environment variable
+%   whose name is given as the second argument. If both methods fail, the
+%   function throws an error.
 
-%   Copyright 2023 The MathWorks, Inc.
+%   Copyright 2023-2024 The MathWorks, Inc.
 
     if isfield(nvp, "ApiKey")
         key = nvp.ApiKey;
     else
-        if isenv("OPENAI_API_KEY")
-            key = getenv("OPENAI_API_KEY");
+        if isenv(envVarName)
+            key = getenv(envVarName);
         else
-            error("llms:keyMustBeSpecified", llms.utils.errorMessageCatalog.getMessage("llms:keyMustBeSpecified"));
+            error("llms:keyMustBeSpecified", llms.utils.errorMessageCatalog.getMessage("llms:keyMustBeSpecified", envVarName));
         end
     end
-end
+end

+llms/+internal/sendRequest.m

@@ -1,6 +1,6 @@
 function [response, streamedText] = sendRequest(parameters, token, endpoint, timeout, streamFun)
 %sendRequest Sends a request to an ENDPOINT using PARAMETERS and
-%   api key TOKEN. TIMEOUT is the nubmer of seconds to wait for initial
+%   api key TOKEN. TIMEOUT is the number of seconds to wait for initial
 %   server connection. STREAMFUN is an optional callback function.
 
 %   Copyright 2023 The MathWorks, Inc.
@@ -15,17 +15,20 @@
 
 % Define the headers for the API request
 
-headers = [matlab.net.http.HeaderField('Content-Type', 'application/json')...
-    matlab.net.http.HeaderField('Authorization', "Bearer " + token)];
+headers = matlab.net.http.HeaderField('Content-Type', 'application/json');
+if ~isempty(token)
+    headers = [headers ...
+        matlab.net.http.HeaderField('Authorization', "Bearer " + token)...
+        matlab.net.http.HeaderField('api-key',token)];
+end
 
 % Define the request message
 request = matlab.net.http.RequestMessage('post',headers,parameters);
 
-% Create a HTTPOptions object;
+% set the timeout
 httpOpts = matlab.net.http.HTTPOptions;
-
-% Set the ConnectTimeout option
 httpOpts.ConnectTimeout = timeout;
+httpOpts.ResponseTimeout = timeout;
 
 % Send the request and store the response
 if isempty(streamFun)

+llms/+internal/textGenerator.m

@@ -0,0 +1,43 @@
+classdef (Abstract) textGenerator
+    %
+
+    % Copyright 2023-2024 The MathWorks, Inc.
+
+    properties
+        %TEMPERATURE   Temperature of generation.
+        Temperature {llms.utils.mustBeValidTemperature} = 1
+
+        %TOPPROBABILITYMASS   Top probability mass to consider for generation.
+        TopProbabilityMass {llms.utils.mustBeValidTopP} = 1
+
+        %STOPSEQUENCES   Sequences to stop the generation of tokens.
+        StopSequences {llms.utils.mustBeValidStop} = {}
+
+        %PRESENCEPENALTY   Penalty for using a token in the response that has already been used.
+        PresencePenalty {llms.utils.mustBeValidPenalty} = 0
+
+        %FREQUENCYPENALTY   Penalty for using a token that is frequent in the training data.
+        FrequencyPenalty {llms.utils.mustBeValidPenalty} = 0
+    end
+
+    properties (SetAccess=protected)
+        %TIMEOUT    Connection timeout in seconds (default 10 secs)
+        TimeOut
+
+        %FUNCTIONNAMES   Names of the functions that the model can request calls
+        FunctionNames
+
+        %SYSTEMPROMPT   System prompt.
+        SystemPrompt = []
+
+        %RESPONSEFORMAT     Response format, "text" or "json"
+        ResponseFormat
+    end
+
+    properties (Access=protected)
+        Tools
+        FunctionsStruct
+        ApiKey
+        StreamFun
+    end
+end