Allow to pass various models to set_llm.

DESCRIPTION

@@ -11,6 +11,8 @@ License: MIT + file LICENSE
 Encoding: UTF-8
 Roxygen: list(markdown = TRUE)
 RoxygenNote: 7.3.2
+Depends:
+    R (>= 4.1.0)
 Imports:
     cli (>= 3.4.0),
     elmer,

NAMESPACE

@@ -10,6 +10,3 @@ export(set_llm)
 export(set_prompt)
 export(verbose_off)
 export(verbose_on)
-importFrom(R6,R6Class)
-importFrom(httr2,with_verbosity)
-importFrom(lubridate,as_datetime)

R/GitAI-package.R

@@ -1,4 +1,4 @@
-#' @importFrom R6 R6Class
-#' @importFrom httr2 with_verbosity
-#' @importFrom lubridate as_datetime
-NULL
+#' Derive knowledge from GitHub or GitLab repositories with the use of AI/LLM
+#'
+#' @name GitAI-package
+"_PACKAGE"

R/add_metadata.R

@@ -17,20 +17,26 @@ add_metadata <- function(result, content) {
 
 get_repo_date <- S7::new_generic("get_repo_date", "repo_api_url")
 
-github_repo <- S7::new_class("github_repo",
-                             properties = list(repo = S7::class_character))
+github_repo <- S7::new_class(
+  "github_repo",
+  properties = list(repo = S7::class_character)
+)
 
-gitlab_repo <- S7::new_class("gitlab_repo",
-                             properties = list(repo = S7::class_character))
+gitlab_repo <- S7::new_class(
+  "gitlab_repo",
+  properties = list(repo = S7::class_character)
+)
 
 S7::method(get_repo_date, github_repo) <- function(repo_api_url) {
   repo_data <- get_response(repo_api_url@repo)
   lubridate::as_datetime(repo_data$updated_at)
 }
 
 S7::method(get_repo_date, gitlab_repo) <- function(repo_api_url) {
-  repo_data <- get_response(endpoint = repo_api_url@repo,
-                            token = Sys.getenv("GITLAB_PAT"))
+  repo_data <- get_response(
+    endpoint = repo_api_url@repo,
+    token = Sys.getenv("GITLAB_PAT")
+  )
   lubridate::as_datetime(repo_data$last_activity_at)
 }
 

R/process_repos.R

@@ -9,10 +9,11 @@ process_repos <- function(gitai, verbose = is_verbose()) {
 
   gitstats <- gitai$gitstats
 
-  gitai$repos_metadata <-
-    GitStats::get_repos(gitstats,
-                        add_contributors = FALSE,
-                        verbose = verbose)
+  gitai$repos_metadata <- GitStats::get_repos(
+      gitstats,
+      add_contributors = FALSE,
+      verbose = verbose
+    )
 
   GitStats::get_files_structure(
     gitstats_object = gitstats,
@@ -22,31 +23,51 @@ process_repos <- function(gitai, verbose = is_verbose()) {
   )
   files_content <- GitStats::get_files_content(gitstats, verbose = verbose)
   repositories <- unique(files_content$repo_name)
-  results <-
-    repositories |>
-    purrr::map(function(repo_name) {
-      if (verbose) {
-        cli::cli_alert_info("Processing repository: {.pkg {repo_name}}")
-      }
-
-      filtered_content <-
-        files_content |>
-        dplyr::filter(repo_name == !!repo_name)
-      content_to_process <-
-        filtered_content |>
-        dplyr::pull(file_content) |>
-        paste(collapse = "\n\n")
-
-      result <- process_content(
-        gitai = gitai,
+  process_repo_content <- function(repo_name) {
+    if (verbose) {
+      cli::cli_alert_info("Processing repository: {.pkg {repo_name}}")
+    }
+
+    filtered_content <- files_content |>
+      dplyr::filter(repo_name == !!repo_name)
+    content_to_process <- filtered_content |>
+      dplyr::pull(file_content) |>
+      paste(collapse = "\n\n")
+
+    result <- gitai |>
+      process_content(
         content = content_to_process
       ) |>
-        add_metadata(
-          content = filtered_content
-        )
+      add_metadata(
+        content = filtered_content
+      )
 
-    }) |>
+  }
+
+  results <- repositories |>
+    purrr::map(process_repo_content) |>
     purrr::set_names(repositories)
 
   results
 }
+
+process_repo_content <- function(repo_name) {
+  if (verbose) {
+    cli::cli_alert_info("Processing repository: {.pkg {repo_name}}")
+  }
+
+  filtered_content <- files_content |>
+    dplyr::filter(repo_name == !!repo_name)
+  content_to_process <- filtered_content |>
+    dplyr::pull(file_content) |>
+    paste(collapse = "\n\n")
+
+  result <- gitai |>
+    process_content(
+      content = content_to_process
+    ) |>
+    add_metadata(
+      content = filtered_content
+    )
+
+}

R/set_llm.R

@@ -1,31 +1,44 @@
 #' Set Large Language Model in `GitAI` object.
+#'
 #' @name set_llm
 #' @param gitai A \code{GitAI} object.
-#' @param provider A LLM provider.
-#' @param model A LLM model.
-#' @param seed An integer to make results more reproducible.
-#' @param ... Other arguments to pass to `elmer::chat_openai()` function.
+#' @param provider Name of LLM provider, a string. Results with setting up LLM using
+#'   \code{elmer::chat_<provider>} function.
+#' @param ... Other arguments to pass to corresponding \code{elmer::chat_<provider>} function.
+#'   Please use \link{get_llm_defaults} to get default model arguments.
 #' @return A \code{GitAI} object.
 #' @export
-set_llm <- function(gitai,
-                    provider = "openai",
-                    model = "gpt-4o-mini",
-                    seed = NULL,
-                    ...) {
+set_llm <- function(gitai, provider = "openai", ...) {
 
-  if (provider == "openai") {
+  provider_method <- rlang::env_get(
+    env = asNamespace("elmer"),
+    nm = glue::glue("chat_{provider}")
+  )
+  provider_args <- purrr::list_modify(
+    get_llm_defaults(provider),
+    !!!rlang::dots_list(...)
+  )
 
-    gitai$llm <- elmer::chat_openai(
-      model = model,
-      echo = "none",
-      seed = seed,
-      ...
-    )
-  }
+  gitai$llm <- rlang::exec(provider_method, !!!provider_args)
 
   invisible(gitai)
 }
 
+llm_default_args <- list(
+  openai = list(model = "gpt-4o-mini", seed = NULL, echo = "none"),
+  ollama = list(model = "llama3.2", seed = NULL),
+  bedrock = list(model = "anthropic.claude-3-5-sonnet-20240620-v1:0")
+)
+
+#' @rdname set_llm
+get_llm_defaults <- function(provider) {
+  llm_defaults <- llm_default_args[[provider]]
+  if (!is.null(llm_defaults)) {
+    return(llm_defaults)
+  }
+  list()
+}
+
 #' Set prompt.
 #' @name set_prompt
 #' @param gitai A \code{GitAI} object.

R/set_repos.R

@@ -20,11 +20,17 @@ set_github_repos <- function(gitai,
     GitStats::set_github_host(
       host = host,
       repos = repos,
+      token = get_github_pat(),
       verbose = verbose
     )
   invisible(gitai)
 }
 
+get_github_pat <- function() {
+  key_get <- get("key_get", envir = asNamespace("elmer"))
+  key_get("GITHUB_PAT")
+}
+
 #' Set GitLab repositories in `GitAI` object.
 #' @name set_gitlab_repos
 #' @param gitai A \code{GitAI} object.
@@ -47,7 +53,13 @@ set_gitlab_repos <- function(gitai,
     GitStats::set_gitlab_host(
       host = host,
       repos = repos,
+      token = get_gitlab_pat(),
       verbose = verbose
     )
   invisible(gitai)
 }
+
+get_gitlab_pat <- function() {
+  key_get <- get("key_get", envir = asNamespace("elmer"))
+  key_get("GITLAB_PAT")
+}

tests/testthat/setup.R

@@ -1 +1,81 @@
 test_mocker <- Mocker$new()
+
+# Override other methods when needed in the future
+ChatMocked <- elmer:::Chat
+ChatMocked$public_methods$chat <- function(..., echo = NULL) {
+  if (self$get_system_prompt() == "You always return only 'Hi there!'") {
+    return("Hi there!")
+  }
+}
+
+# This method allows to skip original checks (e.g. for api or other args structure) and returns
+# object of class ChatMocked that we can modify for our testing purposes.
+mock_chat_method <- function(turns = NULL,
+                             echo = c("none", "text", "all"),
+                             ...,
+                             provider_class) {
+
+  provider_args <- rlang::dots_list(...)
+  provider <- rlang::exec(provider_class, !!!provider_args)
+
+  ChatMocked$new(provider = provider, turns = turns, echo = echo)
+}
+
+chat_openai_mocked <- function(system_prompt = NULL,
+                               turns = NULL,
+                               base_url = "https://api.mocked.com/v1",
+                               api_key = "mocked_key",
+                               model = NULL,
+                               seed = NULL,
+                               api_args = list(),
+                               echo = c("none", "text", "all")) {
+
+  turns <- elmer:::normalize_turns(turns, system_prompt)
+  model <- elmer:::set_default(model, "gpt-4o")
+  echo <- elmer:::check_echo(echo)
+
+  if (is.null(seed)) {
+    seed <- 1014
+  }
+
+  mock_chat_method(
+    turns = turns,
+    echo = echo,
+    base_url = base_url,
+    model = model,
+    seed = seed,
+    extra_args = api_args,
+    api_key = api_key,
+    provider_class = elmer:::ProviderOpenAI
+  )
+}
+
+chat_bedrock_mocked <- function(system_prompt = NULL,
+                                turns = NULL,
+                                model = NULL,
+                                profile = NULL,
+                                echo = NULL) {
+
+  credentials <- list(
+    access_key_id = "access_key_id_mocked",
+    secret_access_key = "access_key_id_mocked",
+    session_token = "session_token_mocked",
+    access_token = "access_token_mocked",
+    expiration = as.numeric(Sys.time() + 3600),
+    region = "eu-central-1"
+  )
+
+  turns <- elmer:::normalize_turns(turns, system_prompt)
+  model <- elmer:::set_default(model, "model_bedrock")
+  echo <- elmer:::check_echo(echo)
+
+  mock_chat_method(
+    turns = turns,
+    echo = echo,
+    base_url = "",
+    model = model,
+    profile = profile,
+    credentials = credentials,
+    provider_class = elmer:::ProviderBedrock
+  )
+}