✨ Experiment extracting out model object ✨

DESCRIPTION

@@ -86,6 +86,7 @@ Collate:
     'import-standalone-types-check.R'
     'interpolate.R'
     'live.R'
+    'model.R'
     'parallel-chat.R'
     'params.R'
     'provider-anthropic.R'

NAMESPACE

@@ -16,6 +16,7 @@ export(ContentText)
 export(ContentThinking)
 export(ContentToolRequest)
 export(ContentToolResult)
+export(Model)
 export(Provider)
 export(SystemTurn)
 export(Turn)

R/batch-chat.R

@@ -193,6 +193,7 @@ BatchJob <- R6::R6Class(
 
     # Internal state
     provider = NULL,
+    model = NULL,
     started_at = NULL,
     stage = NULL,
     batch = NULL,
@@ -208,6 +209,7 @@ BatchJob <- R6::R6Class(
       call = caller_env(2)
     ) {
       self$provider <- chat$get_provider()
+      self$model <- chat$get_model_obj()
       check_has_batch_support(self$provider, call = call)
 
       user_turns <- as_user_turns(prompts, call = call)
@@ -326,7 +328,7 @@ BatchJob <- R6::R6Class(
 
     retrieve = function() {
       self$results <- batch_retrieve(self$provider, self$batch)
-      log_turns(self$provider, self$result_turns())
+      log_turns(self$provider, self$model, self$result_turns())
 
       self$stage <- "done"
       self$save_state()
@@ -335,7 +337,12 @@ BatchJob <- R6::R6Class(
 
     result_turns = function() {
       map2(self$results, self$user_turns, function(result, user_turn) {
-        batch_result_turn(self$provider, result, has_type = !is.null(self$type))
+        batch_result_turn(
+          self$provider,
+          self$model,
+          result,
+          has_type = !is.null(self$type)
+        )
       })
     },
 
@@ -379,7 +386,6 @@ BatchJob <- R6::R6Class(
 provider_hash <- function(x) {
   list(
     name = x@name,
-    model = x@model,
     base_url = x@base_url
   )
 }

R/chat.R

@@ -24,6 +24,7 @@ Chat <- R6::R6Class(
   "Chat",
   public = list(
     #' @param provider A provider object.
+    #' @param model A model object.
     #' @param system_prompt System prompt to start the conversation with.
     #' @param echo One of the following options:
     #'   * `none`: don't emit any output (default when running in a function).
@@ -33,8 +34,14 @@ Chat <- R6::R6Class(
     #'
     #'  Note this only affects the `chat()` method. You can override the default
     #'  by setting the `ellmer_echo` option.
-    initialize = function(provider, system_prompt = NULL, echo = "none") {
+    initialize = function(
+      provider,
+      model,
+      system_prompt = NULL,
+      echo = "none"
+    ) {
       private$provider <- provider
+      private$model <- model
       private$echo <- echo
       private$callback_on_tool_request <- CallbackManager$new(args = "request")
       private$callback_on_tool_result <- CallbackManager$new(args = "result")
@@ -78,7 +85,7 @@ Chat <- R6::R6Class(
       check_turn(assistant)
 
       if (log_tokens) {
-        log_turn(private$provider, assistant)
+        log_turn(private$provider, private$model, assistant)
       }
 
       private$.turns[[length(private$.turns) + 1]] <- user
@@ -97,7 +104,7 @@ Chat <- R6::R6Class(
 
     #' @description Retrieve the model name
     get_model = function() {
-      private$provider@model
+      private$model@name
     },
 
     #' @description Update the system prompt
@@ -381,6 +388,11 @@ Chat <- R6::R6Class(
       private$provider
     },
 
+    #' @description Get the underlying model object. For expert use only.
+    get_model_obj = function() {
+      private$model
+    },
+
     #' @description Retrieve the list of registered tools.
     get_tools = function() {
       private$tools
@@ -422,6 +434,7 @@ Chat <- R6::R6Class(
   ),
   private = list(
     provider = NULL,
+    model = NULL,
 
     .turns = list(),
     echo = NULL,
@@ -579,6 +592,7 @@ Chat <- R6::R6Class(
 
       response <- chat_perform(
         provider = private$provider,
+        model = private$model,
         mode = if (stream) "stream" else "value",
         turns = c(private$.turns, list(user_turn)),
         tools = if (is.null(type)) private$tools,
@@ -603,11 +617,17 @@ Chat <- R6::R6Class(
 
           result <- stream_merge_chunks(private$provider, result, chunk)
         }
-        turn <- value_turn(private$provider, result, has_type = !is.null(type))
+        turn <- value_turn(
+          private$provider,
+          private$model,
+          result,
+          has_type = !is.null(type)
+        )
         turn <- match_tools(turn, private$tools)
       } else {
         turn <- value_turn(
           private$provider,
+          private$model,
           resp_body_json(response),
           has_type = !is.null(type)
         )
@@ -661,6 +681,7 @@ Chat <- R6::R6Class(
     ) {
       response <- chat_perform(
         provider = private$provider,
+        model = private$model,
         mode = if (stream) "async-stream" else "async-value",
         turns = c(private$.turns, list(user_turn)),
         tools = if (is.null(type)) private$tools,
@@ -685,12 +706,18 @@ Chat <- R6::R6Class(
 
           result <- stream_merge_chunks(private$provider, result, chunk)
         }
-        turn <- value_turn(private$provider, result, has_type = !is.null(type))
+        turn <- value_turn(
+          private$provider,
+          private$model,
+          result,
+          has_type = !is.null(type)
+        )
       } else {
         result <- await(response)
 
         turn <- value_turn(
           private$provider,
+          private$model,
           resp_body_json(result),
           has_type = !is.null(type)
         )
@@ -738,6 +765,7 @@ Chat <- R6::R6Class(
 #' @export
 print.Chat <- function(x, ...) {
   provider <- x$get_provider()
+  model <- x$get_model()
   turns <- x$get_turns(include_system_prompt = TRUE)
 
   assistant_turns <- keep(turns, \(x) x@role == "assistant")
@@ -746,7 +774,7 @@ print.Chat <- function(x, ...) {
 
   cat(paste_c(
     "<Chat",
-    c(" ", provider@name, "/", provider@model),
+    c(" ", provider@name, "/", model),
     c(" turns=", length(turns)),
     turn_cost(total_tokens, total_cost, prefix = " "),
     ">\n"

R/httr2.R

@@ -3,6 +3,7 @@
 # We will recconsider this in the future if necessary.
 chat_perform <- function(
   provider,
+  model,
   mode = c("value", "stream", "async-stream", "async-value"),
   turns,
   tools = NULL,
@@ -14,6 +15,7 @@ chat_perform <- function(
 
   req <- chat_request(
     provider = provider,
+    model = model,
     turns = turns,
     tools = tools,
     stream = stream,

R/model.R

@@ -0,0 +1,29 @@
+#' A chatbot model
+#'
+#' A Model captures the details of a specific language model and its
+#' configuration parameters. This includes the model name, parameters like
+#' temperature and max_tokens, and any extra arguments to include in API
+#' requests.
+#'
+#' Model objects are typically created internally by [chat_openai()],
+#' [chat_anthropic()], and other `chat_*()` functions. You generally won't
+#' need to create them directly unless you're implementing a custom provider.
+#'
+#' @export
+#' @param name Name of the model (e.g., "gpt-4", "claude-sonnet-4").
+#' @param params A list of standard parameters created by [params()].
+#' @param extra_args Arbitrary extra arguments to be included in the request body.
+#' @return An S7 Model object.
+#' @examples
+#' Model(
+#'   name = "gpt-4",
+#'   params = params(temperature = 0.7)
+#' )
+Model <- new_class(
+  "Model",
+  properties = list(
+    name = prop_string(),
+    params = class_list,
+    extra_args = class_list
+  )
+)

R/parallel-chat.R

@@ -78,6 +78,7 @@ parallel_chat <- function(
   my_parallel_turns <- function(conversations) {
     parallel_turns(
       provider = chat$get_provider(),
+      model = chat$get_model_obj(),
       conversations = conversations,
       tools = chat$get_tools(),
       max_active = max_active,
@@ -127,7 +128,7 @@ parallel_chat <- function(
   map(seq_along(conversations), function(i) {
     if (is_ok[[i]]) {
       turns <- conversations[[i]]
-      log_turns(chat$get_provider(), turns)
+      log_turns(chat$get_provider(), chat$get_model_obj(), turns)
       chat$clone()$set_turns(turns)
     } else {
       assistant_turns[[i]]
@@ -191,6 +192,7 @@ parallel_chat_structured <- function(
   on_error <- arg_match(on_error)
 
   provider <- chat$get_provider()
+  model <- chat$get_model_obj()
   needs_wrapper <- type_needs_wrapper(type, provider)
 
   # First build up list of cumulative conversations
@@ -200,14 +202,15 @@ parallel_chat_structured <- function(
 
   turns <- parallel_turns(
     provider = provider,
+    model = model,
     conversations = conversations,
     tools = chat$get_tools(),
     type = wrap_type_if_needed(type, needs_wrapper),
     max_active = max_active,
     rpm = rpm,
     on_error = on_error
   )
-  log_turns(provider, turns)
+  log_turns(provider, model, turns)
 
   multi_convert(
     provider,
@@ -227,7 +230,7 @@ multi_convert <- function(
   include_tokens = FALSE,
   include_cost = FALSE
 ) {
-  needs_wrapper <- type_needs_wrapper(type, provider)
+  needs_wrapper <- type_needs_wrapper(type, provider, NULL)
 
   rows <- map(turns, \(turn) {
     if (turn_failed(turn)) {
@@ -306,6 +309,7 @@ turn_failed <- function(turn) {
 
 parallel_turns <- function(
   provider,
+  model,
   conversations,
   tools,
   type = NULL,
@@ -316,6 +320,7 @@ parallel_turns <- function(
   reqs <- map(conversations, function(turns) {
     chat_request(
       provider = provider,
+      model = model,
       turns = turns,
       type = type,
       tools = tools,
@@ -352,7 +357,7 @@ parallel_turns <- function(
       resp
     } else {
       json <- resp_body_json(resp)
-      turn <- value_turn(provider, json, has_type = !is.null(type))
+      turn <- value_turn(provider, model, json, has_type = !is.null(type))
       turn@duration <- resp_timing(resp)[["total"]] %||% NA_real_
       turn
     }