Merge branch 'master' into feature/3654-prompt-caching

Author: peterbanda

build.sbt

@@ -7,7 +7,7 @@ val scala3 = "3.2.2"
 
 ThisBuild / organization := "io.cequence"
 ThisBuild / scalaVersion := scala212
-ThisBuild / version := "1.1.1.RC.10"
+ThisBuild / version := "1.1.1.RC.11"
 ThisBuild / isSnapshot := false
 
 lazy val commonSettings = Seq(

openai-core/src/main/scala/io/cequence/openaiscala/domain/NonOpenAIModelId.scala

@@ -6,6 +6,7 @@ object NonOpenAIModelId {
   // Anthropic
   val claude_3_5_sonnet_20241022 = "claude-3-5-sonnet-20241022"
   val claude_3_5_sonnet_20240620 = "claude-3-5-sonnet-20240620"
+  val claude_3_5_haiku_20241022 = "claude-3-5-haiku-20241022"
   val claude_3_opus_20240229 = "claude-3-opus-20240229"
   val claude_3_sonnet_20240229 = "claude-3-sonnet-20240229"
   val claude_3_haiku_20240307 = "claude-3-haiku-20240307"

openai-core/src/main/scala/io/cequence/openaiscala/service/OpenAIChatCompletionExtra.scala

@@ -1,9 +1,10 @@
 package io.cequence.openaiscala.service
 
 import akka.actor.Scheduler
+import com.fasterxml.jackson.core.JsonParseException
 import io.cequence.openaiscala.JsonFormats.eitherJsonSchemaFormat
 import io.cequence.openaiscala.RetryHelpers.RetrySettings
-import io.cequence.openaiscala.{RetryHelpers, Retryable}
+import io.cequence.openaiscala.{OpenAIScalaClientException, RetryHelpers, Retryable}
 import io.cequence.openaiscala.domain.response.ChatCompletionResponse
 import io.cequence.openaiscala.domain.settings.{
   ChatCompletionResponseFormatType,
@@ -57,7 +58,7 @@ object OpenAIChatCompletionExtra {
     def createChatCompletionWithJSON[T: Format](
       messages: Seq[BaseMessage],
       settings: CreateChatCompletionSettings,
-      failoverModels: Seq[String],
+      failoverModels: Seq[String] = Nil,
       maxRetries: Option[Int] = Some(defaultMaxRetries),
       retryOnAnyError: Boolean = false,
       taskNameForLogging: Option[String] = None
@@ -92,7 +93,7 @@ object OpenAIChatCompletionExtra {
           val content = response.choices.head.message.content
           val contentTrimmed = content.stripPrefix("```json").stripSuffix("```").trim
           val contentJson = contentTrimmed.dropWhile(_ != '{')
-          val json = Json.parse(contentJson)
+          val json = parseJsonOrThrow(contentJson)
 
           logger.debug(
             s"${taskNameForLoggingFinal.capitalize} finished in " + (new java.util.Date().getTime - start.getTime) + " ms."
@@ -102,6 +103,17 @@ object OpenAIChatCompletionExtra {
         }
     }
 
+    private def parseJsonOrThrow(
+      jsonString: String
+    ) = try {
+      Json.parse(jsonString)
+    } catch {
+      case e: JsonParseException =>
+        val message = "Failed to parse JSON response:\n" + jsonString
+        logger.error(message)
+        throw new OpenAIScalaClientException(message, e)
+    }
+
     private def isRetryable(
       retryOnAnyError: Boolean
     ): Throwable => Boolean =

openai-core/src/main/scala/io/cequence/openaiscala/service/adapter/OpenAIServiceAdapters.scala

@@ -3,6 +3,7 @@ package io.cequence.openaiscala.service.adapter
 import akka.actor.Scheduler
 import akka.stream.Materializer
 import io.cequence.openaiscala.RetryHelpers.RetrySettings
+import io.cequence.openaiscala.Retryable
 import io.cequence.openaiscala.domain.BaseMessage
 import io.cequence.openaiscala.domain.settings.CreateChatCompletionSettings
 import io.cequence.openaiscala.service._
@@ -44,13 +45,17 @@ trait OpenAIServiceAdapters[S <: CloseableService] {
 
   def retry(
     underlying: S,
-    log: Option[String => Unit] = None
+    log: Option[String => Unit] = None,
+    isRetryable: Throwable => Boolean = {
+      case Retryable(_) => true
+      case _            => false
+    }
   )(
     implicit ec: ExecutionContext,
     retrySettings: RetrySettings,
     scheduler: Scheduler
   ): S =
-    wrapAndDelegate(new RetryServiceAdapter(underlying, log))
+    wrapAndDelegate(new RetryServiceAdapter(underlying, log, isRetryable))
 
   def log(
     underlying: S,

openai-core/src/main/scala/io/cequence/openaiscala/service/adapter/RetryServiceAdapter.scala

@@ -1,15 +1,16 @@
 package io.cequence.openaiscala.service.adapter
 
 import akka.actor.Scheduler
-import io.cequence.openaiscala.RetryHelpers
+import io.cequence.openaiscala.{RetryHelpers, Retryable}
 import io.cequence.openaiscala.RetryHelpers.RetrySettings
 import io.cequence.wsclient.service.CloseableService
 
 import scala.concurrent.{ExecutionContext, Future}
 
 private class RetryServiceAdapter[+S <: CloseableService](
   underlying: S,
-  log: Option[String => Unit] = None
+  log: Option[String => Unit] = None,
+  isRetryable: Throwable => Boolean
 )(
   implicit ec: ExecutionContext,
   retrySettings: RetrySettings,
@@ -24,7 +25,8 @@ private class RetryServiceAdapter[+S <: CloseableService](
   ): Future[T] =
     fun(underlying).retryOnFailure(
       Some(s"${getFunctionName().capitalize} call failed"),
-      log
+      log,
+      isRetryable
     )
 
   override def close(): Unit =

openai-examples/src/main/scala-2/io/cequence/openaiscala/examples/CreateChatCompletionJsonForCaseClass.scala

@@ -1,51 +1,53 @@
 package io.cequence.openaiscala.examples
 
 import io.cequence.openaiscala.domain._
-import io.cequence.openaiscala.domain.settings.JsonSchemaDef
-import io.cequence.openaiscala.examples.fixtures.TestFixtures
-import io.cequence.openaiscala.service.{JsonSchemaReflectionHelper, OpenAIServiceConsts}
-import play.api.libs.json.Json
+import io.cequence.openaiscala.domain.settings.{CreateChatCompletionSettings, JsonSchemaDef}
+import io.cequence.openaiscala.service.JsonSchemaReflectionHelper
+import play.api.libs.json.{Format, Json}
+import io.cequence.openaiscala.service.OpenAIChatCompletionExtra._
 
 import scala.concurrent.Future
 
-// experimental
-object CreateChatCompletionJsonForCaseClass
-    extends Example
-    with TestFixtures
-    with JsonSchemaReflectionHelper
-    with OpenAIServiceConsts {
+// due to the reflection used in jsonSchemaFor, this example currently works only for Scala 2.12 and 2.13
+object CreateChatCompletionJsonForCaseClass extends Example with JsonSchemaReflectionHelper {
 
-  private val messages = Seq(
-    SystemMessage(capitalsPrompt),
-    UserMessage("List only african countries")
-  )
-
-  // Case class(es)
-  private case class CapitalsResponse(
-    countries: Seq[Country]
-  )
-
-  private case class Country(
+  // data model
+  case class Country(
     country: String,
-    capital: String
+    capital: String,
+    populationMil: Double
   )
+  case class CapitalsResponse(capitals: Seq[Country])
 
-  // json schema def
-  private val jsonSchemaDef: JsonSchemaDef = JsonSchemaDef(
+  // JSON format and schema
+  implicit val countryFormat: Format[Country] = Json.format[Country]
+  implicit val capitalsResponseFormat: Format[CapitalsResponse] = Json.format[CapitalsResponse]
+
+  val jsonSchema: JsonSchemaDef = JsonSchemaDef(
     name = "capitals_response",
     strict = true,
-    // reflective json schema for case class
-    structure = jsonSchemaFor[CapitalsResponse]()
+    jsonSchemaFor[CapitalsResponse]()
+  )
+
+  // messages / prompts
+  val messages: Seq[BaseMessage] = Seq(
+    SystemMessage("You are an expert geographer"),
+    UserMessage("List the most populous African countries in the prescribed JSON format")
   )
 
-  override protected def run: Future[_] =
+  override protected def run: Future[_] = {
+    // chat completion JSON run
     service
-      .createChatCompletion(
-        messages = messages,
-        settings = DefaultSettings.createJsonChatCompletion(jsonSchemaDef)
+      .createChatCompletionWithJSON[CapitalsResponse](
+        messages,
+        settings = CreateChatCompletionSettings(
+          model = ModelId.gpt_4o_2024_08_06,
+          temperature = Some(0),
+          jsonSchema = Some(jsonSchema)
+        )
       )
       .map { response =>
-        val json = Json.parse(messageContent(response))
-        println(Json.prettyPrint(json))
+        response.capitals.foreach(println)
       }
+  }
 }

openai-examples/src/main/scala/io/cequence/openaiscala/examples/Example.scala

@@ -1,6 +1,6 @@
 package io.cequence.openaiscala.examples
 
-import akka.actor.ActorSystem
+import akka.actor.{ActorSystem, Scheduler}
 import akka.stream.Materializer
 import io.cequence.openaiscala.domain.response.ChatCompletionResponse
 import io.cequence.openaiscala.service.{OpenAIService, OpenAIServiceFactory}
@@ -16,6 +16,7 @@ trait ExampleBase[T <: CloseableService] {
 
   implicit val system: ActorSystem = ActorSystem()
   implicit val materializer: Materializer = Materializer(system)
+  implicit val scheduler: Scheduler = system.scheduler
   implicit val ec: ExecutionContext = ExecutionContext.Implicits.global
 
   protected val service: T

openai-examples/src/main/scala/io/cequence/openaiscala/examples/adapter/RetryAdapterExample.scala

@@ -18,7 +18,6 @@ object RetryAdapterExample extends ExampleBase[OpenAIService] {
 
   // implicit retry settings and scheduler
   private implicit val retrySettings: RetrySettings = RetrySettings(maxRetries = 4)
-  private implicit val scheduler: Scheduler = system.scheduler
 
   // regular OpenAI service
   private val regularService = OpenAIServiceFactory()