changes according to comments and feedback

xebia-functional · Sep 20, 2023 · 4fcbc6e · 4fcbc6e
1 parent 0a27174
commit 4fcbc6e
Show file tree

Hide file tree

Showing 4 changed files with 3 additions and 36 deletions.
diff --git a/core/src/commonMain/kotlin/com/xebia/functional/xef/llm/LLM.kt b/core/src/commonMain/kotlin/com/xebia/functional/xef/llm/LLM.kt
@@ -17,7 +17,6 @@ sealed interface LLM : AutoCloseable {
   ): Int { // TODO: naive implementation with magic numbers
     fun Encoding.countTokensFromMessages(tokensPerMessage: Int, tokensPerName: Int): Int =
       messages.sumOf { message ->
-        message.role.name.length
         countTokens(message.role.name) +
           countTokens(message.content) +
           tokensPerMessage +

diff --git a/core/src/commonTest/kotlin/com/xebia/functional/xef/conversation/ConversationSpec.kt b/core/src/commonTest/kotlin/com/xebia/functional/xef/conversation/ConversationSpec.kt
@@ -117,40 +117,6 @@ class ConversationSpec :
       messagesSizePlusMessageResponse shouldBe memories.size
     }
 
-    // considered obsolete since Chat and ChatWithFunctions was separated
-    //    "functionCall should be null when the model doesn't support functions" {
-    //      val messages = generateRandomMessages(1, 40, 60)
-    //      val conversationId = ConversationId(UUID.generateUUID().toString())
-    //      val scope = Conversation(LocalVectorStore(TestEmbeddings()), conversationId =
-    // conversationId)
-    //
-    //      val model = TestModel(modelType = ModelType.ADA, name = "fake-model")
-    //
-    //      model.promptMessage(prompt = Prompt(messages.keys.first()), scope = scope)
-    //
-    //      val lastRequest = model.requests.last()
-    //
-    //      lastRequest.functionCall shouldBe null
-    //    }
-
-    // considered obsolete since Chat and ChatWithFunctions was separated
-    //    "functionCall should be null when the model support functions and the prompt doesn't
-    // contain a function" {
-    //      val messages = generateRandomMessages(1, 40, 60)
-    //      val conversationId = ConversationId(UUID.generateUUID().toString())
-    //      val scope = Conversation(LocalVectorStore(TestEmbeddings()), conversationId =
-    // conversationId)
-    //
-    //      val model =
-    //        TestFunctionsModel(modelType = ModelType.GPT_3_5_TURBO_FUNCTIONS, name = "fake-model")
-    //
-    //      model.createChatCompletion(prompt = Prompt(messages.keys.first()), scope = scope)
-    //
-    //      val lastRequest = model.requests.last()
-    //
-    //      lastRequest.functionCall shouldBe null
-    //    }
-
     "functionCall shouldn't be null when the model support functions and the prompt contain a function" {
       val question = "fake-question"
       val answer = Answer("fake-answer")

diff --git a/openai/src/commonMain/kotlin/com/xebia/functional/xef/conversation/llm/openai/OpenAI.kt b/openai/src/commonMain/kotlin/com/xebia/functional/xef/conversation/llm/openai/OpenAI.kt
@@ -69,7 +69,7 @@ class OpenAI(internal var token: String? = null, internal var host: String? = nu
   val GPT_4 by lazy { autoClose(OpenAIChat(ModelType.GPT_4, defaultClient)) }
 
   val GPT_4_0314 by lazy {
-    autoClose(OpenAIFunChat(ModelType.GPT_4, defaultClient)) // legacy
+    autoClose(OpenAIFunChat(ModelType.GPT_4_0314, defaultClient)) // legacy
   }
 
   val GPT_4_32K by lazy { autoClose(OpenAIChat(ModelType.GPT_4_32K, defaultClient)) }

diff --git a/tokenizer/src/commonMain/kotlin/com/xebia/functional/tokenizer/ModelType.kt b/tokenizer/src/commonMain/kotlin/com/xebia/functional/tokenizer/ModelType.kt
@@ -39,6 +39,8 @@ sealed class ModelType(
   data class LocalModel(override val name: String, override val encodingType: EncodingType, override val maxContextLength: Int) : ModelType(name, encodingType, maxContextLength)
   // chat
   object GPT_4 : ModelType("gpt-4", CL100K_BASE, 8192, tokensPerMessage = 3, tokensPerName = 2, tokenPadding = 5)
+
+  object GPT_4_0314 : ModelType("gpt-4-0314", CL100K_BASE, 8192, tokensPerMessage = 3, tokensPerName = 2, tokenPadding = 5)
   object GPT_4_32K : ModelType("gpt-4-32k", CL100K_BASE, 32768, tokensPerMessage = 3, tokensPerName = 2, tokenPadding = 5)
   object GPT_3_5_TURBO : ModelType("gpt-3.5-turbo", CL100K_BASE, 4097, tokensPerMessage = 4, tokensPerName = 0, tokenPadding = 5)
   object GPT_3_5_TURBO_16_K : ModelType("gpt-3.5-turbo-16k", CL100K_BASE, 4097 * 4)