remove legacy prompt_options

lumni/src/apps/builtin/llm/prompt/src/chat/db/schema.rs

@@ -35,7 +35,6 @@ pub struct Conversation {
     pub parent_conversation_id: Option<ConversationId>,
     pub fork_exchange_id: Option<ExchangeId>,
     pub completion_options: Option<serde_json::Value>,
-    pub prompt_options: Option<serde_json::Value>,
     pub schema_version: i64,
     pub created_at: i64,
     pub updated_at: i64,

lumni/src/apps/builtin/llm/prompt/src/chat/db/schema.sql

@@ -20,7 +20,6 @@ CREATE TABLE conversations (
     parent_conversation_id INTEGER,
     fork_exchange_id INTEGER,
     completion_options TEXT, -- JSON string
-    prompt_options TEXT, -- JSON string
     schema_version INTEGER NOT NULL,
     created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
     updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,

lumni/src/apps/builtin/llm/prompt/src/chat/db/store.rs

@@ -25,7 +25,6 @@ impl ConversationDatabaseStore {
         name: &str,
         parent_id: Option<ConversationId>,
         completion_options: Option<serde_json::Value>,
-        prompt_options: Option<serde_json::Value>,
     ) -> Result<ConversationId, SqliteError> {
         let conversation = Conversation {
             id: ConversationId(-1), // Temporary ID
@@ -35,7 +34,6 @@ impl ConversationDatabaseStore {
             parent_conversation_id: parent_id,
             fork_exchange_id: None,
             completion_options,
-            prompt_options,
             schema_version: 1,
             created_at: 0,
             updated_at: 0,
@@ -74,10 +72,10 @@ impl ConversationDatabaseStore {
             "INSERT INTO conversations (
                 name, metadata, model_id, parent_conversation_id, \
              fork_exchange_id, 
-                completion_options, prompt_options, schema_version, 
+                completion_options, schema_version, 
                 created_at, updated_at, is_deleted
             )
-            VALUES ('{}', {}, {}, {}, {}, {}, {}, {}, {}, {}, {});",
+            VALUES ('{}', {}, {}, {}, {}, {}, {}, {}, {}, {});",
             conversation.name.replace("'", "''"),
             serde_json::to_string(&conversation.metadata)
                 .map(|s| format!("'{}'", s.replace("'", "''")))
@@ -96,13 +94,6 @@ impl ConversationDatabaseStore {
                     serde_json::to_string(v).unwrap().replace("'", "''")
                 )
             ),
-            conversation.prompt_options.as_ref().map_or(
-                "NULL".to_string(),
-                |v| format!(
-                    "'{}'",
-                    serde_json::to_string(v).unwrap().replace("'", "''")
-                )
-            ),
             conversation.schema_version,
             conversation.created_at,
             conversation.updated_at,
@@ -282,26 +273,23 @@ impl ConversationDatabaseStore {
                     completion_options: row
                         .get::<_, Option<String>>(6)?
                         .map(|s| serde_json::from_str(&s).unwrap_or_default()),
-                    prompt_options: row
-                        .get::<_, Option<String>>(7)?
-                        .map(|s| serde_json::from_str(&s).unwrap_or_default()),
-                    schema_version: row.get(8)?,
-                    created_at: row.get(9)?,
-                    updated_at: row.get(10)?,
-                    is_deleted: row.get(11)?,
+                    schema_version: row.get(7)?,
+                    created_at: row.get(8)?,
+                    updated_at: row.get(9)?,
+                    is_deleted: row.get(10)?,
                 };
 
-                let message = if !row.get::<_, Option<i64>>(14)?.is_none() {
+                let message = if !row.get::<_, Option<i64>>(13)?.is_none() {
                     Some(Message {
-                        id: MessageId(row.get(14)?),
+                        id: MessageId(row.get(13)?),
                         conversation_id: conversation.id,
                         exchange_id: ExchangeId(row.get(0)?),
-                        role: row.get(15)?,
-                        message_type: row.get(16)?,
-                        content: row.get(17)?,
-                        has_attachments: row.get(18)?,
-                        token_length: row.get(19)?,
-                        created_at: row.get(20)?,
+                        role: row.get(14)?,
+                        message_type: row.get(15)?,
+                        content: row.get(16)?,
+                        has_attachments: row.get(17)?,
+                        token_length: row.get(18)?,
+                        created_at: row.get(19)?,
                         is_deleted: false,
                     })
                 } else {
@@ -335,7 +323,7 @@ impl ConversationDatabaseStore {
         let query = format!(
             "SELECT id, name, metadata, model_id, parent_conversation_id, \
              fork_exchange_id, 
-             completion_options, prompt_options, schema_version, 
+             completion_options, schema_version, 
              created_at, updated_at, is_deleted
              FROM conversations
              WHERE is_deleted = FALSE
@@ -363,13 +351,10 @@ impl ConversationDatabaseStore {
                     completion_options: row
                         .get::<_, Option<String>>(6)?
                         .map(|s| serde_json::from_str(&s).unwrap_or_default()),
-                    prompt_options: row
-                        .get::<_, Option<String>>(7)?
-                        .map(|s| serde_json::from_str(&s).unwrap_or_default()),
-                    schema_version: row.get(8)?,
-                    created_at: row.get(9)?,
-                    updated_at: row.get(10)?,
-                    is_deleted: row.get(11)?,
+                    schema_version: row.get(7)?,
+                    created_at: row.get(8)?,
+                    updated_at: row.get(9)?,
+                    is_deleted: row.get(10)?,
                 })
             })?;
 

lumni/src/apps/builtin/llm/prompt/src/chat/instruction.rs

@@ -3,20 +3,19 @@ use std::collections::HashMap;
 use lumni::api::error::ApplicationError;
 
 use super::db::{
-    self, ConversationCache, ConversationDatabaseStore, ConversationId,
-    Exchange, ExchangeId, Message,
+    ConversationCache, ConversationDatabaseStore, ConversationId, Exchange,
+    ExchangeId, Message,
 };
 use super::prompt::Prompt;
 use super::{
-    ChatCompletionOptions, ChatMessage, LLMDefinition, PromptOptions,
-    PromptRole, DEFAULT_N_PREDICT, DEFAULT_TEMPERATURE, PERSONAS,
+    ChatCompletionOptions, ChatMessage, PromptRole,
+    DEFAULT_N_PREDICT, DEFAULT_TEMPERATURE, PERSONAS,
 };
 pub use crate::external as lumni;
 
 pub struct PromptInstruction {
     cache: ConversationCache,
     completion_options: ChatCompletionOptions,
-    prompt_options: PromptOptions,
     prompt_template: Option<String>,
 }
 
@@ -31,7 +30,6 @@ impl Default for PromptInstruction {
         PromptInstruction {
             cache: ConversationCache::new(),
             completion_options,
-            prompt_options: PromptOptions::default(),
             prompt_template: None,
         }
     }
@@ -45,8 +43,9 @@ impl PromptInstruction {
         db_conn: &ConversationDatabaseStore,
     ) -> Result<Self, ApplicationError> {
         let mut prompt_instruction = PromptInstruction::default();
+        //let mut prompt_options = PromptOptions::default();
         if let Some(json_str) = options {
-            prompt_instruction.prompt_options.update_from_json(json_str);
+            //prompt_options.update_from_json(json_str);
             prompt_instruction
                 .completion_options
                 .update_from_json(json_str);
@@ -68,7 +67,7 @@ impl PromptInstruction {
                 None,
                 serde_json::to_value(&prompt_instruction.completion_options)
                     .ok(),
-                serde_json::to_value(&prompt_instruction.prompt_options).ok(),
+                //serde_json::to_value(&prompt_options).ok(),
             )?
         };
         prompt_instruction
@@ -152,7 +151,7 @@ impl PromptInstruction {
         // reset by creating a new conversation
         // TODO: clone previous conversation settings
         let current_conversation_id =
-            db_conn.new_conversation("New Conversation", None, None, None)?;
+            db_conn.new_conversation("New Conversation", None, None)?;
         self.cache.set_conversation_id(current_conversation_id);
         Ok(())
     }
@@ -231,9 +230,6 @@ impl PromptInstruction {
         question: &str,
         max_token_length: usize,
     ) -> Vec<ChatMessage> {
-        // token budget for the system prompt
-        let system_prompt_token_length = self.get_n_keep().unwrap_or(0);
-
         // add the partial exchange (question) to the conversation
         let exchange = self.subsequent_exchange();
 
@@ -257,7 +253,7 @@ impl PromptInstruction {
 
         // Collect messages while respecting token limits
         let mut messages: Vec<ChatMessage> = Vec::new();
-        let mut total_tokens = system_prompt_token_length;
+        let mut total_tokens = 0;
 
         let mut system_message: Option<ChatMessage> = None;
 
@@ -273,11 +269,14 @@ impl PromptInstruction {
                     msg.token_length.map(|len| len as usize).unwrap_or(0);
 
                 if msg.role == PromptRole::System {
+                    // store system_prompt for later insertion at the beginning
                     system_message = Some(ChatMessage {
                         role: msg.role,
                         content: msg.content.clone(),
                     });
-                    continue; // system prompt is included separately
+                    // system prompt is always included
+                    total_tokens += msg_token_length;
+                    continue;
                 }
                 if total_tokens + msg_token_length <= max_token_length {
                     total_tokens += msg_token_length;
@@ -309,26 +308,6 @@ impl PromptInstruction {
         &self.completion_options
     }
 
-    pub fn set_model(&mut self, model: &LLMDefinition) {
-        self.completion_options.update_from_model(model);
-    }
-
-    pub fn get_role_prefix(&self, role: PromptRole) -> &str {
-        self.prompt_options.get_role_prefix(role)
-    }
-
-    pub fn get_context_size(&self) -> Option<usize> {
-        self.prompt_options.get_context_size()
-    }
-
-    pub fn set_context_size(&mut self, context_size: usize) {
-        self.prompt_options.set_context_size(context_size);
-    }
-
-    pub fn get_n_keep(&self) -> Option<usize> {
-        self.completion_options.get_n_keep()
-    }
-
     pub fn get_prompt_template(&self) -> Option<&str> {
         self.prompt_template.as_deref()
     }

lumni/src/apps/builtin/llm/prompt/src/chat/options.rs

@@ -111,16 +111,6 @@ impl Default for RolePrefix {
     }
 }
 
-impl RolePrefix {
-    fn get_role_prefix(&self, prompt_role: PromptRole) -> &str {
-        match prompt_role {
-            PromptRole::User => &self.user,
-            PromptRole::Assistant => &self.assistant,
-            PromptRole::System => &self.system,
-        }
-    }
-}
-
 #[derive(Debug, Clone, Deserialize, Serialize)]
 pub struct PromptOptions {
     n_ctx: Option<usize>,
@@ -154,13 +144,4 @@ impl PromptOptions {
     pub fn get_context_size(&self) -> Option<usize> {
         self.n_ctx
     }
-
-    pub fn set_context_size(&mut self, context_size: usize) -> &mut Self {
-        self.n_ctx = Some(context_size);
-        self
-    }
-
-    pub fn get_role_prefix(&self, prompt_role: PromptRole) -> &str {
-        self.role_prefix.get_role_prefix(prompt_role)
-    }
 }

lumni/src/apps/builtin/llm/prompt/src/chat/session.rs

@@ -96,10 +96,7 @@ impl ChatSession {
         tx: mpsc::Sender<Bytes>,
         question: &str,
     ) -> Result<(), ApplicationError> {
-        let max_token_length = self
-            .server
-            .get_context_size(&mut self.prompt_instruction)
-            .await?;
+        let max_token_length = self.server.get_max_context_size().await?;
         let user_question = self.initiate_new_exchange(question).await?;
         let messages = self
             .prompt_instruction

lumni/src/apps/builtin/llm/prompt/src/server/llama/mod.rs

@@ -54,12 +54,8 @@ impl Llama {
 
         let system_prompt = LlamaServerSystemPrompt::new(
             instruction.to_string(),
-            prompt_instruction
-                .get_role_prefix(PromptRole::User)
-                .to_string(),
-            prompt_instruction
-                .get_role_prefix(PromptRole::Assistant)
-                .to_string(),
+            format!("### {}", PromptRole::User.to_string()),
+            format!("### {}", PromptRole::Assistant.to_string()),
         );
         let payload = LlamaServerPayload {
             prompt: "",
@@ -206,23 +202,12 @@ impl ServerTrait for Llama {
         self.model.as_ref()
     }
 
-    async fn get_context_size(
-        &self,
-        prompt_instruction: &mut PromptInstruction,
-    ) -> Result<usize, ApplicationError> {
-        let context_size = prompt_instruction.get_context_size();
-        match context_size {
-            Some(size) => Ok(size), // Return the context size if it's already set
-            None => {
-                // fetch the context size, and store it in the prompt options
-                let context_size = match self.get_props().await {
-                    Ok(props) => props.get_n_ctx(),
-                    Err(_) => DEFAULT_CONTEXT_SIZE,
-                };
-                prompt_instruction.set_context_size(context_size);
-                Ok(context_size)
-            }
-        }
+    async fn get_max_context_size(&self) -> Result<usize, ApplicationError> {
+        let context_size = match self.get_props().await {
+            Ok(props) => props.get_n_ctx(),
+            Err(_) => DEFAULT_CONTEXT_SIZE,
+        };
+        Ok(context_size)
     }
 }
 

lumni/src/apps/builtin/llm/prompt/src/server/mod.rs

@@ -132,23 +132,14 @@ impl ServerTrait for ModelServer {
         }
     }
 
-    async fn get_context_size(
-        &self,
-        prompt_instruction: &mut PromptInstruction,
-    ) -> Result<usize, ApplicationError> {
+    async fn get_max_context_size(&self) -> Result<usize, ApplicationError> {
         match self {
-            ModelServer::Llama(llama) => {
-                llama.get_context_size(prompt_instruction).await
-            }
-            ModelServer::Ollama(ollama) => {
-                ollama.get_context_size(prompt_instruction).await
-            }
+            ModelServer::Llama(llama) => llama.get_max_context_size().await,
+            ModelServer::Ollama(ollama) => ollama.get_max_context_size().await,
             ModelServer::Bedrock(bedrock) => {
-                bedrock.get_context_size(prompt_instruction).await
-            }
-            ModelServer::OpenAI(openai) => {
-                openai.get_context_size(prompt_instruction).await
+                bedrock.get_max_context_size().await
             }
+            ModelServer::OpenAI(openai) => openai.get_max_context_size().await,
         }
     }
 
@@ -260,10 +251,7 @@ pub trait ServerTrait: Send + Sync {
         start_of_stream: bool,
     ) -> Option<CompletionResponse>;
 
-    async fn get_context_size(
-        &self,
-        _prompt_instruction: &mut PromptInstruction,
-    ) -> Result<usize, ApplicationError> {
+    async fn get_max_context_size(&self) -> Result<usize, ApplicationError> {
         Ok(DEFAULT_CONTEXT_SIZE)
     }
 
@@ -285,7 +273,8 @@ pub trait ServerManager: ServerTrait {
     ) -> Result<(), ApplicationError> {
         log::debug!("Initializing server with model: {:?}", model);
         // update completion options from the model, i.e. stop tokens
-        prompt_instruction.set_model(&model);
+        // TODO: prompt_intruction should be re-initialized with the model
+        //prompt_instruction.set_model(&model);
         self.initialize_with_model(model, prompt_instruction).await
     }