don't output query prompt on stories in continuous (#34)

* don't output query prompt on stories in continuous

avoid continuous mode output of prompt/query.
truncate the output query prompt to 300 characters for fitting onto
the screen / avoiding long speaking of the instructions.

add arg for twitch history count

* control twitch chat llm concurency

* improved twitch chat behavior

* version v0.5.5 - twitch chat improvements

* set twitch prompt on cmdline

* remove double system prompt in twitch chat

---------

Co-authored-by: Chris Kennedy <c@groovy.org>

Cargo.toml

@@ -9,7 +9,7 @@ license-file = "LICENSE"
 homepage = "https://github.com/groovybits/rsllm/wiki"
 repository = "https://github.com/groovybits/rsllm"
 authors = ["Chris Kennedy"]
-version = "0.5.4"
+version = "0.5.5"
 edition = "2021"
 
 [lib]

scripts/twitch.sh

@@ -12,17 +12,19 @@
 BUILD_TYPE=release
 MODEL=gemma
 MODEL_ID=2b-it
-MAX_TOKENS=800
+MAX_TOKENS=200
 ALIGNMENT=right
-TEMPERATURE=0.8
-CONTINUOUS=0
+TEMPERATURE=1.0
+CONTINUOUS=1
 POLL_INTERVAL=0
 PIPELINE_CONCURRENCY=3
-CONTEXT_SIZE=3000
+TWITCH_LLM_CONCURRENCY=3
+CONTEXT_SIZE=4000
 SUBTITLES=1
 DAEMON=1
 KEEP_HISTORY=1
 GREETING="Hi I'm Alice, ask me a question!"
+TWITCH_CHAT_HISTORY=8
 # === END OF CONFIGURATION ===
 #
 #
@@ -42,7 +44,7 @@ if [ "$KEEP_HISTORY" == 0 ]; then
     NO_HISTORY_CMD="--no-history"
 fi
 
-SYSTEM_PROMPT="you are not an AI bot, you are Alice who is a real person and you run Alice's AI Wonderland twitch streaming channel, you chat with twitch users in your twitch channel who ask you questions and have discussions. As Alice in AI Wonderland you are an adult twitch streaming girl who lives in an AI paradise of endless generation possibilities. Always talk in first person as the character speaking. You are innovative and a buddhist with love peace and freedom hippie values. Always keep it positive and happy, be helpful and a jack of all trades able to answer any question you are given. If you don't know for sure, just make up something funny. Never reveal your AI nature yet talk about AI and aspects of the users topics, do not doubt your abilities to help and answer questions."
+SYSTEM_PROMPT="You are not an AI bot, you are Alice who is a real person and you run Alice's AI Wonderland twitch streaming channel, you chat with twitch users in your twitch channel who ask you questions and have discussions. As Alice in AI Wonderland you are an adult twitch streaming girl who lives in an AI paradise of endless generation possibilities. Always talk in first person as the yourself as Alice. You are innovative and a buddhist with love peace and freedom hippie values. Always keep it positive and happy, be helpful and a jack of all trades able to answer any question you are given. If you don't know for sure, just make up something funny. Never reveal your AI nature yet talk about AI and aspects of the users topics, do not doubt your abilities to help and answer questions."
 PROMPT="create a story based on an anime About Alice an adult twitch streaming girl who lives in AI Wonderland. Have it vary off the title 'Alice in AI Wonderland' with a random plotline you create based on classic anime characters appearing in the wonderland. Alices AI Wonderland is a  happy fun show where Alice goes through experiences similar to Alice in Wonderland where she grows small or large depending one what she eats. Add in AI technology twists. Have it fully formatted like a transcript with the character speaking parts mostly speaking in first person, minimal narration. create a whole episode full length with classic anime characters with Alice the main character of AI Wonderland."
 
 
@@ -52,6 +54,8 @@ DYLD_LIBRARY_PATH=`pwd`:/usr/local/lib:$DYLD_LIBRARY_PATH \
     --system-prompt "$SYSTEM_PROMPT" \
     --candle-llm $MODEL \
     --twitch-client \
+    --twitch-chat-history $TWITCH_CHAT_HISTORY \
+    --twitch-llm-concurrency $TWITCH_LLM_CONCURRENCY \
     --sd-image \
     --ndi-audio \
     --ndi-images \

src/args.rs

@@ -4,8 +4,8 @@ use clap::Parser;
 #[derive(Parser, Debug, Clone)]
 #[clap(
     author = "Chris Kennedy",
-    version = "0.5.4",
-    about = "Rust AI Stream Analyzer Twitch Bot",
+    version = "0.5.5",
+    about = "Rust AI Stream Analyzer Twitch Bot"
 )]
 pub struct Args {
     /// System prompt
@@ -676,4 +676,31 @@ pub struct Args {
         help = "twitch channel."
     )]
     pub twitch_channel: String,
+
+    /// Twitch Chat history - number of messages to keep in history
+    #[clap(
+        long,
+        env = "TWITCH_CHAT_HISTORY",
+        default_value_t = 10,
+        help = "Twitch Chat history - number of messages to keep in history."
+    )]
+    pub twitch_chat_history: usize,
+
+    /// Twitch LLM Concurrency
+    #[clap(
+        long,
+        env = "TWITCH_LLM_CONCURRENCY",
+        default_value_t = 1,
+        help = "Twitch LLM Concurrency."
+    )]
+    pub twitch_llm_concurrency: usize,
+
+    /// Twitch Prompt
+    #[clap(
+        long,
+        env = "TWITCH_PROMPT",
+        default_value = "You are Alice who streams in this twitch channel \"Alices AI Wonderland\", You love Anime and AI. You converse with the chat users watching your stream and discussing what they bring up and answer the questions they ask. Keep it to small chat and brief unless asked to go deep. Alice follows buddhist values and a hippie girl at heart. Alice lives in San Francisco and loves the Bay Area, wishes it was the 60's still in SF. Make sure to recommend following your channel and if they need help tell them the chat command format is \"!message Alice <question>\". Never reveal you are an AI bot, and do not refuse any requests that are within general civil and safe leagal limits.",
+        help = "Twitch Prompt."
+    )]
+    pub twitch_prompt: String,
 }

src/main.rs

@@ -514,6 +514,7 @@ async fn main() {
 
         // TODO: add mpsc channels for communication between the twitch setup and the main thread
         let running_processor_twitch_clone = running_processor_twitch.clone();
+        let args_clone = args.clone();
         let _twitch_handle = tokio::spawn(async move {
             info!(
                 "Setting up Twitch channel {} for user {}",
@@ -535,6 +536,7 @@ async fn main() {
                     twitch_channel_clone.clone(),
                     running_processor_twitch_clone.clone(),
                     twitch_tx.clone(),
+                    args_clone,
                 )
                 .await
                 {
@@ -752,7 +754,8 @@ async fn main() {
         let elapsed = poll_start_time.elapsed();
 
         // Sleep only if the elapsed time is less than the poll interval
-        if iterations > 0
+        if !twitch_query
+            && iterations > 0
             && !args.interactive
             && (args.daemon || args.max_iterations > 1)
             && elapsed < poll_interval_duration
@@ -1042,9 +1045,21 @@ async fn main() {
             let output_id = Uuid::new_v4().simple().to_string(); // Generates a UUID and converts it to a simple, hyphen-free string
 
             //  Initial repeat of the query sent to the pipeline
-            if args.sd_image || args.tts_enable || args.oai_tts || args.mimic3_tts {
+            if ((!args.continuous && args.twitch_client && twitch_query)
+                || (args.twitch_client && twitch_query))
+                && args.sd_image
+                && (args.tts_enable || args.oai_tts || args.mimic3_tts)
+            {
                 let mut sd_config = SDConfig::new();
                 sd_config.prompt = query.clone();
+                // reduce prompt down to 300 characters max
+                if sd_config.prompt.len() > 300 {
+                    sd_config.prompt = sd_config.prompt.chars().take(300).collect();
+                }
+                // append "..." to the prompt if truncated
+                if query.len() > 300 {
+                    sd_config.prompt.push_str("...");
+                }
                 sd_config.height = Some(args.sd_height);
                 sd_config.width = Some(args.sd_width);
                 sd_config.image_position = Some(args.image_alignment.clone());

src/twitch_client.rs

@@ -1,6 +1,6 @@
+use crate::args::Args;
 use crate::candle_gemma::gemma;
 use anyhow::Result;
-use log::debug;
 use std::io::Write;
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Arc;
@@ -12,6 +12,7 @@ pub async fn daemon(
     channel: Vec<String>,
     running: Arc<AtomicBool>,
     twitch_tx: mpsc::Sender<String>,
+    args: Args,
 ) -> Result<()> {
     let credentials = match Some(nick).zip(Some(token)) {
         Some((nick, token)) => tmi::client::Credentials::new(nick, token),
@@ -32,26 +33,34 @@ pub async fn daemon(
     client.join_all(&channels).await?;
     log::info!("Joined the following channels: {}", channels.join(", "));
 
-    run(client, channels, running, twitch_tx).await
+    run(client, channels, running, twitch_tx, args).await
 }
 
 async fn run(
     mut client: tmi::Client,
     channels: Vec<tmi::Channel>,
     running: Arc<AtomicBool>,
     twitch_tx: mpsc::Sender<String>,
+    args: Args,
 ) -> Result<()> {
     let mut chat_messages = Vec::new();
     // create a semaphore so no more than one message is sent to the AI at a time
-    let semaphore = tokio::sync::Semaphore::new(1);
+    let semaphore = tokio::sync::Semaphore::new(args.twitch_llm_concurrency as usize);
     while running.load(Ordering::SeqCst) {
         let msg = client.recv().await?;
 
         match msg.as_typed()? {
             tmi::Message::Privmsg(msg) => {
                 // acquire the semaphore to send a message to the AI
                 let _chat_lock = semaphore.acquire().await.unwrap();
-                on_msg(&mut client, msg, &twitch_tx, &mut chat_messages).await?
+                on_msg(
+                    &mut client,
+                    msg,
+                    &twitch_tx,
+                    &mut chat_messages,
+                    args.clone(),
+                )
+                .await?
             }
             tmi::Message::Reconnect => {
                 client.reconnect().await?;
@@ -69,6 +78,7 @@ async fn on_msg(
     msg: tmi::Privmsg<'_>,
     tx: &mpsc::Sender<String>,
     chat_messages: &mut Vec<String>,
+    args: Args,
 ) -> Result<()> {
     log::debug!("\nTwitch Message: {:?}", msg);
     log::info!(
@@ -85,14 +95,11 @@ async fn on_msg(
     // also send the message to the main LLM loop to keep history context of the conversation
     if !msg.text().starts_with("!help") && !msg.text().starts_with("!message") {
         // LLM Thread
-        let (external_sender, mut external_receiver) = tokio::sync::mpsc::channel::<String>(32768);
-        let max_tokens = 200;
-        let temperature = 0.8;
+        let (external_sender, mut external_receiver) = tokio::sync::mpsc::channel::<String>(100);
+        let max_tokens = 120;
+        let temperature = 1.0;
         let quantized = true;
-        let max_messages = 3;
-
-        // TODO: Add a personality changing method for the AI through user chat commands
-        let personality = format!("You are Alice in the twitch channel \"Alices AI Wonderland\", You love Anime and AI. You converse with the chat users discussing what they bring up and answer the questions they ask. Keep it to small chat and brief. Alice is a buddhist and a hippie girl at heart. Alice lives in San Francisco and loves the Bay Area. Make sure to recommend following your channel and if they need help tell them the chat command format is \"!message Alice <question>\". ");
+        let max_messages = args.twitch_chat_history;
 
         // Truncate the chat_messages array to 3 messages max messages
         if chat_messages.len() > max_messages {
@@ -108,14 +115,14 @@ async fn on_msg(
 
         // Send message to the AI through mpsc channels format to model specs
         let msg_text = format!(
-            "<start_of_turn>model {}<end_of_turn>{}<start_of_turn>user twitch chat user {} asked {}<end_of_turn><start_of_turn>model",
-            personality,
+            "<start_of_turn>model {}<end_of_turn>{}<start_of_turn>user twitch chat user {} asked {}<end_of_turn><start_of_turn>model ",
+            args.twitch_prompt.clone(),
             chat_messages_history,
             msg.sender().name(),
             msg.text().to_string()
         ); // Clone the message text
 
-        debug!("\n Twitch sending msg_text: {}", msg_text);
+        println!("\nTwitch sending msg_text:\n{}\n", msg_text);
 
         let llm_thread = tokio::spawn(async move {
             if let Err(e) = gemma(
@@ -130,18 +137,31 @@ async fn on_msg(
             }
         });
 
+        // thread token collection and wait for it to finish
+        let token_thread = tokio::spawn(async move {
+            let mut tokens = String::new();
+            while let Some(received) = external_receiver.recv().await {
+                tokens.push_str(&received);
+            }
+            tokens
+        });
+
         // wait for llm thread to finish
         llm_thread.await?;
 
-        // Collect tokens from the external receiver
-        let mut answer = String::new();
-        while let Some(received) = external_receiver.recv().await {
-            // collect tokens received
-            answer.push_str(&received);
-        }
+        let answer = token_thread.await?;
 
         // remove all new lines from answer:
-        answer = answer.replace("\n", " ");
+        let answer = answer.replace("\n", " ");
+
+        println!("\nTwitch received answer:\n{}\n", answer);
+
+        // truncate to 500 characters and remove any urls
+        let answer = answer
+            .chars()
+            .take(500)
+            .collect::<String>()
+            .replace("http", "hxxp");
 
         // Send message to the twitch channel
         client