fix stop sequence for llama

src/modules/chat/chat.rb

@@ -7,7 +7,7 @@
 
 #=============<[ Constants ]>================#
 INST = "A chat between a very important human and an artificial intelligence assistant. The assistant gives quick and truthful answers to the human's questions. The assistant's responses are thorough, but succinct."
-CHAT = "\n@User: Hello.\n@Wayland: Greetings.\n@User: What do you call yourself?\n@Wayland: Wayland.\n@User: What is the closest star to our sun?\n@Wayland: The closest star to our sun Sol is Alpha Centauri."
+CHAT = "\nUser: Hello.\nWayland: Greetings.\nUser: What do you call yourself?\nWayland: Wayland.\nUser: What is the closest star to our sun?\nWayland: The closest star to our sun Sol is Alpha Centauri."
 
 #=============<[ Local Vars ]>================#
 core_config = RedFairy.new("shikigami")
@@ -29,14 +29,14 @@
 
 #=============<[ Methods ]>==================#
 # Parameters passed to llama.cpp running Llama 3
-def format_question(prompt)
+def format_question(prompt, sender)
   i = rand(99)
   request = {
     "stream" => false,        # keep false, breaks if true
     "seed" => i,              # Set the random number generator (RNG) seed.
     "n_predict" => 500,       # notes
     "temperature" => 0.56,    # was:0, def:0-1, higher is more creative (0.49 failed to answer question about unicorns)
-    "stop" => ["\n@User:"],   # notes
+    "stop" => ["\n#{sender}:", "\nUser:"],   # notes
     "repeat_last_n" => 128,   # Last n tokens to consider for penalizing repetition. 0 is disabled and -1 is ctx-size.
     "repeat_penalty" => 1.2,  # Control the repetition of token sequences in the generated text.
     "top_k" => 34,            # def:40, Limit the next token selection to the K most probable tokens.
@@ -55,21 +55,21 @@ def format_question(prompt)
 end #def
 
 # HTTP request interface to llama.cpp server
-def ask_question(q)
-  question = format_question(q)
+def ask_question(q, s)
+  question = format_question(q, s)
   response = HTTP.post("http://localhost:4242/completion", :json => question)
   h = JSON.parse(response.body)
   return h["content"]
 end #def
 
 # Discord chat logic to receive msg and send response
-def get_response(question)
+def get_response(question, sender)
   @sprig.pm2_log("Received msg: #{question}")
-  answer = ask_question(INST + CHAT + "\n" + question + "\n@Wayland:")
+  answer = ask_question(INST + CHAT + "\n" + question + "\n@Wayland:", sender)
   @sprig.pm2_log("Sending msg: #{answer}")
-  if answer.include? "@Wayland:"
-    @sprig.pm2_log("@Wayland string detected, removing..")
-    answer.gsub!("@Wayland:", "").to_s
+  if answer.include? "Wayland:"
+    @sprig.pm2_log("Wayland string detected, removing..")
+    answer.gsub!("Wayland:", "").to_s
   end #if
   return answer.to_s
 end #def
@@ -81,7 +81,7 @@ def get_response(question)
   loop do
     msg_hash = @sprig.get_msg
     begin
-      @answer = get_response(msg_hash["msg"])
+      @answer = get_response(msg_hash["msg"], msg_hash["from"])
     rescue Exception => e
       @sprig.pm2_log("Rescued job: #{e}")
     end #begin

src/modules/discord/discord.rb

@@ -59,7 +59,6 @@ def respond(e)
     @sprig.pm2_log("Received message #{event.message}")
   end
   bot.mention(allow_role_mention: true) do |event|
-    @sprig.pm2_log("Responding to [#{event.message}]")
     respond(event)
   end
   at_exit { bot.stop }