From d76309fa17a54a72bb6d38916d74b14cad5d7e4a Mon Sep 17 00:00:00 2001
From: Sam Saffron <sam.saffron@gmail.com>
Date: Wed, 2 Aug 2023 13:47:56 +1000
Subject: [PATCH] tweaks to llama2 support in chat, fix message so it conforms
 to recommendation by facebook (add <s> )

---
 assets/javascripts/initializers/ai-bot-replies.js |  2 +-
 lib/modules/ai_bot/open_llm_bot.rb                | 10 +++++++---
 lib/modules/summarization/models/llama2.rb        |  6 +++---
 3 files changed, 11 insertions(+), 7 deletions(-)

diff --git a/assets/javascripts/initializers/ai-bot-replies.js b/assets/javascripts/initializers/ai-bot-replies.js
index e6776c14..8e767ff6 100644
--- a/assets/javascripts/initializers/ai-bot-replies.js
+++ b/assets/javascripts/initializers/ai-bot-replies.js
@@ -8,7 +8,7 @@ import { registerWidgetShim } from "discourse/widgets/render-glimmer";
 import { hbs } from "ember-cli-htmlbars";
 
 function isGPTBot(user) {
-  return user && [-110, -111, -112].includes(user.id);
+  return user && [-110, -111, -112, -113].includes(user.id);
 }
 
 function attachHeaderIcon(api) {
diff --git a/lib/modules/ai_bot/open_llm_bot.rb b/lib/modules/ai_bot/open_llm_bot.rb
index 83ea7f02..fc0a2f82 100644
--- a/lib/modules/ai_bot/open_llm_bot.rb
+++ b/lib/modules/ai_bot/open_llm_bot.rb
@@ -15,10 +15,14 @@ module DiscourseAi
 
         # start with system
         result = +""
-        result << "[INST] <<SYS>>\n #{messages.shift[:content]} <</SYS>>\n\n #{messages.shift[:content]} [/INST]"
+        result << "<s>[INST] <<SYS>>\n #{messages.shift[:content]} <</SYS>>\n\n #{messages.shift[:content]} [/INST]"
 
         messages.each do |message|
-          result << "\n\n[INST]#{message[:bot] ? "" : message[:username] + ":"} #{message[:content]} [/INST]"
+          if message[:bot]
+            result << message[:content]
+          else
+            result << "</s><s>[INST]#{message[:bot] ? "" : message[:username] + ":"} #{message[:content]} [/INST]"
+          end
         end
 
         result
@@ -65,7 +69,7 @@ module DiscourseAi
           prompt,
           model_for,
           temperature: 0.4,
-          max_tokens: 200,
+          max_tokens: 600,
           &blk
         )
       end
diff --git a/lib/modules/summarization/models/llama2.rb b/lib/modules/summarization/models/llama2.rb
index 1dc96088..76ca63b3 100644
--- a/lib/modules/summarization/models/llama2.rb
+++ b/lib/modules/summarization/models/llama2.rb
@@ -20,9 +20,9 @@ module DiscourseAi
           )
         end
 
-        def concatenate_summaries(summaries, &on_partial_blk)
-          prompt = <<~TEXT
-            [INST] <<SYS>>
+        def concatenate_summaries(summaries)
+          completion(<<~TEXT)
+            <s>[INST] <<SYS>>
             You are a helpful bot
             <</SYS>>