From 02e13737f39931ece163e2cb270132de2c892d06 Mon Sep 17 00:00:00 2001
From: Adam Treat <treat.adam@gmail.com>
Date: Sun, 9 Apr 2023 01:11:52 -0400
Subject: [PATCH] Don't repeat the prompt in the response.

---
 gptj.cpp | 7 +++++--
 main.qml | 7 ++++---
 2 files changed, 9 insertions(+), 5 deletions(-)

diff --git a/gptj.cpp b/gptj.cpp
index 0858e944..612b1a34 100644
--- a/gptj.cpp
+++ b/gptj.cpp
@@ -700,6 +700,7 @@ void GPTJ::prompt(const std::string &prompt, std::function<bool(const std::strin
     n_predict = std::min(n_predict, d_ptr->model.hparams.n_ctx - (int) embd_inp.size());
 
     std::vector<gpt_vocab::id> embd;
+    std::vector<gpt_vocab::id> resp;
 
     // determine the required inference memory per token:
     size_t mem_per_token = 0;
@@ -720,6 +721,7 @@ void GPTJ::prompt(const std::string &prompt, std::function<bool(const std::strin
 
         n_past += embd.size();
         embd.clear();
+        resp.clear();
 
         if (i >= embd_inp.size()) {
             // sample next token
@@ -738,6 +740,7 @@ void GPTJ::prompt(const std::string &prompt, std::function<bool(const std::strin
 
             // add it to the context
             embd.push_back(id);
+            resp.push_back(id);
         } else {
             // if here, it means we are still processing the input prompt
             for (int k = i; k < embd_inp.size(); k++) {
@@ -750,7 +753,7 @@ void GPTJ::prompt(const std::string &prompt, std::function<bool(const std::strin
         }
 
         // display text
-        for (auto id : embd) {
+        for (auto id : resp) {
             if (!response(d_ptr->vocab.id_to_token[id]))
                 goto stop_generating;
         }
@@ -762,7 +765,7 @@ void GPTJ::prompt(const std::string &prompt, std::function<bool(const std::strin
     }
 
 stop_generating:
-#if 1
+#if 0
     // report timing
     {
         const int64_t t_main_end_us = ggml_time_us();
diff --git a/main.qml b/main.qml
index 2e9038ca..0411306e 100644
--- a/main.qml
+++ b/main.qml
@@ -80,7 +80,7 @@ Window {
                     model: chatModel
                     delegate: TextArea {
                         text: currentResponse ? LLM.response : value
-                        width: parent.width
+                        width: listView.width
                         color: "#d1d5db"
                         wrapMode: Text.WordWrap
                         focus: false
@@ -204,11 +204,12 @@ Window {
                     listElement.currentResponse = false
                     listElement.value = LLM.response
                 }
+                var prompt = textInput.text + "\n"
                 chatModel.append({"name": qsTr("Prompt: "), "currentResponse": false, "value": textInput.text})
-                chatModel.append({"name": qsTr("Response: "), "currentResponse": true, "value": "", "prompt": textInput.text})
+                chatModel.append({"name": qsTr("Response: "), "currentResponse": true, "value": "", "prompt": prompt})
 
                 LLM.resetResponse()
-                LLM.prompt(textInput.text)
+                LLM.prompt(prompt)
                 textInput.text = ""
             }