AI/gpt4all
1
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all.git synced 2024-10-01 01:06:10 -04:00
Code Issues Packages Projects Releases Wiki Activity

chat: fix ChatGPT after #1970 (#2051)

Browse Source 
Signed-off-by: Jared Van Bortel <jared@nomic.ai>
This commit is contained in:
Jared Van Bortel 2024-03-06 14:02:18 -05:00 committed by GitHub
parent 2a91ffd73f
commit 402f515a5d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 31 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
gpt4all-chat/chatgpt.cpp
View File

@ -86,13 +86,33 @@ void ChatGPT::prompt(const std::string &prompt,
Q_UNUSED(promptCallback);
Q_UNUSED(recalculateCallback);
Q_UNUSED(special);
Q_UNUSED(fakeReply); // FIXME(cebtenzzre): I broke ChatGPT
if (!isModelLoaded()) {
std::cerr << "ChatGPT ERROR: prompt won't work with an unloaded model!\n";
return;
}
if (!promptCtx.n_past) { m_queuedPrompts.clear(); }
Q_ASSERT(promptCtx.n_past <= m_context.size());
m_context.resize(promptCtx.n_past);
// FIXME(cebtenzzre): We're assuming people don't try to use %2 with ChatGPT. What would that even mean?
m_queuedPrompts << QString::fromStdString(promptTemplate).arg(QString::fromStdString(prompt));
if (!promptCtx.n_predict && !fakeReply) {
return; // response explicitly suppressed, queue prompt for later
}
QString formattedPrompt = m_queuedPrompts.join("");
m_queuedPrompts.clear();
if (fakeReply) {
promptCtx.n_past += 1;
m_context.append(formattedPrompt);
m_context.append(QString::fromStdString(*fakeReply));
return;
}
// FIXME: We don't set the max_tokens on purpose because in order to do so safely without encountering
// an error we need to be able to count the tokens in our prompt. The only way to do this is to use
// the OpenAI tiktokken library or to implement our own tokenization function that matches precisely
@ -104,8 +124,9 @@ void ChatGPT::prompt(const std::string &prompt,
root.insert("temperature", promptCtx.temp);
root.insert("top_p", promptCtx.top_p);
// conversation history
QJsonArray messages;
for (int i = 0; i < m_context.count() && i < promptCtx.n_past; ++i) {
for (int i = 0; i < m_context.count(); ++i) {
QJsonObject message;
message.insert("role", i % 2 == 0 ? "assistant" : "user");
message.insert("content", m_context.at(i));
@ -114,7 +135,7 @@ void ChatGPT::prompt(const std::string &prompt,
QJsonObject promptObject;
promptObject.insert("role", "user");
promptObject.insert("content", QString::fromStdString(promptTemplate).arg(QString::fromStdString(prompt)));
promptObject.insert("content", formattedPrompt);
messages.append(promptObject);
root.insert("messages", messages);
@ -138,7 +159,7 @@ void ChatGPT::prompt(const std::string &prompt,
workerThread.wait();
promptCtx.n_past += 1;
m_context.append(QString::fromStdString(prompt));
m_context.append(formattedPrompt);
m_context.append(worker.currentResponse());
m_responseCallback = nullptr;

8
gpt4all-chat/chatgpt.h
View File

@ -3,11 +3,14 @@
#include <stdexcept>
#include <QObject>
#include <QNetworkAccessManager>
#include <QNetworkReply>
#include <QNetworkRequest>
#include <QNetworkAccessManager>
#include <QObject>
#include <QString>
#include <QStringList>
#include <QThread>
#include "../gpt4all-backend/llmodel.h"
class ChatGPT;
@ -126,6 +129,7 @@ private:
QString m_modelName;
QString m_apiKey;
QList<QString> m_context;
QStringList m_queuedPrompts;
};
#endif // CHATGPT_H