github/gpt4all
1
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all.git synced 2025-11-03 23:47:16 +00:00
Code Issues Packages Projects Releases Wiki Activity

server: improve correctness of request parsing and responses (#2929)

Browse Source 
Signed-off-by: Jared Van Bortel <jared@nomic.ai>
This commit is contained in:
Jared Van Bortel
2024-09-09 10:48:57 -04:00
committed by GitHub
parent 1aae4ffe0a
commit 39005288c5
22 changed files with 790 additions and 328 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
gpt4all-chat/src/chatllm.cpp
View File

@@ -626,16 +626,16 @@ void ChatLLM::regenerateResponse()
m_ctx.tokens.erase(m_ctx.tokens.end() - m_promptResponseTokens, m_ctx.tokens.end());
m_promptResponseTokens = 0;
m_promptTokens = 0;
m_response = std::string();
emit responseChanged(QString::fromStdString(m_response));
m_response = m_trimmedResponse = std::string();
emit responseChanged(QString::fromStdString(m_trimmedResponse));
}
void ChatLLM::resetResponse()
{
m_promptTokens = 0;
m_promptResponseTokens = 0;
m_response = std::string();
emit responseChanged(QString::fromStdString(m_response));
m_response = m_trimmedResponse = std::string();
emit responseChanged(QString::fromStdString(m_trimmedResponse));
}
void ChatLLM::resetContext()
@@ -645,9 +645,12 @@ void ChatLLM::resetContext()
m_ctx = LLModel::PromptContext();
}
QString ChatLLM::response() const
QString ChatLLM::response(bool trim) const
{
return QString::fromStdString(remove_leading_whitespace(m_response));
std::string resp = m_response;
if (trim)
resp = remove_leading_whitespace(resp);
return QString::fromStdString(resp);
}
ModelInfo ChatLLM::modelInfo() const
@@ -705,7 +708,8 @@ bool ChatLLM::handleResponse(int32_t token, const std::string &response)
// check for error
if (token < 0) {
m_response.append(response);
emit responseChanged(QString::fromStdString(remove_leading_whitespace(m_response)));
m_trimmedResponse = remove_leading_whitespace(m_response);
emit responseChanged(QString::fromStdString(m_trimmedResponse));
return false;
}
@@ -715,7 +719,8 @@ bool ChatLLM::handleResponse(int32_t token, const std::string &response)
m_timer->inc();
Q_ASSERT(!response.empty());
m_response.append(response);
emit responseChanged(QString::fromStdString(remove_leading_whitespace(m_response)));
m_trimmedResponse = remove_leading_whitespace(m_response);
emit responseChanged(QString::fromStdString(m_trimmedResponse));
return !m_stopGenerating;
}
@@ -741,7 +746,7 @@ bool ChatLLM::prompt(const QList<QString> &collectionList, const QString &prompt
bool ChatLLM::promptInternal(const QList<QString> &collectionList, const QString &prompt, const QString &promptTemplate,
int32_t n_predict, int32_t top_k, float top_p, float min_p, float temp, int32_t n_batch, float repeat_penalty,
int32_t repeat_penalty_tokens)
int32_t repeat_penalty_tokens, std::optional<QString> fakeReply)
{
if (!isModelLoaded())
return false;
@@ -751,7 +756,7 @@ bool ChatLLM::promptInternal(const QList<QString> &collectionList, const QString
QList<ResultInfo> databaseResults;
const int retrievalSize = MySettings::globalInstance()->localDocsRetrievalSize();
if (!collectionList.isEmpty()) {
if (!fakeReply && !collectionList.isEmpty()) {
emit requestRetrieveFromDB(collectionList, prompt, retrievalSize, &databaseResults); // blocks
emit databaseResultsChanged(databaseResults);
}
@@ -797,7 +802,8 @@ bool ChatLLM::promptInternal(const QList<QString> &collectionList, const QString
m_ctx.n_predict = old_n_predict; // now we are ready for a response
}
m_llModelInfo.model->prompt(prompt.toStdString(), promptTemplate.toStdString(), promptFunc, responseFunc,
/*allowContextShift*/ true, m_ctx);
/*allowContextShift*/ true, m_ctx, false,
fakeReply.transform(std::mem_fn(&QString::toStdString)));
#if defined(DEBUG)
printf("\n");
fflush(stdout);
@@ -805,9 +811,9 @@ bool ChatLLM::promptInternal(const QList<QString> &collectionList, const QString
m_timer->stop();
qint64 elapsed = totalTime.elapsed();
std::string trimmed = trim_whitespace(m_response);
if (trimmed != m_response) {
m_response = trimmed;
emit responseChanged(QString::fromStdString(m_response));
if (trimmed != m_trimmedResponse) {
m_trimmedResponse = trimmed;
emit responseChanged(QString::fromStdString(m_trimmedResponse));
}
SuggestionMode mode = MySettings::globalInstance()->suggestionMode();
@@ -1078,6 +1084,7 @@ bool ChatLLM::deserialize(QDataStream &stream, int version, bool deserializeKV,
QString response;
stream >> response;
m_response = response.toStdString();
m_trimmedResponse = trim_whitespace(m_response);
QString nameResponse;
stream >> nameResponse;
m_nameResponse = nameResponse.toStdString();
@@ -1306,10 +1313,9 @@ void ChatLLM::processRestoreStateFromText()
auto &response = *it++;
Q_ASSERT(response.first != "Prompt: ");
auto responseText = response.second.toStdString();
m_llModelInfo.model->prompt(prompt.second.toStdString(), promptTemplate.toStdString(), promptFunc, nullptr,
/*allowContextShift*/ true, m_ctx, false, &responseText);
/*allowContextShift*/ true, m_ctx, false, response.second.toUtf8().constData());
}
if (!m_stopGenerating) {