Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

#

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Jump to bottom

backend: fix a crash on inputs greater than n_ctx #2498

Merged
merged 1 commit into from
Jul 1, 2024
Merged

backend: fix a crash on inputs greater than n_ctx #2498

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion gpt4all-backend/llmodel.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -248,7 +248,7 @@ class LLModel {
return true;
}

void decodePrompt(std::function<bool(int32_t)> promptCallback,
bool decodePrompt(std::function<bool(int32_t)> promptCallback,
std::function<bool(int32_t, const std::string&)> responseCallback,
std::function<bool(bool)> recalculateCallback,
PromptContext &promptCtx,
Expand Down
17 changes: 11 additions & 6 deletions gpt4all-backend/llmodel_shared.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -135,14 +135,16 @@ void LLModel::prompt(const std::string &prompt,
promptCtx.n_past = old_n_past; // restore n_past so decodePrompt can increment it

// decode the user prompt
decodePrompt(promptCallback, responseCallback, recalculateCallback, promptCtx, embd_inp);
if (!decodePrompt(promptCallback, responseCallback, recalculateCallback, promptCtx, embd_inp))
return; // error

// decode the assistant's reply, either generated or spoofed
if (fakeReply == nullptr) {
generateResponse(responseCallback, recalculateCallback, promptCtx);
} else {
embd_inp = tokenize(promptCtx, *fakeReply, false);
decodePrompt(promptCallback, responseCallback, recalculateCallback, promptCtx, embd_inp);
if (!decodePrompt(promptCallback, responseCallback, recalculateCallback, promptCtx, embd_inp))
return; // error
}

// decode the rest of the prompt template
Expand All @@ -160,7 +162,8 @@ void LLModel::prompt(const std::string &prompt,
}
}

void LLModel::decodePrompt(std::function<bool(int32_t)> promptCallback,
// returns false on error
bool LLModel::decodePrompt(std::function<bool(int32_t)> promptCallback,
std::function<bool(int32_t, const std::string&)> responseCallback,
std::function<bool(bool)> recalculateCallback,
PromptContext &promptCtx,
Expand All @@ -172,7 +175,7 @@ void LLModel::decodePrompt(std::function<bool(int32_t)> promptCallback,
responseCallback(-1, "ERROR: The prompt size exceeds the context window size and cannot be processed.");
std::cerr << implementation().modelType() << " ERROR: The prompt is " << embd_inp.size() <<
" tokens and the context window is " << promptCtx.n_ctx << "!\n";
return;
return false;
}

promptCtx.n_predict = std::min(promptCtx.n_predict, promptCtx.n_ctx - (int) embd_inp.size());
Expand All @@ -193,7 +196,7 @@ void LLModel::decodePrompt(std::function<bool(int32_t)> promptCallback,

if (!evalTokens(promptCtx, batch)) {
std::cerr << implementation().modelType() << " ERROR: Failed to process prompt\n";
return;
return false;
}

size_t tokens = batch_end - i;
Expand All @@ -203,10 +206,12 @@ void LLModel::decodePrompt(std::function<bool(int32_t)> promptCallback,
promptCtx.tokens.push_back(batch.at(t));
promptCtx.n_past += 1;
if (!promptCallback(batch.at(t)))
return;
return false;
}
i = batch_end;
}

return true;
}

void LLModel::generateResponse(std::function<bool(int32_t, const std::string&)> responseCallback,
Expand Down