Skip to content
New issue

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

Sign up for GitHub

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

Already on GitHub? Sign in to your account

Support for BREAK pseudo-token, to break clip chunks. #285

Open
wants to merge 3 commits into
base: master
Choose a base branch
from
Open

Support for BREAK pseudo-token, to break clip chunks. #285

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
de6ae28
Support for BREAK pseudo-token, to break clip chunks.
ursg Jun 11, 2024
62b3034
Fix a nonsensical comment
ursg Jun 11, 2024
56b94a4
Bring PR up-to-date with master, to fix CI runs
ursg Jun 18, 2024
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
81 changes: 54 additions & 27 deletions clip.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -413,7 +413,7 @@ std::vector<std::pair<std::string, float>> parse_prompt_attention(const std::str
float round_bracket_multiplier = 1.1f;
float square_bracket_multiplier = 1 / 1.1f;

std::regex re_attention(R"(\\\(|\\\)|\\\[|\\\]|\\\\|\\|\(|\[|:([+-]?[.\d]+)\)|\)|\]|[^\\()\[\]:]+|:)");
std::regex re_attention(R"(\\\(|\\\)|\\\[|\\\]|\\\\|\\|\(|\[|:([+-]?[.\d]+)\)|\)|\]|\s*\bBREAK\b\s*|[^\\()\[\]: B]+|:|\s*\bB)");
std::regex re_break(R"(\s*\bBREAK\b\s*)");

auto multiply_range = [&](int start_position, float multiplier) {
Expand All @@ -422,7 +422,7 @@ std::vector<std::pair<std::string, float>> parse_prompt_attention(const std::str
}
};

std::smatch m;
std::smatch m,m2;
std::string remaining_text = text;

while (std::regex_search(remaining_text, m, re_attention)) {
Expand All @@ -446,6 +446,8 @@ std::vector<std::pair<std::string, float>> parse_prompt_attention(const std::str
square_brackets.pop_back();
} else if (text == "\\(") {
res.push_back({text.substr(1), 1.0f});
} else if (std::regex_match(text, m2, re_break)) {
res.push_back({"BREAK", -1.0f});
} else {
res.push_back({text, 1.0f});
}
Expand Down Expand Up @@ -1240,32 +1242,45 @@ struct FrozenCLIPEmbedderWithCustomWords : public GGMLModule {
const std::string& curr_text = item.first;
float curr_weight = item.second;
// printf(" %s: %f \n", curr_text.c_str(), curr_weight);
std::vector<int> curr_tokens = tokenizer.encode(curr_text, on_new_token_cb);

int32_t clean_index = 0;
for (uint32_t i = 0; i < curr_tokens.size(); i++) {
int token_id = curr_tokens[i];
if (token_id == image_token)
class_token_index.push_back(clean_index - 1);
else {
clean_input_ids.push_back(token_id);
clean_index++;
}
}
// GGML_ASSERT(class_token_index.size() == 1); // PhotoMaker currently does not support multiple
// trigger words in a single prompt.
if (class_token_index.size() == 1) {
// Expand the class word token and corresponding mask
int class_token = clean_input_ids[class_token_index[0]];
class_idx = tokens_acc + class_token_index[0];
std::vector<int> clean_input_ids_tmp;
for (uint32_t i = 0; i < class_token_index[0]; i++)
clean_input_ids_tmp.push_back(clean_input_ids[i]);
for (uint32_t i = 0; i < num_input_imgs; i++)
clean_input_ids_tmp.push_back(class_token);
for (uint32_t i = class_token_index[0] + 1; i < clean_input_ids.size(); i++)
clean_input_ids_tmp.push_back(clean_input_ids[i]);
clean_input_ids.clear();
clean_input_ids = clean_input_ids_tmp;
if(curr_text == "BREAK" && curr_weight == -1.0f) {
// Pad token array up to chunk size at this point.
// TODO: This is a hardcoded chunk_len, like in stable-diffusion.cpp, make it a parameter for the future?
// Also, this is 75 instead of 77 to leave room for BOS and EOS tokens.
for(int j=0; j< 75 - (tokens_acc%75); j++) {
clean_input_ids.push_back(EOS_TOKEN_ID);
clean_index++;
}
} else {

// Regular token, process normally.
std::vector<int> curr_tokens = tokenizer.encode(curr_text, on_new_token_cb);
for (uint32_t i = 0; i < curr_tokens.size(); i++) {
int token_id = curr_tokens[i];
if (token_id == image_token)
class_token_index.push_back(clean_index - 1);
else {
clean_input_ids.push_back(token_id);
clean_index++;
}
}
// GGML_ASSERT(class_token_index.size() == 1); // PhotoMaker currently does not support multiple
// trigger words in a single prompt.
if (class_token_index.size() == 1) {
// Expand the class word token and corresponding mask
int class_token = clean_input_ids[class_token_index[0]];
class_idx = tokens_acc + class_token_index[0];
std::vector<int> clean_input_ids_tmp;
for (uint32_t i = 0; i < class_token_index[0]; i++)
clean_input_ids_tmp.push_back(clean_input_ids[i]);
for (uint32_t i = 0; i < num_input_imgs; i++)
clean_input_ids_tmp.push_back(class_token);
for (uint32_t i = class_token_index[0] + 1; i < clean_input_ids.size(); i++)
clean_input_ids_tmp.push_back(clean_input_ids[i]);
clean_input_ids.clear();
clean_input_ids = clean_input_ids_tmp;
}
}
tokens_acc += clean_index;
tokens.insert(tokens.end(), clean_input_ids.begin(), clean_input_ids.end());
Expand Down Expand Up @@ -1341,6 +1356,18 @@ struct FrozenCLIPEmbedderWithCustomWords : public GGMLModule {
for (const auto& item : parsed_attention) {
const std::string& curr_text = item.first;
float curr_weight = item.second;

if(curr_text == "BREAK" && curr_weight == -1.0f) {
// Pad token array up to chunk size at this point.
// TODO: This is a hardcoded chunk_len, like in stable-diffusion.cpp, make it a parameter for the future?
// Also, this is 75 instead of 77 to leave room for BOS and EOS tokens.
LOG_DEBUG("BREAK token encountered, padding current chunk by %i tokens.", 75 - (tokens.size()%75));
while(tokens.size() % 75) {
tokens.push_back(EOS_TOKEN_ID);
weights.push_back(1);
}
continue;
}
std::vector<int> curr_tokens = tokenizer.encode(curr_text, on_new_token_cb);
tokens.insert(tokens.end(), curr_tokens.begin(), curr_tokens.end());
weights.insert(weights.end(), curr_tokens.size(), curr_weight);
Expand Down
Loading