vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

more consistent naming

Browse source
This commit is contained in:
ngxson 2024-06-22 22:57:16 +02:00
parent 317452730d
commit 962be6a834
4 changed files with 6 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

6
common/common.cpp
View file

@ -2980,7 +2980,7 @@ bool llama_chat_verify_template(const std::string & tmpl) {
return res >= 0; return res >= 0;
} }
std::string llama_chat_format(const struct llama_model * model, std::string llama_chat_apply_template(const struct llama_model * model,
const std::string & tmpl, const std::string & tmpl,
const std::vector<llama_chat_msg> & msgs, const std::vector<llama_chat_msg> & msgs,
bool add_ass) { bool add_ass) {
@ -3010,10 +3010,10 @@ std::string llama_chat_format_single(const struct llama_model * model,
const std::vector<llama_chat_msg> & past_msg, const std::vector<llama_chat_msg> & past_msg,
const llama_chat_msg & new_msg, const llama_chat_msg & new_msg,
bool add_ass) { bool add_ass) {
auto fmt_past_msg = llama_chat_format(model, tmpl, past_msg, false); auto fmt_past_msg = llama_chat_apply_template(model, tmpl, past_msg, false);
std::vector<llama_chat_msg> chat_new(past_msg); std::vector<llama_chat_msg> chat_new(past_msg);
chat_new.push_back(new_msg); chat_new.push_back(new_msg);
auto fmt_new_msg = llama_chat_format(model, tmpl, chat_new, add_ass); auto fmt_new_msg = llama_chat_apply_template(model, tmpl, chat_new, add_ass);
auto formatted = fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size()); auto formatted = fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size());
return formatted; return formatted;
} }

2
common/common.h
View file

@ -370,7 +370,7 @@ struct llama_chat_msg {
bool llama_chat_verify_template(const std::string & tmpl); bool llama_chat_verify_template(const std::string & tmpl);
// CPP wrapper for llama_chat_apply_template // CPP wrapper for llama_chat_apply_template
std::string llama_chat_format(const struct llama_model * model, std::string llama_chat_apply_template(const struct llama_model * model,
const std::string & tmpl, const std::string & tmpl,
const std::vector<llama_chat_msg> & chat, const std::vector<llama_chat_msg> & chat,
bool add_ass); bool add_ass);

4
examples/main/main.cpp
View file

@ -876,10 +876,8 @@ int main(int argc, char ** argv) {
? chat_add_and_format("user", buffer) ? chat_add_and_format("user", buffer)
: buffer; : buffer;
// TODO: one inconvenient of current chat template implementation is that we can't distinguish between user input and special tokens (prefix/postfix) // TODO: one inconvenient of current chat template implementation is that we can't distinguish between user input and special tokens (prefix/postfix)
bool accept_special_content = params.conversation;
const auto line_pfx = ::llama_tokenize(ctx, params.input_prefix, false, true); const auto line_pfx = ::llama_tokenize(ctx, params.input_prefix, false, true);
const auto line_inp = ::llama_tokenize(ctx, user_inp, false, accept_special_content); const auto line_inp = ::llama_tokenize(ctx, user_inp, false, params.conversation);
const auto line_sfx = ::llama_tokenize(ctx, params.input_suffix, false, true); const auto line_sfx = ::llama_tokenize(ctx, params.input_suffix, false, true);
LOG("input tokens: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, line_inp).c_str()); LOG("input tokens: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, line_inp).c_str());

2
examples/server/utils.hpp
View file

@ -127,7 +127,7 @@ inline std::string format_chat(const struct llama_model * model, const std::stri
chat.push_back({role, content}); chat.push_back({role, content});
} }
auto formatted_chat = llama_chat_format(model, tmpl, chat, true); auto formatted_chat = llama_chat_apply_template(model, tmpl, chat, true);
LOG_VERBOSE("formatted_chat", {{"text", formatted_chat.c_str()}}); LOG_VERBOSE("formatted_chat", {{"text", formatted_chat.c_str()}});
return formatted_chat; return formatted_chat;
} }