nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 23 Packages Projects Releases Wiki Activity

chat : reserve memory in compute_diffs and improve naming (#17729)

Browse Source
This commit is contained in:
Georgi Gerganov
2025-12-03 17:22:10 +02:00
committed by GitHub
parent e7c2cf1356
commit 190c4838bd
2 changed files with 25 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/chat.cpp
+24 -16
View File
@@ -85,29 +85,36 @@ json common_chat_msg::to_json_oaicompat() const
return message; return message;
} }
std::vector<common_chat_msg_diff> common_chat_msg_diff::compute_diffs(const common_chat_msg & previous_msg, const common_chat_msg & new_msg) { std::vector<common_chat_msg_diff> common_chat_msg_diff::compute_diffs(const common_chat_msg & msg_prv, const common_chat_msg & msg_new) {
std::vector<common_chat_msg_diff> diffs; std::vector<common_chat_msg_diff> diffs;
if (previous_msg.reasoning_content != new_msg.reasoning_content) { if (msg_new.tool_calls.size() > msg_prv.tool_calls.size()) {
auto & diff = diffs.emplace_back(); diffs.reserve(msg_new.tool_calls.size() - msg_prv.tool_calls.size() + 3);
diff.reasoning_content_delta = string_diff(previous_msg.reasoning_content, new_msg.reasoning_content); } else {
} diffs.reserve(3);
if (previous_msg.content != new_msg.content) {
auto & diff = diffs.emplace_back();
diff.content_delta = string_diff(previous_msg.content, new_msg.content);
} }
if (new_msg.tool_calls.size() < previous_msg.tool_calls.size()) { // TODO: these can become expensive for long messages - how to optimize?
if (msg_prv.reasoning_content != msg_new.reasoning_content) {
auto & diff = diffs.emplace_back();
diff.reasoning_content_delta = string_diff(msg_prv.reasoning_content, msg_new.reasoning_content);
}
if (msg_prv.content != msg_new.content) {
auto & diff = diffs.emplace_back();
diff.content_delta = string_diff(msg_prv.content, msg_new.content);
}
if (msg_new.tool_calls.size() < msg_prv.tool_calls.size()) {
throw std::runtime_error("Invalid diff: now finding less tool calls!"); throw std::runtime_error("Invalid diff: now finding less tool calls!");
} }
if (!previous_msg.tool_calls.empty()) { if (!msg_prv.tool_calls.empty()) {
auto idx = previous_msg.tool_calls.size() - 1; const auto idx = msg_prv.tool_calls.size() - 1;
const auto & pref = previous_msg.tool_calls[idx]; const auto & pref = msg_prv.tool_calls[idx];
const auto & newf = new_msg.tool_calls[idx]; const auto & newf = msg_new.tool_calls[idx];
if (pref.name != newf.name) { if (pref.name != newf.name) {
throw std::runtime_error("Invalid diff: tool call mismatch!"); throw std::runtime_error("Invalid diff: tool call mismatch!");
} }
auto args_diff = string_diff(pref.arguments, newf.arguments); const auto args_diff = string_diff(pref.arguments, newf.arguments);
if (!args_diff.empty() || pref.id != newf.id) { if (!args_diff.empty() || pref.id != newf.id) {
auto & diff = diffs.emplace_back(); auto & diff = diffs.emplace_back();
diff.tool_call_index = idx; diff.tool_call_index = idx;
@@ -118,11 +125,12 @@ std::vector<common_chat_msg_diff> common_chat_msg_diff::compute_diffs(const comm
diff.tool_call_delta.arguments = args_diff; diff.tool_call_delta.arguments = args_diff;
} }
} }
for (size_t idx = previous_msg.tool_calls.size(); idx < new_msg.tool_calls.size(); ++idx) { for (size_t idx = msg_prv.tool_calls.size(); idx < msg_new.tool_calls.size(); ++idx) {
auto & diff = diffs.emplace_back(); auto & diff = diffs.emplace_back();
diff.tool_call_index = idx; diff.tool_call_index = idx;
diff.tool_call_delta = new_msg.tool_calls[idx]; diff.tool_call_delta = msg_new.tool_calls[idx];
} }
return diffs; return diffs;
} }
common/chat.h
+1 -1
View File
@@ -77,7 +77,7 @@ struct common_chat_msg_diff {
size_t tool_call_index = std::string::npos; size_t tool_call_index = std::string::npos;
common_chat_tool_call tool_call_delta; common_chat_tool_call tool_call_delta;
static std::vector<common_chat_msg_diff> compute_diffs(const common_chat_msg & previous_msg, const common_chat_msg & new_msg); static std::vector<common_chat_msg_diff> compute_diffs(const common_chat_msg & msg_prv, const common_chat_msg & msg_new);
bool operator==(const common_chat_msg_diff & other) const { bool operator==(const common_chat_msg_diff & other) const {
return content_delta == other.content_delta return content_delta == other.content_delta