mirror of
https://github.com/aaronpo97/the-biergarten-app.git
synced 2026-06-01 01:54:00 +00:00
Code format updates
This commit is contained in:
Aaron Po
@@ -16,135 +16,134 @@
|
||||
#include "data_generation/llama_generator_helpers.h"
|
||||
|
||||
static std::string ExtractFinalJsonPayload(std::string raw_response) {
|
||||
auto trim = [](const std::string_view text) -> std::string_view {
|
||||
const std::size_t first = text.find_first_not_of(" \t\n\r");
|
||||
if (first == std::string_view::npos) {
|
||||
return {};
|
||||
}
|
||||
auto trim = [](const std::string_view text) -> std::string_view {
|
||||
const std::size_t first = text.find_first_not_of(" \t\n\r");
|
||||
if (first == std::string_view::npos) {
|
||||
return {};
|
||||
}
|
||||
|
||||
const std::size_t last = text.find_last_not_of(" \t\n\r");
|
||||
return text.substr(first, last - first + 1);
|
||||
};
|
||||
const std::size_t last = text.find_last_not_of(" \t\n\r");
|
||||
return text.substr(first, last - first + 1);
|
||||
};
|
||||
|
||||
static constexpr std::array<std::string_view, 6> separator_tokens = {
|
||||
"<|think|>", "<think|>", "<|turn|>",
|
||||
"<turn|>", "<channel|>", "<|channel|>"};
|
||||
static constexpr std::array<std::string_view, 6> separator_tokens = {
|
||||
"<|think|>", "<think|>", "<|turn|>",
|
||||
"<turn|>", "<channel|>", "<|channel|>"};
|
||||
|
||||
std::size_t separator_pos = std::string::npos;
|
||||
std::size_t separator_length = 0;
|
||||
for (const std::string_view token : separator_tokens) {
|
||||
const std::size_t candidate_pos = raw_response.rfind(token);
|
||||
if (candidate_pos != std::string::npos &&
|
||||
(separator_pos == std::string::npos ||
|
||||
candidate_pos > separator_pos)) {
|
||||
separator_pos = candidate_pos;
|
||||
separator_length = token.size();
|
||||
}
|
||||
}
|
||||
std::size_t separator_pos = std::string::npos;
|
||||
std::size_t separator_length = 0;
|
||||
for (const std::string_view token : separator_tokens) {
|
||||
const std::size_t candidate_pos = raw_response.rfind(token);
|
||||
if (candidate_pos != std::string::npos &&
|
||||
(separator_pos == std::string::npos || candidate_pos > separator_pos)) {
|
||||
separator_pos = candidate_pos;
|
||||
separator_length = token.size();
|
||||
}
|
||||
}
|
||||
|
||||
if (separator_pos != std::string::npos) {
|
||||
raw_response.erase(0, separator_pos + separator_length);
|
||||
}
|
||||
if (separator_pos != std::string::npos) {
|
||||
raw_response.erase(0, separator_pos + separator_length);
|
||||
}
|
||||
|
||||
const std::string_view trimmed = trim(raw_response);
|
||||
const std::string json_candidate =
|
||||
ExtractLastJsonObjectPublic(std::string(trimmed));
|
||||
const std::string_view trimmed = trim(raw_response);
|
||||
const std::string json_candidate =
|
||||
ExtractLastJsonObjectPublic(std::string(trimmed));
|
||||
|
||||
if (!json_candidate.empty()) {
|
||||
return ExtractLastJsonObjectPublic(std::string(trimmed));
|
||||
}
|
||||
if (!json_candidate.empty()) {
|
||||
return ExtractLastJsonObjectPublic(std::string(trimmed));
|
||||
}
|
||||
|
||||
return std::string(trimmed);
|
||||
return std::string(trimmed);
|
||||
}
|
||||
|
||||
BreweryResult LlamaGenerator::GenerateBrewery(
|
||||
const Location& location, const std::string& region_context) {
|
||||
/**
|
||||
* Preprocess and truncate region context to manageable size
|
||||
*/
|
||||
const std::string safe_region_context =
|
||||
PrepareRegionContextPublic(region_context);
|
||||
/**
|
||||
* Preprocess and truncate region context to manageable size
|
||||
*/
|
||||
const std::string safe_region_context =
|
||||
PrepareRegionContextPublic(region_context);
|
||||
|
||||
const std::string country_suffix =
|
||||
location.country.empty() ? std::string{}
|
||||
: std::format(", {}", location.country);
|
||||
const std::string region_suffix =
|
||||
safe_region_context.empty()
|
||||
? "."
|
||||
: std::format(". Regional context: {}", safe_region_context);
|
||||
const std::string country_suffix =
|
||||
location.country.empty() ? std::string{}
|
||||
: std::format(", {}", location.country);
|
||||
const std::string region_suffix =
|
||||
safe_region_context.empty()
|
||||
? "."
|
||||
: std::format(". Regional context: {}", safe_region_context);
|
||||
|
||||
/**
|
||||
* Load brewery system prompt from file
|
||||
* Falls back to minimal inline prompt if file not found
|
||||
*/
|
||||
const std::string system_prompt =
|
||||
LoadBrewerySystemPrompt("prompts/system.md");
|
||||
/**
|
||||
* Load brewery system prompt from file
|
||||
* Falls back to minimal inline prompt if file not found
|
||||
*/
|
||||
const std::string system_prompt =
|
||||
LoadBrewerySystemPrompt("prompts/system.md");
|
||||
|
||||
/**
|
||||
* User prompt: provides geographic context to guide generation towards
|
||||
* culturally relevant and locally-inspired brewery attributes
|
||||
*/
|
||||
std::string prompt = std::format(
|
||||
"Write a brewery name and place-specific long description for a craft "
|
||||
"brewery in {}{}{}",
|
||||
location.city, country_suffix, region_suffix);
|
||||
/**
|
||||
* User prompt: provides geographic context to guide generation towards
|
||||
* culturally relevant and locally-inspired brewery attributes
|
||||
*/
|
||||
std::string prompt = std::format(
|
||||
"Write a brewery name and place-specific long description for a craft "
|
||||
"brewery in {}{}{}",
|
||||
location.city, country_suffix, region_suffix);
|
||||
|
||||
/**
|
||||
* Store location context for retry prompts (without repeating full context)
|
||||
*/
|
||||
const std::string retry_location =
|
||||
std::format("Location: {}{}", location.city, country_suffix);
|
||||
/**
|
||||
* Store location context for retry prompts (without repeating full context)
|
||||
*/
|
||||
const std::string retry_location =
|
||||
std::format("Location: {}{}", location.city, country_suffix);
|
||||
|
||||
/**
|
||||
* RETRY LOOP with validation and error correction
|
||||
* Attempts to generate valid brewery data up to 3 times, with feedback-based
|
||||
* refinement
|
||||
*/
|
||||
constexpr int max_attempts = 3;
|
||||
std::string raw;
|
||||
std::string last_error;
|
||||
/**
|
||||
* RETRY LOOP with validation and error correction
|
||||
* Attempts to generate valid brewery data up to 3 times, with feedback-based
|
||||
* refinement
|
||||
*/
|
||||
constexpr int max_attempts = 3;
|
||||
std::string raw;
|
||||
std::string last_error;
|
||||
|
||||
// Limit output length to keep it concise and focused
|
||||
for (int attempt = 0; attempt < max_attempts; ++attempt) {
|
||||
constexpr int max_tokens = 1052;
|
||||
// Generate brewery data from LLM
|
||||
raw = this->Infer(system_prompt, prompt, max_tokens);
|
||||
spdlog::debug("LlamaGenerator: raw output (attempt {}): {}", attempt + 1,
|
||||
raw);
|
||||
// Limit output length to keep it concise and focused
|
||||
for (int attempt = 0; attempt < max_attempts; ++attempt) {
|
||||
constexpr int max_tokens = 1052;
|
||||
// Generate brewery data from LLM
|
||||
raw = this->Infer(system_prompt, prompt, max_tokens);
|
||||
spdlog::debug("LlamaGenerator: raw output (attempt {}): {}", attempt + 1,
|
||||
raw);
|
||||
|
||||
// Validate output: parse JSON and check required fields
|
||||
// Validate output: parse JSON and check required fields
|
||||
|
||||
std::string name;
|
||||
std::string description;
|
||||
const std::string json_only = ExtractFinalJsonPayload(raw);
|
||||
const std::optional<std::string> validation_error =
|
||||
ValidateBreweryJsonPublic(json_only, name, description);
|
||||
if (!validation_error.has_value()) {
|
||||
// Success: return parsed brewery data
|
||||
return BreweryResult{.name = std::move(name),
|
||||
.description = std::move(description)};
|
||||
}
|
||||
std::string name;
|
||||
std::string description;
|
||||
const std::string json_only = ExtractFinalJsonPayload(raw);
|
||||
const std::optional<std::string> validation_error =
|
||||
ValidateBreweryJsonPublic(json_only, name, description);
|
||||
if (!validation_error.has_value()) {
|
||||
// Success: return parsed brewery data
|
||||
return BreweryResult{.name = std::move(name),
|
||||
.description = std::move(description)};
|
||||
}
|
||||
|
||||
// Validation failed: log error and prepare corrective feedback
|
||||
// Validation failed: log error and prepare corrective feedback
|
||||
|
||||
last_error = *validation_error;
|
||||
spdlog::warn("LlamaGenerator: malformed brewery JSON (attempt {}): {}",
|
||||
attempt + 1, *validation_error);
|
||||
last_error = *validation_error;
|
||||
spdlog::warn("LlamaGenerator: malformed brewery JSON (attempt {}): {}",
|
||||
attempt + 1, *validation_error);
|
||||
|
||||
// Update prompt with error details to guide LLM toward correct output.
|
||||
prompt = std::format(
|
||||
R"(Your previous response was invalid. Error: {}
|
||||
// Update prompt with error details to guide LLM toward correct output.
|
||||
prompt = std::format(
|
||||
R"(Your previous response was invalid. Error: {}
|
||||
Return ONLY valid JSON with exactly these keys: {{"name": "<brewery name>", "description": "<single-paragraph description>"}}.
|
||||
Do not include markdown, comments, extra keys, or literal placeholder values.
|
||||
|
||||
{})",
|
||||
*validation_error, retry_location);
|
||||
}
|
||||
*validation_error, retry_location);
|
||||
}
|
||||
|
||||
// All retry attempts exhausted: log failure and throw exception
|
||||
spdlog::error(
|
||||
"LlamaGenerator: malformed brewery response after {} attempts: "
|
||||
"{}",
|
||||
max_attempts, last_error.empty() ? raw : last_error);
|
||||
throw std::runtime_error("LlamaGenerator: malformed brewery response");
|
||||
// All retry attempts exhausted: log failure and throw exception
|
||||
spdlog::error(
|
||||
"LlamaGenerator: malformed brewery response after {} attempts: "
|
||||
"{}",
|
||||
max_attempts, last_error.empty() ? raw : last_error);
|
||||
throw std::runtime_error("LlamaGenerator: malformed brewery response");
|
||||
}
|
||||
|
||||
@@ -13,6 +13,6 @@
|
||||
#include "data_generation/llama_generator_helpers.h"
|
||||
|
||||
UserResult LlamaGenerator::GenerateUser(const std::string& locale) {
|
||||
return {.username = "test_user",
|
||||
.bio = "This is a test user profile from " + locale + "."};
|
||||
return {.username = "test_user",
|
||||
.bio = "This is a test user profile from " + locale + "."};
|
||||
}
|
||||
|
||||
@@ -24,14 +24,14 @@
|
||||
* String trimming: removes leading and trailing whitespace
|
||||
*/
|
||||
static std::string Trim(std::string_view value) {
|
||||
constexpr std::string_view whitespace = " \t\n\r\f\v";
|
||||
const std::size_t first_index = value.find_first_not_of(whitespace);
|
||||
if (first_index == std::string_view::npos) {
|
||||
return {};
|
||||
}
|
||||
constexpr std::string_view whitespace = " \t\n\r\f\v";
|
||||
const std::size_t first_index = value.find_first_not_of(whitespace);
|
||||
if (first_index == std::string_view::npos) {
|
||||
return {};
|
||||
}
|
||||
|
||||
const std::size_t last_index = value.find_last_not_of(whitespace);
|
||||
return std::string(value.substr(first_index, last_index - first_index + 1));
|
||||
const std::size_t last_index = value.find_last_not_of(whitespace);
|
||||
return std::string(value.substr(first_index, last_index - first_index + 1));
|
||||
}
|
||||
|
||||
/**
|
||||
@@ -39,26 +39,26 @@ static std::string Trim(std::string_view value) {
|
||||
* spaces
|
||||
*/
|
||||
static std::string CondenseWhitespace(std::string_view text) {
|
||||
std::string out;
|
||||
out.reserve(text.size());
|
||||
std::string out;
|
||||
out.reserve(text.size());
|
||||
|
||||
bool pending_space = false;
|
||||
for (const unsigned char chr : text) {
|
||||
if (std::isspace(chr) != 0) {
|
||||
if (!out.empty()) {
|
||||
pending_space = true;
|
||||
}
|
||||
continue;
|
||||
bool pending_space = false;
|
||||
for (const unsigned char chr : text) {
|
||||
if (std::isspace(chr) != 0) {
|
||||
if (!out.empty()) {
|
||||
pending_space = true;
|
||||
}
|
||||
continue;
|
||||
}
|
||||
|
||||
if (pending_space) {
|
||||
out.push_back(' ');
|
||||
pending_space = false;
|
||||
}
|
||||
out.push_back(static_cast<char>(chr));
|
||||
}
|
||||
if (pending_space) {
|
||||
out.push_back(' ');
|
||||
pending_space = false;
|
||||
}
|
||||
out.push_back(static_cast<char>(chr));
|
||||
}
|
||||
|
||||
return out;
|
||||
return out;
|
||||
}
|
||||
|
||||
/**
|
||||
@@ -67,286 +67,285 @@ static std::string CondenseWhitespace(std::string_view text) {
|
||||
*/
|
||||
static std::string PrepareRegionContext(std::string_view region_context,
|
||||
const size_t max_chars) {
|
||||
std::string normalized = CondenseWhitespace(region_context);
|
||||
if (normalized.size() <= max_chars) {
|
||||
return normalized;
|
||||
}
|
||||
std::string normalized = CondenseWhitespace(region_context);
|
||||
if (normalized.size() <= max_chars) {
|
||||
return normalized;
|
||||
}
|
||||
|
||||
normalized.resize(max_chars);
|
||||
const size_t last_space = normalized.find_last_of(' ');
|
||||
if (last_space != std::string::npos && last_space > max_chars / 2) {
|
||||
normalized.resize(last_space);
|
||||
}
|
||||
normalized.resize(max_chars);
|
||||
const size_t last_space = normalized.find_last_of(' ');
|
||||
if (last_space != std::string::npos && last_space > max_chars / 2) {
|
||||
normalized.resize(last_space);
|
||||
}
|
||||
|
||||
normalized += "...";
|
||||
return normalized;
|
||||
normalized += "...";
|
||||
return normalized;
|
||||
}
|
||||
|
||||
static std::string ToChatPrompt(const llama_model* model,
|
||||
const std::string& system_prompt,
|
||||
const std::string& user_prompt) {
|
||||
std::string combined_prompt;
|
||||
combined_prompt.append(system_prompt);
|
||||
combined_prompt.append("\n\n");
|
||||
combined_prompt.append(user_prompt);
|
||||
const std::string& system_prompt,
|
||||
const std::string& user_prompt) {
|
||||
std::string combined_prompt;
|
||||
combined_prompt.append(system_prompt);
|
||||
combined_prompt.append("\n\n");
|
||||
combined_prompt.append(user_prompt);
|
||||
|
||||
const char* tmpl = llama_model_chat_template(model, nullptr);
|
||||
if (tmpl == nullptr) {
|
||||
// No template found, fallback to raw text
|
||||
spdlog::warn(
|
||||
"LlamaGenerator: missing chat template; using raw prompt fallback");
|
||||
return combined_prompt;
|
||||
}
|
||||
const char* tmpl = llama_model_chat_template(model, nullptr);
|
||||
if (tmpl == nullptr) {
|
||||
// No template found, fallback to raw text
|
||||
spdlog::warn(
|
||||
"LlamaGenerator: missing chat template; using raw prompt fallback");
|
||||
return combined_prompt;
|
||||
}
|
||||
|
||||
const std::array<llama_chat_message, 2> messages = {
|
||||
{{"system", system_prompt.c_str()}, {"user", user_prompt.c_str()}}};
|
||||
const std::array<llama_chat_message, 2> messages = {
|
||||
{{"system", system_prompt.c_str()}, {"user", user_prompt.c_str()}}};
|
||||
|
||||
std::vector<char> buffer(std::max<std::size_t>(
|
||||
1024, (system_prompt.size() + user_prompt.size()) * 4));
|
||||
std::vector<char> buffer(std::max<std::size_t>(
|
||||
1024, (system_prompt.size() + user_prompt.size()) * 4));
|
||||
|
||||
auto apply_template_with_resize =
|
||||
[&](const llama_chat_message* chat_messages,
|
||||
int32_t message_count) -> int32_t {
|
||||
int32_t result = llama_chat_apply_template(
|
||||
tmpl, chat_messages, message_count, true, buffer.data(),
|
||||
static_cast<int32_t>(buffer.size()));
|
||||
|
||||
if (result < 0) {
|
||||
return result;
|
||||
}
|
||||
|
||||
if (result >= static_cast<int32_t>(buffer.size())) {
|
||||
buffer.resize(static_cast<std::size_t>(result) + 1);
|
||||
result = llama_chat_apply_template(
|
||||
tmpl, chat_messages, message_count, true, buffer.data(),
|
||||
static_cast<int32_t>(buffer.size()));
|
||||
}
|
||||
auto apply_template_with_resize = [&](const llama_chat_message* chat_messages,
|
||||
int32_t message_count) -> int32_t {
|
||||
int32_t result = llama_chat_apply_template(
|
||||
tmpl, chat_messages, message_count, true, buffer.data(),
|
||||
static_cast<int32_t>(buffer.size()));
|
||||
|
||||
if (result < 0) {
|
||||
return result;
|
||||
};
|
||||
}
|
||||
|
||||
int32_t template_result = apply_template_with_resize(messages.data(), 2);
|
||||
if (result >= static_cast<int32_t>(buffer.size())) {
|
||||
buffer.resize(static_cast<std::size_t>(result) + 1);
|
||||
result = llama_chat_apply_template(tmpl, chat_messages, message_count,
|
||||
true, buffer.data(),
|
||||
static_cast<int32_t>(buffer.size()));
|
||||
}
|
||||
|
||||
if (template_result >= 0) {
|
||||
return {buffer.data(), static_cast<std::size_t>(template_result)};
|
||||
}
|
||||
return result;
|
||||
};
|
||||
|
||||
spdlog::warn(
|
||||
"LlamaGenerator: chat template rejected system/user messages (result "
|
||||
"{}); trying single user fallback",
|
||||
template_result);
|
||||
int32_t template_result = apply_template_with_resize(messages.data(), 2);
|
||||
|
||||
// FALLBACK: If the template fails (e.g., Model rejecting the "system" role),
|
||||
// combine the system and user prompts into a single "user" message.
|
||||
const std::array<llama_chat_message, 1> fallback_msg = {
|
||||
{{"user", combined_prompt.c_str()}}};
|
||||
if (template_result >= 0) {
|
||||
return {buffer.data(), static_cast<std::size_t>(template_result)};
|
||||
}
|
||||
|
||||
template_result = apply_template_with_resize(fallback_msg.data(), 1);
|
||||
spdlog::warn(
|
||||
"LlamaGenerator: chat template rejected system/user messages (result "
|
||||
"{}); trying single user fallback",
|
||||
template_result);
|
||||
|
||||
// Ultimate fallback: if GGUF template parsing still fails, use raw text.
|
||||
if (template_result < 0) {
|
||||
spdlog::warn(
|
||||
"LlamaGenerator: chat template fallback failed (result {}); using "
|
||||
"raw prompt text",
|
||||
template_result);
|
||||
return combined_prompt;
|
||||
}
|
||||
// FALLBACK: If the template fails (e.g., Model rejecting the "system" role),
|
||||
// combine the system and user prompts into a single "user" message.
|
||||
const std::array<llama_chat_message, 1> fallback_msg = {
|
||||
{{"user", combined_prompt.c_str()}}};
|
||||
|
||||
return {buffer.data(), static_cast<std::size_t>(template_result)};
|
||||
template_result = apply_template_with_resize(fallback_msg.data(), 1);
|
||||
|
||||
// Ultimate fallback: if GGUF template parsing still fails, use raw text.
|
||||
if (template_result < 0) {
|
||||
spdlog::warn(
|
||||
"LlamaGenerator: chat template fallback failed (result {}); using "
|
||||
"raw prompt text",
|
||||
template_result);
|
||||
return combined_prompt;
|
||||
}
|
||||
|
||||
return {buffer.data(), static_cast<std::size_t>(template_result)};
|
||||
}
|
||||
|
||||
static void AppendTokenPiece(const llama_vocab* vocab, llama_token token,
|
||||
std::string& output) {
|
||||
std::array<char, 256> buffer{};
|
||||
int32_t bytes = llama_token_to_piece(vocab, token, buffer.data(),
|
||||
buffer.size(), 0, true);
|
||||
std::array<char, 256> buffer{};
|
||||
int32_t bytes =
|
||||
llama_token_to_piece(vocab, token, buffer.data(), buffer.size(), 0, true);
|
||||
|
||||
if (bytes < 0) {
|
||||
std::vector<char> dynamic_buffer(static_cast<std::size_t>(-bytes));
|
||||
bytes = llama_token_to_piece(vocab, token, dynamic_buffer.data(),
|
||||
static_cast<int32_t>(dynamic_buffer.size()),
|
||||
0, true);
|
||||
if (bytes < 0) {
|
||||
throw std::runtime_error(
|
||||
"LlamaGenerator: failed to decode sampled token piece");
|
||||
}
|
||||
if (bytes < 0) {
|
||||
std::vector<char> dynamic_buffer(static_cast<std::size_t>(-bytes));
|
||||
bytes = llama_token_to_piece(vocab, token, dynamic_buffer.data(),
|
||||
static_cast<int32_t>(dynamic_buffer.size()), 0,
|
||||
true);
|
||||
if (bytes < 0) {
|
||||
throw std::runtime_error(
|
||||
"LlamaGenerator: failed to decode sampled token piece");
|
||||
}
|
||||
|
||||
output.append(dynamic_buffer.data(), static_cast<std::size_t>(bytes));
|
||||
return;
|
||||
}
|
||||
output.append(dynamic_buffer.data(), static_cast<std::size_t>(bytes));
|
||||
return;
|
||||
}
|
||||
|
||||
output.append(buffer.data(), static_cast<std::size_t>(bytes));
|
||||
output.append(buffer.data(), static_cast<std::size_t>(bytes));
|
||||
}
|
||||
|
||||
static bool ExtractLastJsonObject(const std::string& text,
|
||||
std::string& json_out) {
|
||||
std::size_t start = std::string::npos;
|
||||
int depth = 0;
|
||||
bool in_string = false;
|
||||
bool escaped = false;
|
||||
bool found = false;
|
||||
std::string candidate;
|
||||
std::size_t start = std::string::npos;
|
||||
int depth = 0;
|
||||
bool in_string = false;
|
||||
bool escaped = false;
|
||||
bool found = false;
|
||||
std::string candidate;
|
||||
|
||||
for (std::size_t i = 0; i < text.size(); ++i) {
|
||||
const char ch = text[i];
|
||||
for (std::size_t i = 0; i < text.size(); ++i) {
|
||||
const char ch = text[i];
|
||||
|
||||
if (in_string) {
|
||||
if (escaped) {
|
||||
escaped = false;
|
||||
} else if (ch == '\\') {
|
||||
escaped = true;
|
||||
} else if (ch == '"') {
|
||||
in_string = false;
|
||||
}
|
||||
continue;
|
||||
if (in_string) {
|
||||
if (escaped) {
|
||||
escaped = false;
|
||||
} else if (ch == '\\') {
|
||||
escaped = true;
|
||||
} else if (ch == '"') {
|
||||
in_string = false;
|
||||
}
|
||||
continue;
|
||||
}
|
||||
|
||||
if (ch == '"') {
|
||||
in_string = true;
|
||||
continue;
|
||||
if (ch == '"') {
|
||||
in_string = true;
|
||||
continue;
|
||||
}
|
||||
|
||||
if (ch == '{') {
|
||||
if (depth == 0) {
|
||||
start = i;
|
||||
}
|
||||
++depth;
|
||||
continue;
|
||||
}
|
||||
|
||||
if (ch == '{') {
|
||||
if (depth == 0) {
|
||||
start = i;
|
||||
}
|
||||
++depth;
|
||||
continue;
|
||||
if (ch == '}') {
|
||||
if (depth == 0) {
|
||||
continue;
|
||||
}
|
||||
|
||||
if (ch == '}') {
|
||||
if (depth == 0) {
|
||||
continue;
|
||||
}
|
||||
--depth;
|
||||
if (depth == 0 && start != std::string::npos) {
|
||||
candidate = text.substr(start, i - start + 1);
|
||||
found = true;
|
||||
}
|
||||
--depth;
|
||||
if (depth == 0 && start != std::string::npos) {
|
||||
candidate = text.substr(start, i - start + 1);
|
||||
found = true;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (!found) {
|
||||
return false;
|
||||
}
|
||||
if (!found) {
|
||||
return false;
|
||||
}
|
||||
|
||||
json_out = std::move(candidate);
|
||||
return true;
|
||||
json_out = std::move(candidate);
|
||||
return true;
|
||||
}
|
||||
|
||||
std::string ExtractLastJsonObjectPublic(const std::string& text) {
|
||||
std::string extracted;
|
||||
if (ExtractLastJsonObject(text, extracted)) {
|
||||
return extracted;
|
||||
}
|
||||
std::string extracted;
|
||||
if (ExtractLastJsonObject(text, extracted)) {
|
||||
return extracted;
|
||||
}
|
||||
|
||||
return {};
|
||||
return {};
|
||||
}
|
||||
|
||||
static std::optional<std::string> ValidateBreweryJson(
|
||||
const std::string& raw, std::string& name_out,
|
||||
std::string& description_out) {
|
||||
auto validate_object = [&](const boost::json::value& jv,
|
||||
std::string& error_out) -> bool {
|
||||
if (!jv.is_object()) {
|
||||
error_out = "JSON root must be an object";
|
||||
return false;
|
||||
}
|
||||
auto validate_object = [&](const boost::json::value& jv,
|
||||
std::string& error_out) -> bool {
|
||||
if (!jv.is_object()) {
|
||||
error_out = "JSON root must be an object";
|
||||
return false;
|
||||
}
|
||||
|
||||
const auto& obj = jv.get_object();
|
||||
if (!obj.contains("name") || !obj.at("name").is_string()) {
|
||||
error_out = "JSON field 'name' is missing or not a string";
|
||||
return false;
|
||||
}
|
||||
const auto& obj = jv.get_object();
|
||||
if (!obj.contains("name") || !obj.at("name").is_string()) {
|
||||
error_out = "JSON field 'name' is missing or not a string";
|
||||
return false;
|
||||
}
|
||||
|
||||
if (!obj.contains("description") || !obj.at("description").is_string()) {
|
||||
error_out = "JSON field 'description' is missing or not a string";
|
||||
return false;
|
||||
}
|
||||
if (!obj.contains("description") || !obj.at("description").is_string()) {
|
||||
error_out = "JSON field 'description' is missing or not a string";
|
||||
return false;
|
||||
}
|
||||
|
||||
const auto& name_value = obj.at("name").as_string();
|
||||
const auto& description_value = obj.at("description").as_string();
|
||||
name_out = Trim(std::string_view(name_value.data(), name_value.size()));
|
||||
description_out = Trim(
|
||||
std::string_view(description_value.data(), description_value.size()));
|
||||
const auto& name_value = obj.at("name").as_string();
|
||||
const auto& description_value = obj.at("description").as_string();
|
||||
name_out = Trim(std::string_view(name_value.data(), name_value.size()));
|
||||
description_out = Trim(
|
||||
std::string_view(description_value.data(), description_value.size()));
|
||||
|
||||
if (name_out.empty()) {
|
||||
error_out = "JSON field 'name' must not be empty";
|
||||
return false;
|
||||
}
|
||||
if (name_out.empty()) {
|
||||
error_out = "JSON field 'name' must not be empty";
|
||||
return false;
|
||||
}
|
||||
|
||||
if (description_out.empty()) {
|
||||
error_out = "JSON field 'description' must not be empty";
|
||||
return false;
|
||||
}
|
||||
if (description_out.empty()) {
|
||||
error_out = "JSON field 'description' must not be empty";
|
||||
return false;
|
||||
}
|
||||
|
||||
std::string name_lower = name_out;
|
||||
std::string description_lower = description_out;
|
||||
std::transform(
|
||||
name_lower.begin(), name_lower.end(), name_lower.begin(),
|
||||
[](unsigned char c) { return static_cast<char>(std::tolower(c)); });
|
||||
std::transform(description_lower.begin(), description_lower.end(),
|
||||
description_lower.begin(), [](unsigned char c) {
|
||||
return static_cast<char>(std::tolower(c));
|
||||
});
|
||||
std::string name_lower = name_out;
|
||||
std::string description_lower = description_out;
|
||||
std::transform(
|
||||
name_lower.begin(), name_lower.end(), name_lower.begin(),
|
||||
[](unsigned char c) { return static_cast<char>(std::tolower(c)); });
|
||||
std::transform(description_lower.begin(), description_lower.end(),
|
||||
description_lower.begin(), [](unsigned char c) {
|
||||
return static_cast<char>(std::tolower(c));
|
||||
});
|
||||
|
||||
if (name_lower == "string" || description_lower == "string") {
|
||||
error_out = "JSON appears to be a schema placeholder, not content";
|
||||
return false;
|
||||
}
|
||||
if (name_lower == "string" || description_lower == "string") {
|
||||
error_out = "JSON appears to be a schema placeholder, not content";
|
||||
return false;
|
||||
}
|
||||
|
||||
error_out.clear();
|
||||
return true;
|
||||
};
|
||||
error_out.clear();
|
||||
return true;
|
||||
};
|
||||
|
||||
boost::system::error_code ec;
|
||||
boost::json::value jv = boost::json::parse(raw, ec);
|
||||
std::string validation_error;
|
||||
if (ec) {
|
||||
std::string extracted;
|
||||
if (!ExtractLastJsonObject(raw, extracted)) {
|
||||
return "JSON parse error: " + ec.message();
|
||||
}
|
||||
boost::system::error_code ec;
|
||||
boost::json::value jv = boost::json::parse(raw, ec);
|
||||
std::string validation_error;
|
||||
if (ec) {
|
||||
std::string extracted;
|
||||
if (!ExtractLastJsonObject(raw, extracted)) {
|
||||
return "JSON parse error: " + ec.message();
|
||||
}
|
||||
|
||||
ec.clear();
|
||||
jv = boost::json::parse(extracted, ec);
|
||||
if (ec) {
|
||||
return "JSON parse error: " + ec.message();
|
||||
}
|
||||
ec.clear();
|
||||
jv = boost::json::parse(extracted, ec);
|
||||
if (ec) {
|
||||
return "JSON parse error: " + ec.message();
|
||||
}
|
||||
|
||||
if (!validate_object(jv, validation_error)) {
|
||||
return validation_error;
|
||||
}
|
||||
|
||||
return std::nullopt;
|
||||
}
|
||||
|
||||
if (!validate_object(jv, validation_error)) {
|
||||
if (!validate_object(jv, validation_error)) {
|
||||
return validation_error;
|
||||
}
|
||||
}
|
||||
|
||||
return std::nullopt;
|
||||
return std::nullopt;
|
||||
}
|
||||
|
||||
if (!validate_object(jv, validation_error)) {
|
||||
return validation_error;
|
||||
}
|
||||
|
||||
return std::nullopt;
|
||||
}
|
||||
|
||||
// Forward declarations for helper functions exposed to other translation units
|
||||
std::string PrepareRegionContextPublic(std::string_view region_context,
|
||||
std::size_t max_chars) {
|
||||
return PrepareRegionContext(region_context, max_chars);
|
||||
return PrepareRegionContext(region_context, max_chars);
|
||||
}
|
||||
|
||||
std::string ToChatPromptPublic(const llama_model* model,
|
||||
const std::string& system_prompt,
|
||||
const std::string& user_prompt) {
|
||||
return ToChatPrompt(model, system_prompt, user_prompt);
|
||||
return ToChatPrompt(model, system_prompt, user_prompt);
|
||||
}
|
||||
|
||||
void AppendTokenPiecePublic(const llama_vocab* vocab, llama_token token,
|
||||
std::string& output) {
|
||||
AppendTokenPiece(vocab, token, output);
|
||||
AppendTokenPiece(vocab, token, output);
|
||||
}
|
||||
|
||||
std::optional<std::string> ValidateBreweryJsonPublic(
|
||||
const std::string& raw, std::string& name_out,
|
||||
std::string& description_out) {
|
||||
return ValidateBreweryJson(raw, name_out, description_out);
|
||||
return ValidateBreweryJson(raw, name_out, description_out);
|
||||
}
|
||||
|
||||
@@ -14,32 +14,32 @@
|
||||
#include "llama.h"
|
||||
|
||||
void LlamaGenerator::Load(const std::string& model_path) {
|
||||
if (context_ != nullptr) {
|
||||
llama_free(context_);
|
||||
context_ = nullptr;
|
||||
}
|
||||
if (model_ != nullptr) {
|
||||
llama_model_free(model_);
|
||||
model_ = nullptr;
|
||||
}
|
||||
if (context_ != nullptr) {
|
||||
llama_free(context_);
|
||||
context_ = nullptr;
|
||||
}
|
||||
if (model_ != nullptr) {
|
||||
llama_model_free(model_);
|
||||
model_ = nullptr;
|
||||
}
|
||||
|
||||
const llama_model_params model_params = llama_model_default_params();
|
||||
model_ = llama_model_load_from_file(model_path.c_str(), model_params);
|
||||
if (model_ == nullptr) {
|
||||
throw std::runtime_error(
|
||||
"LlamaGenerator: failed to load model from path: " + model_path);
|
||||
}
|
||||
const llama_model_params model_params = llama_model_default_params();
|
||||
model_ = llama_model_load_from_file(model_path.c_str(), model_params);
|
||||
if (model_ == nullptr) {
|
||||
throw std::runtime_error(
|
||||
"LlamaGenerator: failed to load model from path: " + model_path);
|
||||
}
|
||||
|
||||
llama_context_params context_params = llama_context_default_params();
|
||||
context_params.n_ctx = n_ctx_;
|
||||
context_params.n_batch = std::min(n_ctx_, static_cast<uint32_t>(5000));
|
||||
llama_context_params context_params = llama_context_default_params();
|
||||
context_params.n_ctx = n_ctx_;
|
||||
context_params.n_batch = std::min(n_ctx_, static_cast<uint32_t>(5000));
|
||||
|
||||
context_ = llama_init_from_model(model_, context_params);
|
||||
if (context_ == nullptr) {
|
||||
llama_model_free(model_);
|
||||
model_ = nullptr;
|
||||
throw std::runtime_error("LlamaGenerator: failed to create context");
|
||||
}
|
||||
context_ = llama_init_from_model(model_, context_params);
|
||||
if (context_ == nullptr) {
|
||||
llama_model_free(model_);
|
||||
model_ = nullptr;
|
||||
throw std::runtime_error("LlamaGenerator: failed to create context");
|
||||
}
|
||||
|
||||
spdlog::info("[LlamaGenerator] Loaded model: {}", model_path);
|
||||
spdlog::info("[LlamaGenerator] Loaded model: {}", model_path);
|
||||
}
|
||||
|
||||
@@ -21,40 +21,39 @@ namespace fs = std::filesystem;
|
||||
* @return Prompt text loaded from disk.
|
||||
*/
|
||||
std::string LlamaGenerator::LoadBrewerySystemPrompt(
|
||||
const std::string& prompt_file_path) {
|
||||
// Return cached version if already loaded
|
||||
if (!brewery_system_prompt_.empty()) {
|
||||
return brewery_system_prompt_;
|
||||
}
|
||||
const std::string& prompt_file_path) {
|
||||
// Return cached version if already loaded
|
||||
if (!brewery_system_prompt_.empty()) {
|
||||
return brewery_system_prompt_;
|
||||
}
|
||||
|
||||
// Try the provided path only
|
||||
const fs::path prompt_path(prompt_file_path);
|
||||
std::ifstream prompt_file(prompt_path);
|
||||
if (!prompt_file.is_open()) {
|
||||
spdlog::error(
|
||||
"LlamaGenerator: Failed to open brewery system prompt file '{}'",
|
||||
prompt_path.string());
|
||||
throw std::runtime_error(
|
||||
"LlamaGenerator: missing brewery system prompt file: " +
|
||||
prompt_path.string());
|
||||
}
|
||||
// Try the provided path only
|
||||
const fs::path prompt_path(prompt_file_path);
|
||||
std::ifstream prompt_file(prompt_path);
|
||||
if (!prompt_file.is_open()) {
|
||||
spdlog::error(
|
||||
"LlamaGenerator: Failed to open brewery system prompt file '{}'",
|
||||
prompt_path.string());
|
||||
throw std::runtime_error(
|
||||
"LlamaGenerator: missing brewery system prompt file: " +
|
||||
prompt_path.string());
|
||||
}
|
||||
|
||||
const std::string prompt((std::istreambuf_iterator(prompt_file)),
|
||||
std::istreambuf_iterator<char>());
|
||||
prompt_file.close();
|
||||
const std::string prompt((std::istreambuf_iterator(prompt_file)),
|
||||
std::istreambuf_iterator<char>());
|
||||
prompt_file.close();
|
||||
|
||||
if (prompt.empty()) {
|
||||
spdlog::error(
|
||||
"LlamaGenerator: Brewery system prompt file '{}' is empty",
|
||||
prompt_path.string());
|
||||
throw std::runtime_error(
|
||||
"LlamaGenerator: empty brewery system prompt file: " +
|
||||
prompt_path.string());
|
||||
}
|
||||
if (prompt.empty()) {
|
||||
spdlog::error("LlamaGenerator: Brewery system prompt file '{}' is empty",
|
||||
prompt_path.string());
|
||||
throw std::runtime_error(
|
||||
"LlamaGenerator: empty brewery system prompt file: " +
|
||||
prompt_path.string());
|
||||
}
|
||||
|
||||
spdlog::info(
|
||||
spdlog::info(
|
||||
"LlamaGenerator: Loaded brewery system prompt from '{}' ({} chars)",
|
||||
prompt_path.string(), prompt.length());
|
||||
brewery_system_prompt_ = prompt;
|
||||
return brewery_system_prompt_;
|
||||
brewery_system_prompt_ = prompt;
|
||||
return brewery_system_prompt_;
|
||||
}
|
||||
@@ -9,8 +9,8 @@
|
||||
#include "data_generation/mock_generator.h"
|
||||
|
||||
size_t MockGenerator::DeterministicHash(const Location& location) {
|
||||
size_t seed = 0;
|
||||
boost::hash_combine(seed, location.city);
|
||||
boost::hash_combine(seed, location.country);
|
||||
return seed;
|
||||
size_t seed = 0;
|
||||
boost::hash_combine(seed, location.city);
|
||||
boost::hash_combine(seed, location.country);
|
||||
return seed;
|
||||
}
|
||||
|
||||
@@ -12,31 +12,31 @@
|
||||
|
||||
BreweryResult MockGenerator::GenerateBrewery(
|
||||
const Location& location, const std::string& /*region_context*/) {
|
||||
const std::size_t hash = DeterministicHash(location);
|
||||
const std::size_t hash = DeterministicHash(location);
|
||||
|
||||
const std::string_view adjective =
|
||||
kBreweryAdjectives.at(hash % kBreweryAdjectives.size());
|
||||
const std::string_view noun =
|
||||
kBreweryNouns.at(hash / 7 % kBreweryNouns.size());
|
||||
const std::string_view base_description =
|
||||
kBreweryDescriptions.at((hash / 13) % kBreweryDescriptions.size());
|
||||
const std::string_view adjective =
|
||||
kBreweryAdjectives.at(hash % kBreweryAdjectives.size());
|
||||
const std::string_view noun =
|
||||
kBreweryNouns.at(hash / 7 % kBreweryNouns.size());
|
||||
const std::string_view base_description =
|
||||
kBreweryDescriptions.at((hash / 13) % kBreweryDescriptions.size());
|
||||
|
||||
const std::string name =
|
||||
std::format("{} {} {}", location.city, adjective, noun);
|
||||
const std::string name =
|
||||
std::format("{} {} {}", location.city, adjective, noun);
|
||||
|
||||
const std::string state_suffix =
|
||||
location.state_province.empty()
|
||||
? std::string{}
|
||||
: std::format(", {}", location.state_province);
|
||||
const std::string country_suffix =
|
||||
location.country.empty() ? std::string{}
|
||||
: std::format(", {}", location.country);
|
||||
const std::string description = std::format(
|
||||
"{} Located in {}{}{}.", base_description, location.city,
|
||||
state_suffix, country_suffix);
|
||||
const std::string state_suffix =
|
||||
location.state_province.empty()
|
||||
? std::string{}
|
||||
: std::format(", {}", location.state_province);
|
||||
const std::string country_suffix =
|
||||
location.country.empty() ? std::string{}
|
||||
: std::format(", {}", location.country);
|
||||
const std::string description =
|
||||
std::format("{} Located in {}{}{}.", base_description, location.city,
|
||||
state_suffix, country_suffix);
|
||||
|
||||
return {
|
||||
.name = name,
|
||||
.description = description,
|
||||
};
|
||||
return {
|
||||
.name = name,
|
||||
.description = description,
|
||||
};
|
||||
}
|
||||
|
||||
@@ -11,12 +11,12 @@
|
||||
#include "data_generation/mock_generator.h"
|
||||
|
||||
UserResult MockGenerator::GenerateUser(const std::string& locale) {
|
||||
const std::size_t hash = std::hash<std::string>{}(locale);
|
||||
const std::size_t hash = std::hash<std::string>{}(locale);
|
||||
|
||||
UserResult result;
|
||||
const std::string_view username = kUsernames[hash % kUsernames.size()];
|
||||
const std::string_view bio = kBios[hash / 11 % kBios.size()];
|
||||
result.username = username;
|
||||
result.bio = bio;
|
||||
return result;
|
||||
UserResult result;
|
||||
const std::string_view username = kUsernames[hash % kUsernames.size()];
|
||||
const std::string_view bio = kBios[hash / 11 % kBios.size()];
|
||||
result.username = username;
|
||||
result.bio = bio;
|
||||
return result;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user