LLM inference in C/C++
C++ ggml
server, ui : Add support for HTTP ETags in llama-server (#23701)
Markus Tavenrathd205df6812e2
diff --git a/tools/server/server-http.cpp b/tools/server/server-http.cppindex c0a9af8f09..00290b0782 100644--- a/tools/server/server-http.cpp+++ b/tools/server/server-http.cpp@@ -318,12 +318,19 @@ bool server_http_context::init(const common_params & params) {} else {#if defined(LLAMA_UI_HAS_ASSETS)auto serve_asset = [](const std::string & name, const char * mime, bool with_isolation_headers) {- return [name, mime, with_isolation_headers](const httplib::Request & /*req*/, httplib::Response & res) {+ return [name, mime, with_isolation_headers](const httplib::Request & req, httplib::Response & res) {const llama_ui_asset * a = llama_ui_find_asset(name.c_str());if (!a) {res.status = 404;return false;}+ res.set_header("ETag", a->etag);+ // Check If-None-Match for conditional GET (304 Not Modified)+ if (const std::string & inm = req.get_header_value("If-None-Match");+ !inm.empty() && inm == a->etag) {+ res.status = 304;+ return false;+ }if (with_isolation_headers) {// COEP and COOP headers, required by pyodide (python interpreter)res.set_header("Cross-Origin-Embedder-Policy", "require-corp");diff --git a/tools/ui/embed.cpp b/tools/ui/embed.cppindex 41227868e6..c88faefa14 100644--- a/tools/ui/embed.cpp+++ b/tools/ui/embed.cpp@@ -9,6 +9,19 @@#include <fstream>#include <string>#include <vector>+#include <cstdint>++// Computes FNV-1a hash of the data+static uint64_t fnv_hash(const uint8_t * data, size_t len) {+ const uint64_t fnv_prime = 0x100000001b3ULL;+ uint64_t hash = 0xcbf29ce484222325ULL;++ for (size_t i = 0; i < len; ++i) {+ hash ^= data[i];+ hash *= fnv_prime;+ }+ return hash;+}static bool read_file(const std::string & path, std::vector<unsigned char> & out) {std::ifstream f(path, std::ios::binary | std::ios::ate);@@ -95,6 +108,7 @@ int main(int argc, char ** argv) {" const char * name;\n"" const unsigned char * data;\n"" size_t size;\n"+ " const char * etag;\n""};\n\n""const llama_ui_asset * llama_ui_find_asset(const char * name);\n";@@ -110,14 +124,18 @@ int main(int argc, char ** argv) {}cpp += fmt("static const unsigned char asset_%d_data[] = {", i);append_bytes_hex(cpp, bytes);- cpp += fmt("};\nstatic const size_t asset_%d_size = %lu;\n\n",+ const auto hash = fnv_hash(bytes.data(), bytes.size());++ cpp += fmt("};\nstatic const size_t asset_%d_size = %lu;\n",i, static_cast<unsigned long>(bytes.size()));+ cpp += fmt("static const char asset_%d_etag[] = \"\\\"0x%016lx\\\"\";\n\n",+ i, static_cast<unsigned long>(hash));}cpp += "static const llama_ui_asset g_assets[] = {\n";for (int i = 0; i < n_assets; i++) {- const char * name = argv[3 + i * 2];- cpp += fmt(" { \"%s\", asset_%d_data, asset_%d_size },\n", name, i, i);+ cpp += fmt(" { \"%s\", asset_%d_data, asset_%d_size, asset_%d_etag },\n",+ argv[3 + i * 2], i, i, i);}cpp += "};\n\n";