Function context_overflow_message

server/src/server/http_server.cpp:89–96

#endif

namespace dflash::common {

static std::string context_overflow_message(int max_ctx, int prompt_tokens, int max_output) {
    const int requested_tokens = prompt_tokens + max_output;
    return "This model's maximum context length is " + std::to_string(max_ctx) +
           " tokens. However, you requested " + std::to_string(requested_tokens) +
           " tokens (" + std::to_string(prompt_tokens) + " in the messages, " +
           std::to_string(max_output) +
           " in the completion). Please reduce the length of the messages or completion.";
}