I tried to replicate this guys test with gemma4. I don't have his code, but I used the largest code file in OpenSSL I could find: https://github.com/openssl/openssl/blob/master/test/sslapitest.c

This is even larger than his code, 461kB, 15k lines of code, it took 17k tokens prompt processing. This was my prompt:

I tried the prompt from the video:

reproduce the 20 lines immediately following the opening bracket { of the function static int test_keylog_output, copy verbatim, do not paraphrase, do not comment

Answer:

char *token = NULL;
unsigned char actual_client_random[SSL3_RANDOM_SIZE] = { 0 };
size_t client_random_size = SSL3_RANDOM_SIZE;
unsigned char actual_master_key[SSL_MAX_MASTER_KEY_LENGTH] = { 0 };
size_t master_key_size = SSL_MAX_MASTER_KEY_LENGTH;
unsigned int rsa_key_exchange_count = 0;
unsigned int master_secret_count = 0;
unsigned int client_early_secret_count = 0;
unsigned int client_handshake_secret_count = 0;
unsigned int server_handshake_secret_count = 0;
unsigned int client_application_secret_count = 0;
unsigned int server_application_secret_count = 0;
unsigned int early_exporter_secret_count = 0;
unsigned int exporter_secret_count = 0;

for (token = strtok(buffer, " \n"); token != NULL;
    token = strtok(NULL, " \n")) {
    if (strcmp(token, "RSA") == 0) {
        /*
         * Premaster secret. Tokens should be: 16 ASCII bytes of

Diff found only removing of four empty space characters at the start of each line, other than that 100% reproduced.

One caviat though is that I'm using the full BF16 version of the Gemma4 26B A4B model. In his video the author had the Q4 version and had some workaround for the insufficient context length out of the box. I wonder whether this had caused the problem.