vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

common : change defaults for escape and n_ctx

Browse source
This commit is contained in:
Georgi Gerganov 2024-06-04 13:04:31 +03:00
parent c4b6b83811
commit ea4665e6ef
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
3 changed files with 12 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

7
common/common.cpp
View file

@ -353,6 +353,10 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
params.escape = true;
return true;
}
if (arg == "--no-escape") {
params.escape = false;
return true;
}
if (arg == "--prompt-cache") {
if (++i >= argc) {
invalid_param = true;
@ -1629,7 +1633,8 @@ void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & param
options.push_back({ "*", "-p, --prompt PROMPT", "prompt to start generation with (default: '%s')", params.prompt.c_str() });
options.push_back({ "*", "-f, --file FNAME", "a file containing the prompt (default: none)" });
options.push_back({ "*", "-bf, --binary-file FNAME", "binary file containing the prompt (default: none)" });
options.push_back({ "*", "-e, --escape", "process escapes sequences (\\n, \\r, \\t, \\', \\\", \\\\)" });
options.push_back({ "*", "-e, --escape", "process escapes sequences (\\n, \\r, \\t, \\', \\\", \\\\) (default: %s)", params.escape ? "true" : "false" });
options.push_back({ "*", " --no-escape", "do not process escape sequences" });
options.push_back({ "main", "-ptc, --print-token-count N", "print token count every N tokens (default: %d)", params.n_print });
options.push_back({ "main", " --prompt-cache FNAME", "file to cache prompt state for faster startup (default: none)" });
options.push_back({ "main", " --prompt-cache-all", "if specified, saves user input and generations to cache as well\n"

4
common/common.h
View file

@ -60,7 +60,7 @@ struct gpt_params {
int32_t n_threads_batch = -1; // number of threads to use for batch processing (-1 = use n_threads)
int32_t n_threads_batch_draft = -1;
int32_t n_predict = -1; // new tokens to predict
int32_t n_ctx = 512; // context size
int32_t n_ctx = 0; // context size
int32_t n_batch = 2048; // logical batch size for prompt processing (must be >=32 to use BLAS)
int32_t n_ubatch = 512; // physical batch size for prompt processing (must be >=32 to use BLAS)
int32_t n_keep = 0; // number of tokens to keep from initial prompt
@ -153,7 +153,7 @@ struct gpt_params {
bool prompt_cache_ro = false; // open the prompt cache read-only and do not update it
bool embedding = false; // get only sentence embedding
bool escape = false; // escape "\n", "\r", "\t", "\'", "\"", and "\\"
bool escape = true; // escape "\n", "\r", "\t", "\'", "\"", and "\\"
bool multiline_input = false; // reverse the usage of `\`
bool simple_io = false; // improves compatibility with subprocesses and limited consoles
bool cont_batching = true; // insert new sequences for decoding on-the-fly

6
examples/perplexity/perplexity.cpp
View file

@ -1964,13 +1964,14 @@ static void kl_divergence(llama_context * ctx, const gpt_params & params) {
int main(int argc, char ** argv) {
gpt_params params;
params.n_ctx = 512;
params.logits_all = true;
if (!gpt_params_parse(argc, argv, params)) {
gpt_params_print_usage(argc, argv, params);
return 1;
}
params.logits_all = true;
const int32_t n_ctx = params.n_ctx;
if (n_ctx <= 0) {
@ -2025,6 +2026,7 @@ int main(int argc, char ** argv) {
}
const int n_ctx_train = llama_n_ctx_train(model);
if (params.n_ctx > n_ctx_train) {
fprintf(stderr, "%s: warning: model was trained on only %d context tokens (%d specified)\n",
__func__, n_ctx_train, params.n_ctx);