server : add --override-kv parameter (#4710)

* Changes to server to allow metadata override

* documentation

* flake.nix: expose full scope in legacyPackages

* flake.nix: rocm not yet supported on aarch64, so hide the output

* flake.nix: expose checks

* workflows: nix-ci: init; build flake outputs

* workflows: nix-ci: add a job for eval

* workflows: weekly `nix flake update`

* workflows: nix-flakestry: drop tag filters

...and add a job for flakehub.com

* workflows: nix-ci: add a qemu job for jetsons

* flake.nix: suggest the binary caches

* flake.lock: update

to a commit recently cached by nixpkgs-cuda-ci

---------

Co-authored-by: John <john@jLap.lan>
Co-authored-by: Someone Serge <sergei.kozlukov@aalto.fi>
This commit is contained in:
minarchist 2024-01-02 04:38:15 -06:00 committed by GitHub
parent 26f3071d71
commit 5d7002d437
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -2016,6 +2016,10 @@ static void server_print_usage(const char *argv0, const gpt_params &params,
printf(" --mmproj MMPROJ_FILE path to a multimodal projector file for LLaVA.\n");
printf(" --log-disable disables logging to a file.\n");
printf("\n");
printf(" --override-kv KEY=TYPE:VALUE\n");
printf(" advanced option to override model metadata by key. may be specified multiple times.\n");
printf(" types: int, float, bool. example: --override-kv tokenizer.ggml.add_bos_token=bool:false\n");
printf("\n");
}
static void server_params_parse(int argc, char **argv, server_params &sparams,
@ -2379,6 +2383,49 @@ static void server_params_parse(int argc, char **argv, server_params &sparams,
log_set_target(stdout);
LOG_INFO("logging to file is disabled.", {});
}
else if (arg == "--override-kv")
{
if (++i >= argc) {
invalid_param = true;
break;
}
char * sep = strchr(argv[i], '=');
if (sep == nullptr || sep - argv[i] >= 128) {
fprintf(stderr, "error: Malformed KV override: %s\n", argv[i]);
invalid_param = true;
break;
}
struct llama_model_kv_override kvo;
std::strncpy(kvo.key, argv[i], sep - argv[i]);
kvo.key[sep - argv[i]] = 0;
sep++;
if (strncmp(sep, "int:", 4) == 0) {
sep += 4;
kvo.tag = LLAMA_KV_OVERRIDE_INT;
kvo.int_value = std::atol(sep);
} else if (strncmp(sep, "float:", 6) == 0) {
sep += 6;
kvo.tag = LLAMA_KV_OVERRIDE_FLOAT;
kvo.float_value = std::atof(sep);
} else if (strncmp(sep, "bool:", 5) == 0) {
sep += 5;
kvo.tag = LLAMA_KV_OVERRIDE_BOOL;
if (std::strcmp(sep, "true") == 0) {
kvo.bool_value = true;
} else if (std::strcmp(sep, "false") == 0) {
kvo.bool_value = false;
} else {
fprintf(stderr, "error: Invalid boolean value for KV override: %s\n", argv[i]);
invalid_param = true;
break;
}
} else {
fprintf(stderr, "error: Invalid type for KV override: %s\n", argv[i]);
invalid_param = true;
break;
}
params.kv_overrides.push_back(kvo);
}
else
{
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
@ -2386,6 +2433,10 @@ static void server_params_parse(int argc, char **argv, server_params &sparams,
exit(1);
}
}
if (!params.kv_overrides.empty()) {
params.kv_overrides.emplace_back(llama_model_kv_override());
params.kv_overrides.back().key[0] = 0;
}
if (invalid_param)
{