CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

server : --n-predict option document and cap to max value (#5549)

Browse Source 
* server: document --n-predict

* server: ensure client request cannot override n_predict if set

* server: fix print usage LF in new --n-predict option
This commit is contained in:
Pierrick Hymbert
2024-02-18 17:30:09 +01:00
committed by GitHub
parent 66c1968f7a
commit 36376abe05
2 changed files with 15 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
examples/server/README.md
View File

@@ -39,6 +39,7 @@ see https://github.com/ggerganov/llama.cpp/issues/1437
- `--mmproj MMPROJ_FILE`: Path to a multimodal projector file for LLaVA. - `--mmproj MMPROJ_FILE`: Path to a multimodal projector file for LLaVA.
- `--grp-attn-n`: Set the group attention factor to extend context size through self-extend(default: 1=disabled), used together with group attention width `--grp-attn-w` - `--grp-attn-n`: Set the group attention factor to extend context size through self-extend(default: 1=disabled), used together with group attention width `--grp-attn-w`
- `--grp-attn-w`: Set the group attention width to extend context size through self-extend(default: 512), used together with group attention factor `--grp-attn-n` - `--grp-attn-w`: Set the group attention width to extend context size through self-extend(default: 512), used together with group attention factor `--grp-attn-n`
- `-n, --n-predict`: Set the maximum tokens to predict (default: -1)
## Build ## Build

15
examples/server/server.cpp
View File

@@ -159,6 +159,7 @@ struct llama_client_slot
int32_t n_decoded = 0; int32_t n_decoded = 0;
int32_t n_remaining = -1; int32_t n_remaining = -1;
int32_t i_batch = -1; int32_t i_batch = -1;
int32_t n_predict = -1;
int32_t num_prompt_tokens = 0; int32_t num_prompt_tokens = 0;
int32_t num_prompt_tokens_processed = 0; int32_t num_prompt_tokens_processed = 0;
@@ -410,6 +411,7 @@ struct llama_server_context
slot.id = i; slot.id = i;
slot.n_ctx = n_ctx_slot; slot.n_ctx = n_ctx_slot;
slot.n_predict = params.n_predict;
LOG_TEE(" -> Slot %i - max context: %i\n", slot.id, n_ctx_slot); LOG_TEE(" -> Slot %i - max context: %i\n", slot.id, n_ctx_slot);
@@ -546,6 +548,15 @@ struct llama_server_context
slot->sparams.grammar = json_value(data, "grammar", default_sparams.grammar); slot->sparams.grammar = json_value(data, "grammar", default_sparams.grammar);
slot->sparams.n_probs = json_value(data, "n_probs", default_sparams.n_probs); slot->sparams.n_probs = json_value(data, "n_probs", default_sparams.n_probs);
if (slot->n_predict > 0 && slot->params.n_predict > slot->n_predict) {
// Might be better to reject the request with a 400 ?
LOG_WARNING("Max tokens to predict exceeds server configuration", {
{"params.n_predict", slot->params.n_predict},
{"slot.n_predict", slot->n_predict},
});
slot->params.n_predict = slot->n_predict;
}
// infill // infill
if (data.count("input_prefix") != 0) if (data.count("input_prefix") != 0)
{ {
@@ -1053,6 +1064,7 @@ struct llama_server_context
return json { return json {
{"n_ctx", slot.n_ctx}, {"n_ctx", slot.n_ctx},
{"n_predict", slot.n_predict},
{"model", params.model_alias}, {"model", params.model_alias},
{"seed", slot.params.seed}, {"seed", slot.params.seed},
{"temperature", slot.sparams.temp}, {"temperature", slot.sparams.temp},
@@ -1915,13 +1927,14 @@ 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(" --mmproj MMPROJ_FILE path to a multimodal projector file for LLaVA.\n");
printf(" --log-disable disables logging to a file.\n"); printf(" --log-disable disables logging to a file.\n");
printf("\n"); printf("\n");
printf(" -n, --n-predict maximum tokens to predict (default: %d)\n", params.n_predict);
printf(" --override-kv KEY=TYPE:VALUE\n"); printf(" --override-kv KEY=TYPE:VALUE\n");
printf(" advanced option to override model metadata by key. may be specified multiple times.\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(" types: int, float, bool. example: --override-kv tokenizer.ggml.add_bos_token=bool:false\n");
printf(" -gan N, --grp-attn-n N set the group attention factor to extend context size through self-extend(default: 1=disabled), used together with group attention width `--grp-attn-w`"); printf(" -gan N, --grp-attn-n N set the group attention factor to extend context size through self-extend(default: 1=disabled), used together with group attention width `--grp-attn-w`");
printf(" -gaw N, --grp-attn-w N set the group attention width to extend context size through self-extend(default: 512), used together with group attention factor `--grp-attn-n`"); printf(" -gaw N, --grp-attn-w N set the group attention width to extend context size through self-extend(default: 512), used together with group attention factor `--grp-attn-n`");
printf(" --chat-template FORMAT_NAME"); printf(" --chat-template FORMAT_NAME");
printf(" set chat template, possible valus is: llama2, chatml (default %s)", sparams.chat_template.c_str()); printf(" set chat template, possible value is: llama2, chatml (default %s)", sparams.chat_template.c_str());
printf("\n"); printf("\n");
} }