CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-05 09:36:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

server : simplify context limit logic

Browse Source
This commit is contained in:
Georgi Gerganov
2025-10-28 12:38:17 +02:00
parent 7ebe7f77a1
commit edd2e8cfc7
1 changed files with 3 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
tools/server/server.cpp
View File

@@ -2866,10 +2866,12 @@ struct server_context {
// if context shifting is disabled, make sure that we don't run out of context // if context shifting is disabled, make sure that we don't run out of context
if (!params_base.ctx_shift && slot.n_past + 1 >= slot.n_ctx) { if (!params_base.ctx_shift && slot.n_past + 1 >= slot.n_ctx) {
slot.truncated = true;
slot.stop = STOP_TYPE_LIMIT; slot.stop = STOP_TYPE_LIMIT;
slot.has_next_token = false; slot.has_next_token = false;
SLT_DBG(slot, "stopped due to running out of context, n_past = %d, n_ctx = %d\n", slot.n_past, slot.n_ctx); SLT_DBG(slot, "stopped due to running out of context capacity, n_past = %d, n_prompt_tokens = %d, n_decoded = %d, n_ctx = %d\n",
slot.n_decoded, slot.n_prompt_tokens(), slot.n_past, slot.n_ctx);
} }
// check the limits // check the limits
@@ -2929,16 +2931,6 @@ struct server_context {
} }
} }
// if context shift is disabled, we stop when it reaches the context limit
if (slot.n_past >= slot.n_ctx) {
slot.truncated = true;
slot.stop = STOP_TYPE_LIMIT;
slot.has_next_token = false;
SLT_DBG(slot, "stopped due to running out of context capacity, n_past = %d, n_prompt_tokens = %d, n_decoded = %d, n_ctx = %d\n",
slot.n_decoded, slot.n_prompt_tokens(), slot.n_past, slot.n_ctx);
}
if (llama_vocab_is_eog(vocab, result.tok)) { if (llama_vocab_is_eog(vocab, result.tok)) {
slot.stop = STOP_TYPE_EOS; slot.stop = STOP_TYPE_EOS;
slot.has_next_token = false; slot.has_next_token = false;
@@ -2946,17 +2938,6 @@ struct server_context {
SLT_DBG(slot, "%s", "stopped by EOS\n"); SLT_DBG(slot, "%s", "stopped by EOS\n");
} }
if (slot.task->params.n_predict < 1 && slot.n_prompt_tokens() + slot.n_decoded >= slot.n_ctx) {
slot.truncated = true;
slot.stop = STOP_TYPE_LIMIT;
slot.has_next_token = false; // stop prediction
SLT_WRN(slot,
"n_predict (%d) is set for infinite generation. "
"Limiting generated tokens to slot.n_ctx (%d) to avoid EOS-less generation infinite loop\n",
slot.task->params.n_predict, slot.n_ctx);
}
SLT_DBG(slot, "n_decoded = %d, n_remaining = %d, next token: %5d '%s'\n", slot.n_decoded, slot.n_remaining, result.tok, token_str.c_str()); SLT_DBG(slot, "n_decoded = %d, n_remaining = %d, next token: %5d '%s'\n", slot.n_decoded, slot.n_remaining, result.tok, token_str.c_str());
return slot.has_next_token; // continue return slot.has_next_token; // continue