sdgoij/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-03-17 16:44:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

server : do not create checkpoints right after mtmd chunks (#20232)

Browse Source
This commit is contained in:
Georgi Gerganov
2026-03-08 22:16:46 +02:00
committed by GitHub
parent 35bee031e1
commit d417bc43dd
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tools/server/server-context.cpp
View File

@@ -2438,6 +2438,8 @@ private:
slot.n_prompt_tokens_cache = 0;
}
bool do_checkpoint = params_base.n_ctx_checkpoints > 0;
// check if we should process the image
if (slot.prompt.n_tokens() < slot.task->n_tokens() && input_tokens[slot.prompt.n_tokens()] == LLAMA_TOKEN_NULL) {
// process the image
@@ -2457,6 +2459,8 @@ private:
const auto & chunk = input_tokens.find_chunk(slot.prompt.n_tokens());
slot.prompt.tokens.push_back(chunk.get()); // copy
}
do_checkpoint = false; // do not checkpoint right after an image chunk
}
// If using an alora, there may be uncached tokens that come
@@ -2473,8 +2477,6 @@ private:
alora_disabled_id = enabled_loras[0];
}
bool do_checkpoint = params_base.n_ctx_checkpoints > 0;
// make checkpoints only for completion tasks
do_checkpoint = do_checkpoint && slot.task->type == SERVER_TASK_TYPE_COMPLETION;