upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-08-06 01:50:50 -05:00
Code Issues Packages Projects Releases Wiki Activity

sampling : expand support (wip)

Browse Source
This commit is contained in:
Georgi Gerganov
2025-12-05 22:02:48 +02:00
parent e652566139
commit 30742a6ff5
16 changed files with 167 additions and 149 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tools/server/server-context.cpp
View File

@@ -1003,7 +1003,12 @@ struct server_context_impl {
return false;
}
llama_set_sampler(ctx, slot.id, common_sampler_get(slot.smpl.get()));
// TODO: tmp until backend sampling is fully implemented
if (task.params.sampling.backend_sampling) {
llama_set_sampler(ctx, slot.id, common_sampler_get(slot.smpl.get()));
} else {
llama_set_sampler(ctx, slot.id, nullptr);
}
SLT_INF(slot, "sampler chain: %s\n", common_sampler_print(slot.smpl.get()).c_str());
}
@@ -1156,6 +1161,7 @@ struct server_context_impl {
return slot.has_next_token; // continue
}
// TODO: does not work with backend sampling
void populate_token_probs(const server_slot & slot, completion_token_output & result, bool post_sampling, bool special, int idx) const {
size_t n_probs = slot.task->params.sampling.n_probs;
size_t n_vocab = llama_vocab_n_tokens(vocab);