Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-27 14:49:25 +01:00
Code Issues Packages Projects Releases Wiki Activity

server : fix segfault on long system prompt (#8987)

Browse Source 
* server : fix segfault on long system prompt

* server : fix parallel generation with very small batch sizes

* server : fix typo in comment
This commit is contained in:
compilade 2024-08-14 02:51:02 -04:00 committed by GitHub
parent 43bdd3ce18
commit 98a532d474
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 12 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
examples/server/server.cpp
View File

@ -754,13 +754,13 @@ struct server_context {
default_generation_settings_for_props = get_formated_generation(slots.front()); default_generation_settings_for_props = get_formated_generation(slots.front());
default_generation_settings_for_props["seed"] = -1; default_generation_settings_for_props["seed"] = -1;
// the update_slots() logic will always submit a maximum of n_batch tokens // the update_slots() logic will always submit a maximum of n_batch or n_parallel tokens
// note that n_batch can be > n_ctx (e.g. for non-causal attention models such as BERT where the KV cache is not used) // note that n_batch can be > n_ctx (e.g. for non-causal attention models such as BERT where the KV cache is not used)
{ {
const int32_t n_batch = llama_n_batch(ctx); const int32_t n_batch = llama_n_batch(ctx);
// only a single seq_id per token is needed // only a single seq_id per token is needed
batch = llama_batch_init(n_batch, 0, 1); batch = llama_batch_init(std::max(n_batch, params.n_parallel), 0, 1);
} }
metrics.init(); metrics.init();
@ -1137,28 +1137,19 @@ struct server_context {
if (!system_prompt.empty()) { if (!system_prompt.empty()) {
system_tokens = ::llama_tokenize(ctx, system_prompt, true); system_tokens = ::llama_tokenize(ctx, system_prompt, true);
const int32_t n_batch = llama_n_batch(ctx);
const int32_t n_tokens_prompt = system_tokens.size();
for (int32_t i = 0; i < n_tokens_prompt; i += n_batch) {
const int32_t n_tokens = std::min(n_batch, n_tokens_prompt - i);
llama_batch_clear(batch); llama_batch_clear(batch);
for (int i = 0; i < (int)system_tokens.size(); ++i) { for (int32_t j = 0; j < n_tokens; ++j) {
llama_batch_add(batch, system_tokens[i], i, { 0 }, false); llama_batch_add(batch, system_tokens[i + j], i + j, { 0 }, false);
} }
const int32_t n_batch = llama_n_batch(ctx); if (llama_decode(ctx, batch) != 0) {
for (int32_t i = 0; i < batch.n_tokens; i += n_batch) {
const int32_t n_tokens = std::min(params.n_batch, batch.n_tokens - i);
llama_batch batch_view = {
n_tokens,
batch.token + i,
nullptr,
batch.pos + i,
batch.n_seq_id + i,
batch.seq_id + i,
batch.logits + i,
0, 0, 0, // unused
};
if (llama_decode(ctx, batch_view) != 0) {
LOG_ERROR("llama_decode() failed", {}); LOG_ERROR("llama_decode() failed", {});
return; return;
} }