Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-08-30 09:00:49 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into concedo_experimental

Browse Source 
# Conflicts:
#	Makefile
#	ggml-cuda.cu
#	tests/test-tokenizer-0-falcon.py
#	tests/test-tokenizer-0-llama.py
This commit is contained in:
Concedo
2023-11-18 11:10:45 +08:00
parent 35ce2b054f bbecf3f415
commit 6bf8ee4aea
29 changed files with 448 additions and 113 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/server/server.cpp
+6 -3
View File
@@ -502,6 +502,7 @@ struct llama_server_context
bool multimodal = false;
bool clean_kv_cache = true;
bool all_slots_are_idle = false;
bool add_bos_token = true;
int32_t id_gen;
int32_t n_ctx; // total context for all clients / slots
@@ -574,6 +575,8 @@ struct llama_server_context
n_ctx = llama_n_ctx(ctx);
add_bos_token = llama_should_add_bos_token(model);
return true;
}
@@ -865,7 +868,7 @@ struct llama_server_context
}
void update_system_prompt() {
system_tokens = ::llama_tokenize(ctx, system_prompt, true);
system_tokens = ::llama_tokenize(ctx, system_prompt, add_bos_token);
llama_batch_clear(batch);
@@ -1553,7 +1556,7 @@ struct llama_server_context
}
else
{
prompt_tokens = tokenize(slot.prompt, system_prompt.empty()); // add BOS if there isn't system prompt
prompt_tokens = tokenize(slot.prompt, system_prompt.empty() && add_bos_token); // add BOS if there isn't system prompt
}
slot.num_prompt_tokens = prompt_tokens.size();
@@ -1630,7 +1633,7 @@ struct llama_server_context
const bool has_images = process_images(slot);
// process the prefix of first image
std::vector<llama_token> prefix_tokens = has_images ? tokenize(slot.images[0].prefix_prompt, true) : prompt_tokens;
std::vector<llama_token> prefix_tokens = has_images ? tokenize(slot.images[0].prefix_prompt, add_bos_token) : prompt_tokens;
for (; slot.n_past < (int) prefix_tokens.size(); ++slot.n_past)
{
llama_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot.n_past, { slot.id }, false);