jdelony/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-06-27 23:50:20 -05:00
Code Issues Packages Projects Releases Wiki Activity

Some renames to make @CISC happy :>

Browse Source
This commit is contained in:
Piotr Wilkin 2026-06-14 19:52:23 +02:00
parent f2bb114a32
commit 4c4a3e2596
3 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
common/chat.cpp
View File

@ -1979,7 +1979,7 @@ static common_chat_params common_chat_params_init_deepseek_v3_2(const common_cha
return data;
}
// Cohere2-MoE (a.k.a. "North Code") parser.
// Cohere2 MoE (a.k.a. "North Code") parser.
//
// The assistant turn is fully marker-wrapped:
// <|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>
@ -1995,7 +1995,7 @@ static common_chat_params common_chat_params_init_deepseek_v3_2(const common_cha
// parser literal therefore only covers the stable <|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|> prefix
// and the reasoning rule consumes the <|START_THINKING|> ... <|END_THINKING|> markers itself,
// regardless of whether they came from the generation prompt or the generated text.
static common_chat_params common_chat_params_init_cohere2_moe(const common_chat_template & tmpl,
static common_chat_params common_chat_params_init_cohere2moe(const common_chat_template & tmpl,
const autoparser::generation_params & inputs) {
common_chat_params data;
@ -2367,13 +2367,13 @@ std::optional<common_chat_params> common_chat_try_specialized_template(
return common_chat_params_init_kimi_k2(tmpl, params);
}
// Cohere2-MoE / North Code - marker-wrapped format with <|START_TEXT|> content and
// Cohere2 MoE / North Code - marker-wrapped format with <|START_TEXT|> content and
// <|START_ACTION|> JSON tool calls. <|START_TEXT|> is unique to this template (the older
// Command-R templates use <|START_RESPONSE|>).
if (src.find("<|START_TEXT|>") != std::string::npos &&
src.find("<|START_ACTION|>") != std::string::npos) {
LOG_DBG("Using specialized template: Cohere2-MoE\n");
return common_chat_params_init_cohere2_moe(tmpl, params);
LOG_DBG("Using specialized template: Cohere2 MoE\n");
return common_chat_params_init_cohere2moe(tmpl, params);
}
if (is_lfm2_template(src)) {

0
models/templates/Cohere2-MoE.jinja → models/templates/Cohere2MoE.jinja
View File

4
tests/test-chat.cpp
View File

@ -2645,12 +2645,12 @@ static void test_template_output_peg_parsers(bool detailed_debug) {
}
{
// Cohere2-MoE (North Code) - dedicated parser.
// Cohere2 MoE (North Code) - dedicated parser.
// Marker-wrapped format: <|START_THINKING|>...<|END_THINKING|> then either
// <|START_TEXT|>...<|END_TEXT|> (content) or <|START_ACTION|>[json]<|END_ACTION|> (tools).
// The generation prompt forces a leading <|START_THINKING|>, so model output begins inside
// the thinking block: test inputs start with the reasoning body, not the <|START_THINKING|> tag.
auto tst = peg_tester("models/templates/Cohere2-MoE.jinja", detailed_debug);
auto tst = peg_tester("models/templates/Cohere2MoE.jinja", detailed_debug);
// Content with reasoning, extracted.
tst.test("I'm\nthinking<|END_THINKING|><|START_TEXT|>Hello, world!\nWhat's up?<|END_TEXT|>")