llama-cpp-bin 9628.0.0__tar.gz → 9631.0.0__tar.gz

{llama_cpp_bin-9628.0.0 → llama_cpp_bin-9631.0.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: llama-cpp-bin
-Version: 9628.0.0
+Version: 9631.0.0
 Summary: llama.cpp server binary built from source
 Author-email: Vladyslav Tkachenko <vladlearns@gmail.com>
 License: AGPL-3.0-or-later

{llama_cpp_bin-9628.0.0 → llama_cpp_bin-9631.0.0}/llama.cpp/src/llama-vocab.cpp

@@ -2280,7 +2280,8 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                 clean_spaces = false;
                 ignore_merges = true;
             } else if (
-                tokenizer_pre == "tiny_aya") {
+                tokenizer_pre == "tiny_aya" ||
+                tokenizer_pre == "cohere2moe") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_TINY_AYA;
                 clean_spaces = false;
             } else if (

{llama_cpp_bin-9628.0.0 → llama_cpp_bin-9631.0.0}/llama.cpp/tools/cli/cli.cpp

@@ -97,11 +97,18 @@ struct cli_context {
                 task.params.chat_parser_params.parser.load(chat_params.parser);
             }
 
+            // Copy the preserved tokens into the sampling params
+            const llama_vocab * vocab = llama_model_get_vocab(
+                llama_get_model(ctx_server.get_llama_context()));
+            for (const auto & token : chat_params.preserved_tokens) {
+                auto ids = common_tokenize(vocab, token, false, true);
+                if (ids.size() == 1) {
+                    task.params.sampling.preserved_tokens.insert(ids[0]);
+                }
+            }
+
             // reasoning budget sampler
             if (!chat_params.thinking_end_tag.empty()) {
-                const llama_vocab * vocab = llama_model_get_vocab(
-                    llama_get_model(ctx_server.get_llama_context()));
-
                 task.params.sampling.reasoning_budget_tokens = defaults.sampling.reasoning_budget_tokens;
                 task.params.sampling.generation_prompt = chat_params.generation_prompt;
 

{llama_cpp_bin-9628.0.0 → llama_cpp_bin-9631.0.0}/src/llama_cpp_bin/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '9628.0.0'
-__version_tuple__ = version_tuple = (9628, 0, 0)
+__version__ = version = '9631.0.0'
+__version_tuple__ = version_tuple = (9631, 0, 0)
 
-__commit_id__ = commit_id = 'g230bc8dab'
+__commit_id__ = commit_id = 'g81a532888'

{llama_cpp_bin-9628.0.0 → llama_cpp_bin-9631.0.0}/src/llama_cpp_bin.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: llama-cpp-bin
-Version: 9628.0.0
+Version: 9631.0.0
 Summary: llama.cpp server binary built from source
 Author-email: Vladyslav Tkachenko <vladlearns@gmail.com>
 License: AGPL-3.0-or-later