feat: fix rebase confict.

Author: Kang-Meng

xllm_service/chat_template/CMakeLists.txt

@@ -6,10 +6,8 @@ cc_library (
     chat_template
   HDRS
     jinja_chat_template.h
-    chat_template_factory.h
   SRCS
     jinja_chat_template.cpp
-    chat_template_factory.cpp
   DEPS
     :minja
     :tokenizer

xllm_service/chat_template/chat_template_factory.cpp

@@ -4,7 +4,7 @@
 #include <array>
 #include <nlohmann/json.hpp>
 
-#include "chat_template/chat_template_factory.h"
+#include "chat_template/jinja_chat_template.h"
 #include "common.pb.h"
 #include "common/hash_util.h"
 #include "tokenizer/tokenizer_factory.h"
@@ -20,9 +20,9 @@ Scheduler::Scheduler(const RpcServiceConfig& rpc_config,
     : rpc_config_(rpc_config),
       model_config_(model_config),
       http_config_(http_config) {
-  tokenizer_ = create_tokenizer(model_config_, &tokenizer_args_);
-  chat_template_ =
-      create_chat_template(model_config_.model_type, tokenizer_args_);
+  tokenizer_ = TokenizerFactory::create_tokenizer(model_config_.tokenizer_path,
+                                                  &tokenizer_args_);
+  chat_template_ = std::make_unique<JinjaChatTemplate>(tokenizer_args_);
 
   etcd_client_ = std::make_shared<EtcdClient>(rpc_config_.etcd_addr);
 

xllm_service/chat_template/chat_template_factory.h

@@ -7,7 +7,7 @@
 #include <unordered_map>
 #include <unordered_set>
 
-#include "chat_template/chat_template.h"
+#include "chat_template/jinja_chat_template.h"
 #include "common/hash_util.h"
 #include "common/macros.h"
 #include "common/types.h"
@@ -68,7 +68,7 @@ class Scheduler {
   HttpServiceConfig http_config_;
 
   // chat template instance
-  std::unique_ptr<ChatTemplate> chat_template_;
+  std::unique_ptr<JinjaChatTemplate> chat_template_;
 
   std::shared_ptr<EtcdClient> etcd_client_;
 

xllm_service/rpc_service/scheduler.cpp

@@ -27,7 +27,7 @@ std::optional<std::string> load_chat_template_file(const std::string& dir) {
 }
 }  // namespace
 
-bool load_tokenizer_args(const std::string& model_weights_path,
+void load_tokenizer_args(const std::string& model_weights_path,
                          TokenizerArgs& tokenizer_args) {
   // tokenizer args from tokenizer_config.json
   JsonReader tokenizer_reader;
@@ -68,8 +68,6 @@ bool load_tokenizer_args(const std::string& model_weights_path,
       tokenizer_args.pad_token() = v.value();
     }
   }
-
-  return true;
 }
 
 }  // namespace xllm_service

xllm_service/rpc_service/scheduler.h

@@ -93,7 +93,7 @@ inline std::ostream& operator<<(std::ostream& os, const TokenizerArgs& args) {
   return os;
 }
 
-bool load_tokenizer_args(const std::string& model_weights_path,
+void load_tokenizer_args(const std::string& model_weights_path,
                          TokenizerArgs& tokenizer_args);
 
 }  // namespace xllm_service

xllm_service/tokenizer/tokenizer_args.cpp

@@ -2,28 +2,32 @@
 
 #include <filesystem>
 
+#include "tokenizer_args.h"
+
 namespace xllm_service {
 
 std::unique_ptr<Tokenizer> TokenizerFactory::create_tokenizer(
     const std::string& model_weights_path,
-    TokenizerArgs tokenizer_args) {
+    TokenizerArgs* tokenizer_args) {
+  load_tokenizer_args(model_weights_path, *tokenizer_args);
+
   const std::string tokenizer_json_path =
       model_weights_path + "/tokenizer.json";
   if (std::filesystem::exists(tokenizer_json_path)) {
     // 1. fast tokenizer
     LOG(INFO) << "Create fast tokenizer.";
     return std::make_unique<FastTokenizer>(tokenizer_json_path);
-  } else if (tokenizer_args.tokenizer_type() == "tiktoken" ||
-             tokenizer_args.tokenizer_class() == "TikTokenTokenizer") {
+  } else if (tokenizer_args->tokenizer_type() == "tiktoken" ||
+             tokenizer_args->tokenizer_class() == "TikTokenTokenizer") {
     // 2. create tiktoken tokenizer
     LOG(INFO) << "Create Tiktoken tokenizer.";
     return std::make_unique<TiktokenTokenizer>(model_weights_path,
-                                               tokenizer_args);
+                                               *tokenizer_args);
   } else {
     // 3. create sentencepiece tokenizer
     LOG(INFO) << "Create SentencePiece tokenizer.";
     return std::make_unique<SentencePieceTokenizer>(model_weights_path,
-                                                    tokenizer_args);
+                                                    *tokenizer_args);
   }
 }
 

xllm_service/tokenizer/tokenizer_args.h

@@ -11,7 +11,7 @@ class TokenizerFactory {
  public:
   static std::unique_ptr<Tokenizer> create_tokenizer(
       const std::string& model_weights_path,
-      TokenizerArgs tokenizer_args);
+      TokenizerArgs* tokenizer_args);
 };
 
 }  // namespace xllm_service