common: Refactoring sampler parameters (#20429) (#22233)

This change refactors the reasoning_budget_message parameter from the
common params into the sampling parameters specifically. It also removes
the reasoning_budget common parameter and standardizes on the existing
reasoning_budget_tokens parameter in the sampling configuration.

Issue: https://github.com/ggml-org/llama.cpp/issues/20429
Original PR: https://github.com/ggml-org/llama.cpp/pull/20297

common/arg.cpp

@@ -3122,14 +3122,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         "token budget for thinking: -1 for unrestricted, 0 for immediate end, N>0 for token budget (default: -1)",
         [](common_params & params, int value) {
             if (value < -1) { throw std::invalid_argument("invalid value"); }
-            params.reasoning_budget = value;
+            params.sampling.reasoning_budget_tokens = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_COMPLETION, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_THINK_BUDGET"));
     add_opt(common_arg(
         {"--reasoning-budget-message"}, "MESSAGE",
         "message injected before the end-of-thinking tag when reasoning budget is exhausted (default: none)",
         [](common_params & params, const std::string & value) {
-            params.reasoning_budget_message = value;
+            params.sampling.reasoning_budget_message = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_COMPLETION, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_THINK_BUDGET_MESSAGE"));
     add_opt(common_arg(

common/common.h

@@ -274,6 +274,7 @@ struct common_params_sampling {
     std::vector<llama_token> reasoning_budget_start;           // start tag token sequence
     std::vector<llama_token> reasoning_budget_end;             // end tag token sequence
     std::vector<llama_token> reasoning_budget_forced;          // forced sequence (message + end tag)
+    std::string              reasoning_budget_message;         // message injected before end tag when budget exhausted
 
     bool backend_sampling = false;
 
@@ -581,8 +582,6 @@ struct common_params {
     bool force_pure_content_parser = false;
     common_reasoning_format reasoning_format = COMMON_REASONING_FORMAT_DEEPSEEK;
     int enable_reasoning = -1; // -1 = auto, 0 = disable, 1 = enable
-    int reasoning_budget = -1;
-    std::string reasoning_budget_message; // message injected before end tag when budget exhausted
     bool prefill_assistant = true; // if true, any trailing assistant message will be prefilled into the response
     int sleep_idle_seconds = -1;   // if >0, server will sleep after this many seconds of idle time
 

tools/cli/cli.cpp

@@ -77,8 +77,8 @@ struct cli_context {
         // defaults.return_progress = true; // TODO: show progress
 
         verbose_prompt = params.verbose_prompt;
-        reasoning_budget = params.reasoning_budget;
+        reasoning_budget = params.sampling.reasoning_budget_tokens;
-        reasoning_budget_message = params.reasoning_budget_message;
+        reasoning_budget_message = params.sampling.reasoning_budget_message;
     }
 
     std::string generate_completion(result_timings & out_timings) {

tools/server/server-context.cpp

@@ -1045,8 +1045,8 @@ private:
                 /* allow_image           */ mctx ? mtmd_support_vision(mctx) : false,
                 /* allow_audio           */ mctx ? mtmd_support_audio (mctx) : false,
                 /* enable_thinking       */ enable_thinking,
-                /* reasoning_budget      */ params_base.reasoning_budget,
+                /* reasoning_budget      */ params_base.sampling.reasoning_budget_tokens,
-                /* reasoning_budget_msg  */ params_base.reasoning_budget_message,
+                /* reasoning_budget_msg  */ params_base.sampling.reasoning_budget_message,
                 /* media_path            */ params_base.media_path,
                 /* force_pure_content    */ params_base.force_pure_content_parser
             };