common : support manually triggering the reasoning budget end sequence (#23949)

common/reasoning-budget.cpp

@@ -247,3 +247,24 @@ common_reasoning_budget_state common_reasoning_budget_get_state(const struct lla
     }
     return ((const common_reasoning_budget_ctx *)smpl->ctx)->state;
 }
+
+bool common_reasoning_budget_force(struct llama_sampler * smpl) {
+    if (!smpl) {
+        return false;
+    }
+
+    auto * ctx = (common_reasoning_budget_ctx *) smpl->ctx;
+
+    // only a sampler that is actively counting down the budget may be forced;
+    // any other state (idle, already forcing/waiting, or done) is left untouched
+    if (ctx->state != REASONING_BUDGET_COUNTING) {
+        return false;
+    }
+
+    ctx->state = REASONING_BUDGET_FORCING;
+    ctx->force_pos = 0;
+    ctx->end_matcher.reset();
+    LOG_INF("reasoning-budget: forced into forcing state (manual transition)\n");
+
+    return true;
+}

common/reasoning-budget.h

@@ -40,3 +40,7 @@ struct llama_sampler * common_reasoning_budget_init(
         common_reasoning_budget_state    initial_state = REASONING_BUDGET_IDLE);
 
 common_reasoning_budget_state common_reasoning_budget_get_state(const struct llama_sampler * smpl);
+
+// Manually transition the reasoning budget sampler into the FORCING state.
+// Returns true if the transition occurred.
+bool common_reasoning_budget_force(struct llama_sampler * smpl);

common/sampling.cpp

@@ -661,6 +661,14 @@ uint32_t common_sampler_get_seed(const struct common_sampler * gsmpl) {
     return llama_sampler_get_seed(gsmpl->chain);
 }
 
+bool common_sampler_reasoning_budget_force(struct common_sampler * gsmpl) {
+    if (!gsmpl) {
+        return false;
+    }
+
+    return common_reasoning_budget_force(gsmpl->rbudget);
+}
+
 // helpers
 
 llama_token_data_array * common_sampler_get_candidates(struct common_sampler * gsmpl, bool do_sort) {

common/sampling.h

@@ -87,6 +87,9 @@ std::vector<llama_token> common_sampler_sample_and_accept_n(struct common_sample
 
 uint32_t common_sampler_get_seed(const struct common_sampler * gsmpl);
 
+// force the reasoning budget sampler (if any) to begin forcing its end sequence now.
+bool common_sampler_reasoning_budget_force(struct common_sampler * gsmpl);
+
 // helpers
 
 // access the internal list of current candidate tokens