[TRTLLM-12669][chore] simplify populate_sampling_params_for_one_model docstring

Signed-off-by: ZhaoyangWang <zhaoyangw@nvidia.com>

Author: zhaoyangwang-nvidia

tensorrt_llm/_torch/speculative/interface.py

@@ -522,17 +522,9 @@ def populate_sampling_params_for_one_model(
         """
         Set up topp/topk/temperatures for 1-model sampler.
 
-        Two-phase implementation:
-          Phase 1: scan request sampling configs to compute skip_*/has_greedy
-                   flags and decide whether the batch is all-greedy. No per-token
-                   list construction.
-          Phase 2 (only when is_all_greedy_sample=False): build per-token /
-                   per-request lists and issue 6 H->D copies into the sampling
-                   parameter buffers.
-
-        For all-greedy batches we skip phase 2 entirely (no list building, no
-        H->D copies). This is the common case when callers do not pass
-        temperature/top_k/top_p.
+        Scans sampling configs to set skip_*/is_all_greedy_sample flags. When
+        any request needs sampling, also builds per-token/per-request lists
+        and copies them to GPU buffers; all-greedy batches skip this entirely.
         """
         from tensorrt_llm._torch.pyexecutor.llm_request import LlmRequestState
         from tensorrt_llm.sampling_params import SamplingParams