Remove VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION (#4860)

VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION is enabled by default for long
time. Let's remove it now.

- vLLM version: v0.12.0
- vLLM main:
vllm-project/vllm@ad32e3e

Signed-off-by: wangxiyuan <[email protected]>

Author: wangxiyuan

vllm_ascend/envs.py

@@ -92,11 +92,6 @@
     "VLLM_ASCEND_KV_CACHE_MEGABYTES_FLOATING_TOLERANCE":
     lambda: int(
         os.getenv("VLLM_ASCEND_KV_CACHE_MEGABYTES_FLOATING_TOLERANCE", 64)),
-    # Whether to enable the topk optimization. It's enabled by default. Please set to False if you hit any issue.
-    # We'll remove this flag in the future once it's stable enough.
-    "VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION":
-    lambda: bool(
-        int(os.getenv("VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION", '1'))),
     # Whether to enable mla_pa for deepseek mla decode, this flag will be removed after its available torch_npu is public accessible
     # and the mla_pa will be the default path of deepseek decode path.
     "VLLM_ASCEND_MLA_PA":

vllm_ascend/worker/model_runner_v1.py

@@ -141,6 +141,7 @@
 from vllm_ascend.platform import NPUPlatform
 from vllm_ascend.sample.logits_processor import build_logitsprocs
 from vllm_ascend.sample.rejection_sampler import AscendRejectionSampler
+from vllm_ascend.sample.sampler import AscendSampler
 from vllm_ascend.spec_decode import get_spec_decode_method
 from vllm_ascend.spec_decode.eagle_proposer import EagleProposer
 from vllm_ascend.spec_decode.interface import SpecDcodeType
@@ -312,15 +313,7 @@ def __init__(self, vllm_config: VllmConfig, device: torch.device):
         else:
             self.prefetch_stream = None
         self.dtype = self.model_config.dtype
-        if envs_ascend.VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION:
-            # TODO: drop the env config to use ascend sampler by default
-            from vllm_ascend.sample.sampler import AscendSampler
-
-            self.sampler = AscendSampler()
-        else:
-            from vllm.v1.sample.sampler import Sampler
-
-            self.sampler = Sampler()
+        self.sampler = AscendSampler()
         self.reorder_batch_threshold: Optional[int] = None
 
         # Lazy initialization, these will be set after __init__