[Refactor]Refactor sampler (#2050)

Refactor Sampler implementation from patch way to inherit from vLLM
Sampler interface.

Next step: Make the op `TopKTopPSampler` in vLLM support custom ops
register mechanism

- vLLM version: v0.10.0
- vLLM main:
61a6905ab0

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

vllm_ascend/patch/worker/patch_common/__init__.py

@@ -21,4 +21,3 @@ import vllm_ascend.patch.worker.patch_common.patch_utils  # noqa isort:skip
 import vllm_ascend.patch.worker.patch_common.patch_distributed  # noqa
 import vllm_ascend.patch.worker.patch_common.patch_linear  # noqa
 import vllm_ascend.patch.worker.patch_common.patch_minicpm  # noqa
-import vllm_ascend.patch.worker.patch_common.patch_sampler  # noqa

vllm_ascend/patch/worker/patch_common/patch_sampler.py

@@ -1,83 +0,0 @@
-#
-# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
-# SPDX-License-Identifier: Apache-2.0
-# This file is a part of the vllm-ascend project.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-
-from typing import Optional
-
-import torch
-import torch_npu
-from vllm.v1.sample.ops.topk_topp_sampler import TopKTopPSampler, random_sample
-
-from vllm_ascend import envs
-
-
-def apply_top_k_top_p(
-    logits: torch.Tensor,
-    k: torch.Tensor,
-    p: torch.Tensor,
-) -> torch.Tensor:
-    if p is not None and k is not None:
-        # npu_top_k_top_p's parameter order is (logits, p, k), not (logits, k, p)
-        return torch_npu.npu_top_k_top_p(logits, p, k)
-
-    probs = logits.softmax(dim=-1)
-    probs_sort, _ = probs.sort(dim=-1, descending=False)
-
-    if k is not None:
-        top_k_count = probs_sort.size(1) - k.to(torch.long)  # shape: (batch, )
-        top_k_count = top_k_count.unsqueeze(dim=1)
-        top_k_cutoff = probs_sort.gather(-1, top_k_count)
-
-        # Make sure the no top-k rows are no-op.
-        no_top_k_mask = (k == logits.shape[1]).unsqueeze(dim=1)
-        top_k_cutoff.masked_fill_(no_top_k_mask, -float("inf"))
-
-        elements_to_discard = probs < top_k_cutoff
-        logits.masked_fill_(elements_to_discard, -float("inf"))
-
-    if p is not None:
-        cumprob = torch.cumsum(probs_sort, dim=-1)
-        top_p_mask = cumprob <= 1 - p.unsqueeze(dim=1)
-        top_p_mask[:, -1] = False  # at least one
-
-        top_p_count = top_p_mask.sum(dim=-1).unsqueeze(1)
-        top_p_cutoff = probs_sort.gather(-1, top_p_count)
-        elements_to_discard = probs < top_p_cutoff
-        logits.masked_fill_(elements_to_discard, -float("inf"))
-
-    return logits
-
-
-def topk_topp_forward_native(
-    self,
-    logits: torch.Tensor,
-    generators: dict[int, torch.Generator],
-    k: Optional[torch.Tensor],
-    p: Optional[torch.Tensor],
-) -> torch.Tensor:
-    """
-    PyTorch-native implementation of top-k and top-p sampling.
-
-    The logits tensor may be updated in-place.
-    """
-    logits = apply_top_k_top_p(logits, k, p)
-    probs = logits.softmax(dim=-1, dtype=torch.float32)
-    return random_sample(probs, generators)
-
-
-if envs.VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION:
-    TopKTopPSampler.forward_native = topk_topp_forward_native