[New model] Qwen3-next support (#2917)

### What this PR does / why we need it?
Add Qwen3-next support.

### Does this PR introduce _any_ user-facing change?
Yes, users can use Qwen3 next.
Related doc: https://github.com/vllm-project/vllm-ascend/pull/2916 the
tutorial will be ready in
[here](https://vllm-ascend.readthedocs.io/en/latest/tutorials/multi_npu_qwen3_next.html)

### How was this patch tested?
Doc CI passed

Related: https://github.com/vllm-project/vllm-ascend/issues/2884

Co-Authored-By: Angazenn <supperccell@163.com>
Co-Authored-By: zzzzwwjj <1183291235@qq.com>
Co-Authored-By: MengqingCao <cmq0113@163.com>
Co-Authored-By: linfeng-yuan <1102311262@qq.com>
Co-Authored-By: hust17yixuan <303660421@qq.com>
Co-Authored-By: SunnyLee219 <3294305115@qq.com>
Co-Authored-By: maoxx241 <maoxx241@umn.edu>


- vLLM version: v0.10.2
- vLLM main:
b834b4cbf1

---------

Signed-off-by: MengqingCao <cmq0113@163.com>
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Signed-off-by: Angazenn <supperccell@163.com>
Signed-off-by: Your Name <you@example.com>
Signed-off-by: zzzzwwjj <1183291235@qq.com>
Signed-off-by: linfeng-yuan <1102311262@qq.com>
Signed-off-by: hust17yixuan <303660421@qq.com>
Co-authored-by: MengqingCao <cmq0113@163.com>
Co-authored-by: Angazenn <supperccell@163.com>
Co-authored-by: Your Name <you@example.com>
Co-authored-by: zzzzwwjj <1183291235@qq.com>
Co-authored-by: linfeng-yuan <1102311262@qq.com>
Co-authored-by: hust17yixuan <303660421@qq.com>

vllm_ascend/attention/attention_v1.py

@@ -17,7 +17,7 @@
 
 from dataclasses import dataclass
 from enum import Enum
-from typing import List, Optional, Tuple, Type
+from typing import ClassVar, List, Optional, Tuple, Type
 
 import torch
 import torch.nn as nn
@@ -32,12 +32,12 @@ from vllm.distributed.kv_transfer import (get_kv_transfer_group,
 from vllm.forward_context import ForwardContext, get_forward_context
 from vllm.utils import cdiv, direct_register_custom_op
 from vllm.v1.core.sched.output import SchedulerOutput
+from vllm.v1.kv_cache_interface import AttentionSpec
 
 from vllm_ascend.attention.utils import AscendCommonAttentionMetadata
 from vllm_ascend.ops.attention import vanilla_chunked_prefill
 from vllm_ascend.utils import (ACL_FORMAT_FRACTAL_NZ, aligned_16, is_310p,
                                nd_to_nz_2d, nd_to_nz_spec)
-from vllm_ascend.worker.npu_input_batch import InputBatch
 
 
 def wait_for_kv_layer_from_connector(layer_name: str):
@@ -145,6 +145,10 @@ class AscendAttentionBackend(AttentionBackend):
             key_caches[dst_indices] = key_caches[src_indices]
             value_caches[dst_indices] = value_caches[src_indices]
 
+    @staticmethod
+    def get_supported_block_size() -> list[int]:
+        return [64]
+
 
 class AscendAttentionState(Enum):
     PrefillNoCache = 0
@@ -193,24 +197,29 @@ class AscendMetadata:
 
 
 class AscendAttentionMetadataBuilder:
+    reorder_batch_threshold: ClassVar[int] = 1
 
     def __init__(
         self,
+        kv_cache_spec: AttentionSpec,
+        layer_names: list[str],
         vllm_config: VllmConfig,
         device: torch.device,
     ):
         self.vllm_config = vllm_config
         self.model_config = vllm_config.model_config
         self.device = device
-        self.max_num_blocks_per_req = cdiv(self.model_config.max_model_len,
-                                           vllm_config.cache_config.block_size)
+        self.max_num_blocks_per_req = cdiv(
+            self.model_config.max_model_len,
+            AscendAttentionBackend.get_supported_block_size()[0])
 
-    def reorder_batch(self, input_batch: "InputBatch",
+    def reorder_batch(self, input_batch,
                       scheduler_output: "SchedulerOutput") -> bool:
         return False
 
     def build(
         self,
+        common_prefix_len: int,
         common_attn_metadata: AscendCommonAttentionMetadata,
         model: nn.Module,
     ):
@@ -219,11 +228,7 @@ class AscendAttentionMetadataBuilder:
         query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu[:
                                                                        num_reqs
                                                                        + 1]
-
         block_table = common_attn_metadata.block_table_tensor
-        block_table[:num_reqs, :self.max_num_blocks_per_req] = (
-            block_table[:num_reqs])
-
         query_lens = query_start_loc_cpu[1:] - query_start_loc_cpu[:-1]
         seq_lens = common_attn_metadata.seq_lens_cpu[:num_reqs]
         slot_mapping = common_attn_metadata.slot_mapping_cpu[:
@@ -574,6 +579,8 @@ def unified_ascend_attention_with_output(
     wait_for_kv_layer_from_connector(layer_name)
     forward_context: ForwardContext = get_forward_context()
     attn_metadata = forward_context.attn_metadata
+    if isinstance(attn_metadata, dict):
+        attn_metadata = attn_metadata[layer_name]
     self = forward_context.no_compile_layers[layer_name]
     kv_cache = self.kv_cache[forward_context.virtual_engine]
     self.impl.forward(self,

vllm_ascend/attention/mla_v1.py

@@ -171,6 +171,8 @@ class AscendMLAMetadataBuilder:
 
     # _attn_mask_builder = None
     def __init__(self,
+                 kv_cache_spec,
+                 layer_names,
                  vllm_config: VllmConfig,
                  device: torch.device,
                  metadata_cls: Optional[AscendMLAMetadata] = None):
@@ -265,6 +267,7 @@ class AscendMLAMetadataBuilder:
 
     def build(
         self,
+        common_prefix_len: int,
         common_attn_metadata: AscendCommonAttentionMetadata,
         model: nn.Module,
     ) -> AscendMLAMetadata:

vllm_ascend/attention/utils.py

@@ -21,6 +21,13 @@ class AscendCommonAttentionMetadata:
     """(batch_size,), the length of each request including both computed tokens
     and newly scheduled tokens"""
 
+    seq_lens: torch.Tensor
+    """same to seq_lens_cpu, for compatibility with some new attn metadata
+    (such as GDN)."""
+
+    num_computed_tokens_cpu: torch.Tensor
+    """(batch_size,), the number of computed tokens for each request"""
+
     num_reqs: int
     """Number of requests"""
     num_actual_tokens: int