change to cann version

Signed-off-by: Angazenn <supperccell@163.com>

Author: Angazenn

vllm_ascend/attention/attention_mask.py

@@ -50,8 +50,7 @@ def __init__(
         self._seq_len_cached = attn_mask.shape[0]
         self.attn_mask_cache = attn_mask
         self.device = device
-        if self.device:
-
+        if torch.version.cann.startswith("8.3"):
             assigned_mask_dim = 2048
             self.chunked_prefill_attn_mask = torch.triu(torch.ones(assigned_mask_dim, assigned_mask_dim), diagonal=1
             ).to(torch.int8).to(device)

vllm_ascend/utils.py

@@ -60,7 +60,6 @@
 _SLEEP_MODE_ENABLED = None
 _CURRENT_STREAM = None
 _ASCEND_CUSTOMOP_IS_REIGISTERED = False
-_CURRENT_TORCH_NPU_VERSION = torch_npu.__version__
 
 def is_310p():
     global _IS_310P

vllm_ascend/worker/model_runner_v1.py

@@ -112,7 +112,7 @@
 from vllm_ascend.utils import (ACL_FORMAT_FRACTAL_ND, ACL_FORMAT_FRACTAL_NZ,
                                AscendSocVersion, ProfileExecuteDuration,
                                get_ascend_soc_version, is_310p,
-                               lmhead_tp_enable, verify_torch_npu_version)
+                               lmhead_tp_enable)
 from vllm_ascend.worker.npu_input_batch import CachedRequestState, InputBatch
 
 if TYPE_CHECKING:
@@ -301,10 +301,7 @@ def __init__(self, vllm_config: VllmConfig, device: torch.device):
             use_mla=self.model_config.use_mla,
         )
 
-        pta_version_support_compressed_mask = "2.7.1.dev20250918"
-        self.compressed_mask = verify_torch_npu_version(pta_version_support_compressed_mask, "compressed mask")
-
-        if self.compressed_mask:
+        if torch.version.cann.startswith("8.3"):
             self.attn_mask_builder = AttentionMaskBuilder(
                 self.scheduler_config.max_num_batched_tokens, self.dtype, self.device)
         else:
@@ -825,7 +822,7 @@ def _make_attention_mask(self, seq_lens, position,
                              attn_state) -> torch.Tensor:
         # Chunk Prefill situation.
         if attn_state == AscendAttentionState.ChunkedPrefill and not self.vllm_config.model_config.use_mla:
-            if self.compressed_mask:
+            if torch.version.cann.startswith("8.3"):
                 return self.attn_mask_builder.get_splitfuse_attn_mask()
             else:
                 return self.attn_mask_builder.get_splitfuse_attn_mask(