init v0.23.0

Signed-off-by: Sun Ruoxi <sunruoxi@4paradigm.com>

vllm_ascend/_310p/npu_input_batch.py

@@ -0,0 +1,59 @@
+import torch
+from vllm.v1.kv_cache_interface import KVCacheGroupSpec
+from vllm.v1.sample.logits_processor import LogitsProcessors
+
+from vllm_ascend._310p.block_table import MultiGroupBlockTable
+from vllm_ascend.worker.npu_input_batch import NPUInputBatch
+
+
+class NPUInputBatch310(NPUInputBatch):
+    def __init__(
+        self,
+        max_num_reqs: int,
+        max_model_len: int,
+        max_num_batched_tokens: int,
+        device: torch.device,
+        pin_memory: bool,
+        vocab_size: int,
+        block_sizes: list[int],
+        kernel_block_sizes: list[list[int]],
+        max_num_blocks_per_req: list[int] | None = None,
+        logitsprocs: LogitsProcessors | None = None,
+        logitsprocs_need_output_token_ids: bool = False,
+        is_spec_decode: bool = False,
+        is_pooling_model: bool = False,
+        num_speculative_tokens: int = 0,
+        cp_kv_cache_interleave_size: int = 1,
+        kv_cache_groups: list[KVCacheGroupSpec] | None = None,
+    ):
+        super().__init__(
+            max_num_reqs=max_num_reqs,
+            max_model_len=max_model_len,
+            max_num_batched_tokens=max_num_batched_tokens,
+            device=device,
+            pin_memory=pin_memory,
+            vocab_size=vocab_size,
+            block_sizes=block_sizes,
+            kernel_block_sizes=kernel_block_sizes,
+            max_num_blocks_per_req=max_num_blocks_per_req,
+            logitsprocs=logitsprocs,
+            logitsprocs_need_output_token_ids=logitsprocs_need_output_token_ids,
+            is_spec_decode=is_spec_decode,
+            is_pooling_model=is_pooling_model,
+            num_speculative_tokens=num_speculative_tokens,
+            cp_kv_cache_interleave_size=cp_kv_cache_interleave_size,
+            kv_cache_groups=kv_cache_groups,
+        )
+        self.block_table = MultiGroupBlockTable(
+            max_num_reqs=max_num_reqs,
+            max_model_len=max_model_len,
+            max_num_batched_tokens=max_num_batched_tokens,
+            pin_memory=pin_memory,
+            device=device,
+            block_sizes=block_sizes,
+            max_num_blocks=max_num_blocks_per_req,
+            num_speculative_tokens=num_speculative_tokens,
+            kernel_sizes=kernel_block_sizes,
+            cp_kv_cache_interleave_size=cp_kv_cache_interleave_size,
+            kv_cache_groups=kv_cache_groups,
+        )