EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Bugfix][kvcache] revert multiple kv cache groups (#923)

Browse Source 
Revert multiple kv cache groups related changes as this feature is
reverted in vllm https://github.com/vllm-project/vllm/pull/18459

---------

Signed-off-by: MengqingCao <cmq0113@163.com>
This commit is contained in:
Mengqing Cao
2025-05-22 15:15:33 +08:00
committed by GitHub
parent b4d6672d01
commit 7aa4f85f10
3 changed files with 34 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
vllm_ascend/worker/model_runner_v1.py
View File

@@ -164,6 +164,24 @@ class NPUModelRunner:
raise NotImplementedError(
"Non-Attention backend is not supported by V1 NPUModelRunner.")
self.attn_backend = get_attn_backend(
self.head_size,
self.dtype,
self.kv_cache_dtype,
self.block_size,
self.model_config.is_attention_free,
use_mla=self.model_config.use_mla,
)
if self.attn_backend is None:
error_msg = (
f"Error with get_att_backend: {self.head_size=}, "
f"{self.dtype=}, {self.kv_cache_dtype=}, {self.block_size=}, "
f"{self.model_config.is_attention_free=}, "
f"{self.model_config.use_mla=}")
logger.error(error_msg)
raise NotImplementedError(
"Non-Attention backend is not supported by V1 GPUModelRunner.")
self.attn_metadata_builder = self.attn_backend.get_builder_cls()(
weakref.proxy(self))
@@ -196,6 +214,17 @@ class NPUModelRunner:
pin_memory=True,
vocab_size=self.model_config.get_vocab_size(),
)
else:
self.input_batch = InputBatch(
max_num_reqs=self.max_num_reqs,
max_model_len=self.model_config.max_model_len,
max_num_blocks_per_req=self.max_num_blocks_per_req,
max_num_batched_tokens=self.max_num_tokens,
device=self.device,
pin_memory=True,
vocab_size=self.model_config.get_vocab_size(),
)
self.input_ids = torch.zeros(self.max_num_tokens,
dtype=torch.int32,
device=self.device)
@@ -542,10 +571,7 @@ class NPUModelRunner:
block_table_indices = (req_indices * self.max_num_blocks_per_req +
positions_np // self.block_size)
if vllm_version_is("0.8.5") or vllm_version_is("0.8.5.post1"):
block_table_cpu = self.input_batch.block_table.get_cpu_tensor()
else:
block_table_cpu = self.input_batch.block_table[0].get_cpu_tensor()
block_table_cpu = self.input_batch.block_table.get_cpu_tensor()
block_numbers = block_table_cpu.flatten()[block_table_indices].numpy()
block_offsets = positions_np % self.block_size
np.add(block_numbers * self.block_size,
@@ -960,16 +986,6 @@ class NPUModelRunner:
"""
import torch_npu
kv_caches: Dict[str, torch.Tensor] = {}
if not (vllm_version_is("0.8.5") or vllm_version_is("0.8.5.post1")):
self.input_batch = InputBatch(
max_num_reqs=self.max_num_reqs,
max_model_len=self.model_config.max_model_len,
max_num_batched_tokens=self.max_num_tokens,
device=self.device,
pin_memory=True,
vocab_size=self.model_config.get_vocab_size(),
kv_cache_config=kv_cache_config,
)
for kv_cache_group in kv_cache_config.kv_cache_groups:
kv_cache_spec = kv_cache_group.kv_cache_spec