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

[HybridKV][Bugfix] Fix Hybrid kvcache sharing bug in same attention type (#3760)

Browse Source 
### What this PR does / why we need it?
Part of https://github.com/vllm-project/vllm-ascend/pull/3106
Fix Hybrid kvcache sharing bug in same attention type
Change the `shared_by` logic so that the same attention spec could share
the same buffer instead of allocating more hbm.
After this pr, kvcache memory saved 50% in qwen3-next compared with
before (`self_attn:linear_attn=1:3` in an `attn_group`), and
`gpu_memory_utilization` could increase to `0.8` on Qwen3-Next when
running on A2 64G/card with tp4

<img width="2833" height="1540" alt="image"
src="https://github.com/user-attachments/assets/2a91fa99-fb0f-447c-9e8b-acd587890fbe"
/>

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?
Test pass with the latest e2e test case on qwen3-next

- vLLM version: v0.11.0rc3
- vLLM main:
c9461e05a4

---------

Signed-off-by: MengqingCao <cmq0113@163.com>
This commit is contained in:
Mengqing Cao
2025-10-29 14:18:52 +08:00
committed by GitHub
parent 789ba4c5c2
commit 900086fdc6
2 changed files with 26 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/e2e/multicard/test_qwen3_next.py
View File

@@ -27,12 +27,12 @@ from tests.e2e.conftest import VllmRunner
def test_models_distributed_Qwen3_NEXT_TP4():
example_prompts = [
"Hello, my name is",
]
] * 4
max_tokens = 5
with VllmRunner("Qwen/Qwen3-Next-80B-A3B-Instruct",
tensor_parallel_size=4,
max_model_len=4096,
gpu_memory_utilization=0.7,
gpu_memory_utilization=0.8,
distributed_executor_backend="mp",
enforce_eager=True) as vllm_model:
vllm_model.generate_greedy(example_prompts, max_tokens)

42
vllm_ascend/worker/model_runner_v1.py
View File

@@ -3225,25 +3225,26 @@ class NPUModelRunner(LoRAModelRunnerMixin):
# TODO: REFACTOR ME to sharing hybrid cache
for idx in range(len(kv_cache_tensor.shared_by)):
layer_name = kv_cache_tensor.shared_by[idx]
if "linear_attn" in layer_name:
if "linear_attn" in layer_name and layer_name not in kv_cache_raw_tensors.keys(
):
# for mamba linear attention
if self.vllm_config.kv_transfer_config is None:
tensor = torch.zeros(kv_cache_tensor.size,
dtype=torch.int8,
device=self.device)
else:
cache_size_aligned = kv_cache_tensor.size + alignment
tensor = torch.zeros(cache_size_aligned,
dtype=torch.int8,
device=self.device)
tensor = self._align_memory(
tensor, alignment)[:kv_cache_tensor.size]
for layer_name_inner in kv_cache_tensor.shared_by:
if ("attn" in layer_name_inner and "linear_attn" not in layer_name_inner) or \
layer_name_inner in kv_cache_raw_tensors.keys():
continue
if self.vllm_config.kv_transfer_config is None:
tensor = torch.zeros(kv_cache_tensor.size,
dtype=torch.int8,
device=self.device)
else:
cache_size_aligned = kv_cache_tensor.size + alignment
tensor = torch.zeros(cache_size_aligned,
dtype=torch.int8,
device=self.device)
tensor = self._align_memory(
tensor, alignment)[:kv_cache_tensor.size]
kv_cache_raw_tensors[layer_name_inner] = tensor
elif "attn" in layer_name:
# shared the kvcache between the linear_attn specs in the same group
if "linear_attn" in layer_name_inner:
kv_cache_raw_tensors[layer_name_inner] = tensor
elif "attn" in layer_name and layer_name not in kv_cache_raw_tensors.keys(
):
# for other attentions, e.g., self_attn, sliding window attn
if self.vllm_config.kv_transfer_config is None:
k_tensor = torch.zeros(kv_cache_tensor.size // 2,
@@ -3265,7 +3266,12 @@ class NPUModelRunner(LoRAModelRunnerMixin):
alignment)[:cache_size]
v_tensor = self._align_memory(v_tensor,
alignment)[:cache_size]
kv_cache_raw_tensors[layer_name] = (k_tensor, v_tensor)
for layer_name_inner in kv_cache_tensor.shared_by:
# shared the kvcache between the self_attn specs in the same group
if ("attn" in layer_name_inner
and "linear_attn" not in layer_name_inner):
kv_cache_raw_tensors[layer_name_inner] = (k_tensor,
v_tensor)
layer_names = set()
for group in kv_cache_config.kv_cache_groups: