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

Add DeepSeek V3.2 support (#3270)

Browse Source 
### What this PR does / why we need it?

This PR added the initial DeepSeek V3.2 support with [vLLM
v0.11.0](https://github.com/vllm-project/vllm/tree/releases/v0.11.0)
(not released yet). We will complete vLLM adaptation as soon as
possible. This feature will be ready in recent 1-2 days.

Related doc: https://github.com/vllm-project/vllm-ascend/pull/3223 .

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

### How was this patch tested?
CI passed and Run deepseek doc soon.


- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/releases/v0.11.0

---------

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Signed-off-by: zzzzwwjj <1183291235@qq.com>
Signed-off-by: linfeng-yuan <1102311262@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
Signed-off-by: MengqingCao <cmq0113@163.com>
Co-authored-by: zzzzwwjj <1183291235@qq.com>
Co-authored-by: linfeng-yuan <1102311262@qq.com>
Co-authored-by: wxsIcey <1790571317@qq.com>
Co-authored-by: MengqingCao <cmq0113@163.com>
This commit is contained in:
wangxiyuan
2025-09-30 03:25:58 +08:00
committed by GitHub
parent 5503a3142f
commit 81bd6e4c99
27 changed files with 4354 additions and 70 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
vllm_ascend/platform.py
View File

@@ -300,6 +300,7 @@ class NPUPlatform(Platform):
block_size,
use_v1,
use_mla,
use_sfa,
has_sink=False):
if not use_v1:
raise ValueError("vLLM Ascend does not support V0 engine.")
@@ -307,21 +308,28 @@ class NPUPlatform(Platform):
ascend_config = get_ascend_config()
if use_mla and ascend_config.enable_shared_expert_dp:
return "vllm_ascend.torchair.torchair_mla.AscendMLATorchairBackend"
if use_mla and not use_sfa:
return "vllm_ascend.torchair.torchair_mla.AscendMLATorchairBackend"
if use_mla and use_sfa:
return "vllm_ascend.torchair.torchair_sfa.AscendSFATorchairBackend"
use_torchair = ascend_config.torchair_graph_config.enabled
# choose attention backend based on use_mla and use_torchair
backend_map = {
(True, True):
(True, False, True):
"vllm_ascend.torchair.torchair_mla.AscendMLATorchairBackend",
(True, False):
(True, False, False):
"vllm_ascend.attention.mla_v1.AscendMLABackend",
(False, True):
(False, False, True):
"vllm_ascend.torchair.torchair_attention.AscendAttentionTorchairBackend",
(False, False):
"vllm_ascend.attention.attention_v1.AscendAttentionBackend"
(False, False, False):
"vllm_ascend.attention.attention_v1.AscendAttentionBackend",
(True, True, False):
"vllm_ascend.attention.sfa_v1.AscendSFABackend",
(True, True, True):
"vllm_ascend.torchair.torchair_sfa.AscendSFATorchairBackend",
}
return backend_map[(use_mla, use_torchair)]
return backend_map[(use_mla, use_sfa, use_torchair)]
@classmethod
def get_punica_wrapper(cls) -> str: