[Model][3/N] Refactor sfa into mla and remove deepseek_v3_2.py (#3769)

This is the follow-up PR to PR #3189, which continues to refactor sfa
into mla and finally remove deepseek_v3_2.py. This is the last PR of
deepseek modeling refactoring. After this, all deepseek-related model
codes are removed from vllm_ascend.

FurtherMore, after this PR deepseek v3.2 can run chunk-prefill with
correct accuracy.

- vLLM version: v0.11.0rc3
- vLLM main:
83f478bb19

---------

Signed-off-by: whx-sjtu <2952154980@qq.com>
This commit is contained in:
whx
2025-10-30 17:06:38 +08:00
committed by GitHub
parent eff3e5fc6f
commit f6149f3894
10 changed files with 751 additions and 1935 deletions

View File

@@ -29,10 +29,6 @@ def register_model():
"vllm_ascend.models.qwen2_5_vl_without_padding:AscendQwen2_5_VLForConditionalGeneration_Without_Padding"
)
ModelRegistry.register_model(
"DeepseekV32ForCausalLM",
"vllm_ascend.models.deepseek_v3_2:CustomDeepseekV3ForCausalLM")
# There is no PanguProMoEForCausalLM in vLLM, so we should register it before vLLM config initialization
# to make sure the model can be loaded correctly. This register step can be removed once vLLM support PanguProMoEForCausalLM.
ModelRegistry.register_model(