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

[Model][1/N] Delete deepseek v2/v3 modeling codes. (#3189)

Browse Source 
This PR deletes model codes of deepseek_v2 and deepseek_v3 to reuse the
model file from vLLM.

vLLM Ascend now uses custom ops register way instead of model file
hard-coding.

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

---------

Signed-off-by: whx-sjtu <2952154980@qq.com>
This commit is contained in:
whx
2025-10-20 15:31:34 +08:00
committed by GitHub
parent 918ded9155
commit f8b52fe950
13 changed files with 143 additions and 843 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm_ascend/quantization/quant_config.py
View File

@@ -176,12 +176,14 @@ packed_modules_model_mapping = {
"deepseek_v2": {
"gate_up_proj": ["gate_proj", "up_proj"],
"experts":
["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"]
["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"],
"fused_qkv_a_proj": ["q_a_proj", "kv_a_proj_with_mqa"]
},
"deepseek_v3": {
"gate_up_proj": ["gate_proj", "up_proj"],
"experts":
["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"]
["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"],
"fused_qkv_a_proj": ["q_a_proj", "kv_a_proj_with_mqa"]
},
"deepseek_v32": {
"gate_up_proj": ["gate_proj", "up_proj"],
@@ -274,6 +276,7 @@ class AscendLinearMethod(LinearMethodBase):
# disable warning
param.ignore_warning = True
layer.register_parameter(pertensor_name, param)
param.weight_loader = extra_weight_attrs.get("weight_loader")
perchannel_dict = self.quant_method.get_perchannel_param(
output_size_per_partition, params_dtype)