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

[Misc] Clean up useless patch (#3320)

Browse Source 
### What this PR does / why we need it?
1. clean up v0.10.2 support in ut and e2e test
2. remove v0.11.0 period job, we're at v0.11.0 now.
3. remove uesless patch for deepseek v3.2. They have been done in vLLM
already.
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?


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

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-10-09 14:07:26 +08:00
committed by GitHub
parent a43e2f61e1
commit 1c5b302f0d
10 changed files with 29 additions and 412 deletions
Download Patch File Download Diff File Expand all files Collapse all files

79
vllm_ascend/patch/platform/patch_common/patch_config.py
View File

@@ -1,87 +1,10 @@
import ast
import vllm.envs as envs
from transformers import PretrainedConfig
from vllm.config import ModelConfig
from vllm.config.speculative import SpeculativeConfig
from vllm.logger import logger
# mypy: ignore-errors
@property
def is_deepseek_mla(self: ModelConfig):
if not hasattr(self.hf_text_config, "model_type"):
return False
elif self.hf_text_config.model_type in \
('deepseek_v2', 'deepseek_v3', 'deepseek_mtp',
'kimi_k2', 'longcat_flash', 'deepseek_v32'):
return self.hf_text_config.kv_lora_rank is not None
elif self.hf_text_config.model_type == 'eagle':
# if the model is an EAGLE module, check for the
# underlying architecture
return self.hf_text_config.model.model_type in \
('deepseek_v2', 'deepseek_v3', 'deepseek_v32') \
and self.hf_text_config.kv_lora_rank is not None
return False
@staticmethod
def hf_config_override(hf_config: PretrainedConfig) -> PretrainedConfig:
if hf_config.model_type in ("deepseek_v3", "deepseek_v32"):
hf_config.model_type = "deepseek_mtp"
if hf_config.model_type == "deepseek_mtp":
n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
hf_config.update({
"n_predict": n_predict,
"architectures": ["DeepSeekMTPModel"]
})
if hf_config.architectures[0] == "MiMoForCausalLM":
hf_config.model_type = "mimo_mtp"
n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
hf_config.update({
"num_hidden_layers": 0,
"n_predict": n_predict,
"architectures": ["MiMoMTPModel"]
})
if hf_config.architectures[0] == "Glm4MoeForCausalLM":
hf_config.model_type = "glm4_moe_mtp"
n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
hf_config.update({
"num_hidden_layers": 0,
"n_predict": n_predict,
"architectures": ["Glm4MoeMTPModel"]
})
if hf_config.model_type == "ernie4_5_moe":
hf_config.model_type = "ernie_mtp"
if hf_config.model_type == "ernie_mtp":
n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
hf_config.update({
"n_predict": n_predict,
"architectures": ["ErnieMTPModel"]
})
if hf_config.model_type == "qwen3_next":
hf_config.model_type = "qwen3_next_mtp"
if hf_config.model_type == "qwen3_next_mtp":
n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
hf_config.update({
"n_predict": n_predict,
"architectures": ["Qwen3NextMTP"]
})
if hf_config.model_type == "longcat_flash":
hf_config.model_type = "longcat_flash_mtp"
n_predict = getattr(hf_config, "num_nextn_predict_layers", 1)
hf_config.update({
"n_predict": n_predict,
"architectures": ["LongCatFlashMTPModel"]
})
return hf_config
def __post_init__(self):
# Note: "method" is a new parameter that helps to extend the
@@ -308,6 +231,4 @@ def __post_init__(self):
self.draft_tensor_parallel_size))
ModelConfig.is_deepseek_mla = is_deepseek_mla
SpeculativeConfig.__post_init__ = __post_init__
SpeculativeConfig.hf_config_override = hf_config_override