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

[Quickfix] update CachedRequestState as NewRequestData changed (#2367)

Browse Source 
### What this PR does / why we need it?
1. update `CachedRequestState` as `NewRequestData` changed in
https://github.com/vllm-project/vllm/pull/22570
2. drop maintenance of vllm v0.10.0 in the branch main

### Does this PR introduce _any_ user-facing change?
N/A

### How was this patch tested?
CI passed with existing test.


- vLLM version: v0.10.0
- vLLM main:
92ff41abea

---------

Signed-off-by: MengqingCao <cmq0113@163.com>
This commit is contained in:
Mengqing Cao
2025-08-15 07:35:27 +08:00
committed by GitHub
parent 2ad7e1251e
commit 61866b8ac6
18 changed files with 77 additions and 285 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
vllm_ascend/models/qwen2_5_vl.py
View File

@@ -30,8 +30,7 @@ from transformers.models.qwen2_5_vl.configuration_qwen2_5_vl import (
from vllm.config import VllmConfig
from vllm.distributed import parallel_state
from vllm.distributed import utils as dist_utils
from vllm.model_executor.layers.activation import (_ACTIVATION_REGISTRY,
get_act_and_mul_fn)
from vllm.model_executor.layers.activation import get_act_and_mul_fn
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
@@ -43,8 +42,6 @@ from vllm.model_executor.models.qwen2_5_vl import (
from vllm.model_executor.models.utils import maybe_prefix
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm_ascend.utils import vllm_version_is
MIN_PAD_SIZE = 64 # min_size to pad weight
MAX_PAD_SIZE = 128 # max_size to pad weight
@@ -202,8 +199,6 @@ class AscendQwen2_5_VisionTransformer(Qwen2_5_VisionTransformer):
)
act_fn = get_act_and_mul_fn(vision_config.hidden_act)
if vllm_version_is("0.10.0"):
act_fn = _ACTIVATION_REGISTRY[vision_config.hidden_act]
self.blocks = nn.ModuleList([
AscendQwen2_5_VisionBlock(
dim=self.hidden_size,
@@ -303,12 +298,9 @@ class AscendQwen2_5_VisionTransformer(Qwen2_5_VisionTransformer):
("qkv_proj", "q_proj", "q"),
("qkv_proj", "k_proj", "k"),
("qkv_proj", "v_proj", "v"),
("mlp.gate_up_proj.", "mlp.gate_proj.", 0),
("mlp.gate_up_proj.", "mlp.up_proj.", 1),
]
if not vllm_version_is("0.10.0"):
stacked_params_mapping.extend([
("mlp.gate_up_proj.", "mlp.gate_proj.", 0),
("mlp.gate_up_proj.", "mlp.up_proj.", 1),
])
params_dict = dict(self.named_parameters(remove_duplicate=False))
loaded_params: Set[str] = set()
for name, loaded_weight in weights:

6
vllm_ascend/models/qwen2_5_vl_without_padding.py
View File

@@ -30,8 +30,7 @@ from transformers.models.qwen2_5_vl.configuration_qwen2_5_vl import (
from vllm.config import VllmConfig
from vllm.distributed import parallel_state
from vllm.distributed import utils as dist_utils
from vllm.model_executor.layers.activation import (_ACTIVATION_REGISTRY,
get_act_and_mul_fn)
from vllm.model_executor.layers.activation import get_act_and_mul_fn
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.models.qwen2_5_vl import (
@@ -43,7 +42,6 @@ from vllm.model_executor.models.utils import maybe_prefix
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm_ascend.models.qwen2_5_vl import AscendQwen2_5_VisionRotaryEmbedding
from vllm_ascend.utils import vllm_version_is
class AscendQwen2_5_VisionAttention_Without_Padding(Qwen2_5_VisionAttention):
@@ -175,8 +173,6 @@ class AscendQwen2_5_VisionTransformer_Without_Padding(Qwen2_5_VisionTransformer
)
act_fn = get_act_and_mul_fn(vision_config.hidden_act)
if vllm_version_is("0.10.0"):
act_fn = _ACTIVATION_REGISTRY[vision_config.hidden_act]
self.blocks = nn.ModuleList([
AscendQwen2_5_VisionBlock_Without_Padding(
dim=self.hidden_size,