biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Support other quantization methods in glm41v (#36321)

Browse Source 
Signed-off-by: g00887675/loganJane <g00887675/loganJane73@hotmail.com>
Co-authored-by: g00887675/loganJane <g00887675/loganJane73@hotmail.com>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
LoganJane
2026-03-11 17:48:05 +08:00
committed by GitHub
parent e661b9ee83
commit 545d18d81b
1 changed files with 6 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/model_executor/models/glm4_1v.py
View File

@@ -63,6 +63,9 @@ from vllm.model_executor.layers.linear import (
RowParallelLinear, RowParallelLinear,
) )
from vllm.model_executor.layers.quantization import QuantizationConfig from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.quantization.compressed_tensors import (
compressed_tensors,
)
from vllm.model_executor.layers.rotary_embedding import get_rope from vllm.model_executor.layers.rotary_embedding import get_rope
from vllm.model_executor.layers.rotary_embedding.common import ( from vllm.model_executor.layers.rotary_embedding.common import (
ApplyRotaryEmb, ApplyRotaryEmb,
@@ -280,7 +283,9 @@ class Glm4vVisionAttention(nn.Module):
bias=False, bias=False,
quant_config=quant_config, quant_config=quant_config,
# Change qkv prefix to align with GLM-4.5V-FP8 quantization cfg # Change qkv prefix to align with GLM-4.5V-FP8 quantization cfg
prefix=f"{prefix}.qkv_proj" if quant_config else f"{prefix}.qkv", prefix=f"{prefix}.qkv_proj"
if isinstance(quant_config, compressed_tensors.CompressedTensorsConfig)
else f"{prefix}.qkv",
disable_tp=use_data_parallel, disable_tp=use_data_parallel,
) )
self.proj = RowParallelLinear( self.proj = RowParallelLinear(