xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-22 16:15:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[CI/Build][CPU] Fix CPU CI by lazy importing triton FP8 kernels (#11618)

Browse Source 
Signed-off-by: jiang1.li <jiang1.li@intel.com>
This commit is contained in:
Li, Jiang 2024-12-30 18:17:04 +08:00 committed by GitHub
parent 970d6d0776
commit 5dbf854553
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/layers/quantization/fp8.py
View File

@ -15,8 +15,6 @@ from vllm.model_executor.layers.linear import (LinearBase, LinearMethodBase,
from vllm.model_executor.layers.quantization.base_config import ( from vllm.model_executor.layers.quantization.base_config import (
QuantizationConfig, QuantizeMethodBase) QuantizationConfig, QuantizeMethodBase)
from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod
from vllm.model_executor.layers.quantization.utils.fp8_utils import (
apply_w8a8_block_fp8_linear)
from vllm.model_executor.layers.quantization.utils.marlin_utils_fp8 import ( from vllm.model_executor.layers.quantization.utils.marlin_utils_fp8 import (
apply_fp8_marlin_linear, prepare_fp8_layer_for_marlin) apply_fp8_marlin_linear, prepare_fp8_layer_for_marlin)
from vllm.model_executor.layers.quantization.utils.quant_utils import ( from vllm.model_executor.layers.quantization.utils.quant_utils import (
@ -337,6 +335,9 @@ class Fp8LinearMethod(LinearMethodBase):
size_k=layer.input_size_per_partition, size_k=layer.input_size_per_partition,
bias=bias) bias=bias)
# Note: lazy import to avoid triton import error.
from vllm.model_executor.layers.quantization.utils.fp8_utils import (
apply_w8a8_block_fp8_linear)
if self.block_quant: if self.block_quant:
assert self.quant_config.weight_block_size is not None assert self.quant_config.weight_block_size is not None
return apply_w8a8_block_fp8_linear( return apply_w8a8_block_fp8_linear(