[Cleanup] Remove online quantization support in fp8.py in favor of online shorthands (#53585)

Signed-off-by: Felix Marty <Felix.Marty@amd.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
fxmarty-amd
2026-09-24 12:48:14 -07:00
committed by GitHub
co-authored by mgoin mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 58aa2f1ef1
commit 90a9515006
15 changed files with 240 additions and 269 deletions
+13 -1
View File
@@ -9,6 +9,12 @@ from transformers import PreTrainedConfig
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.quantization.fp8 import Fp8Config
from vllm.model_executor.layers.quantization.online.base import OnlineQuantizationConfig
from vllm.model_executor.layers.quantization.utils.quant_utils import (
kFp8Static128BlockSym,
kFp8StaticChannelSym,
kFp8StaticTensorSym,
)
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.transformers_utils.processors.internvl import (
InternVLImageProcessor,
@@ -71,7 +77,12 @@ class QianfanOCRForConditionalGeneration(InternVLChatModel):
) -> None:
super()._patch_quant_config(config, quant_config)
# ignore vit layers to preserve model performance
if isinstance(quant_config, Fp8Config):
if isinstance(quant_config, Fp8Config) or (
isinstance(quant_config, OnlineQuantizationConfig)
and quant_config.args.linear is not None
and quant_config.args.linear.weight
in (kFp8StaticTensorSym, kFp8Static128BlockSym, kFp8StaticChannelSym)
):
_FP8_IGNORED_LAYERS = [
*(
layer
@@ -87,6 +98,7 @@ class QianfanOCRForConditionalGeneration(InternVLChatModel):
"mlp1.1",
"mlp1.3",
]
quant_config.ignored_layers = list(quant_config.ignored_layers)
for layer in _FP8_IGNORED_LAYERS:
if layer not in quant_config.ignored_layers:
quant_config.ignored_layers.append(layer)