[Quantization] Support online quantization with partially pre-quantized checkpoints (#51392)

Signed-off-by: Felix Marty <Felix.Marty@amd.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
fxmarty-amd
2026-09-04 11:46:06 -07:00
committed by GitHub
co-authored by Codex mgoin
parent a11dfcff89
commit 5690b02c03
25 changed files with 846 additions and 175 deletions
+5 -2
View File
@@ -69,7 +69,10 @@ from vllm.model_executor.layers.linear import (
RowParallelLinear,
UnquantizedLinearMethod,
)
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.quantization import (
QuantizationConfig,
resolve_quant_method,
)
from vllm.model_executor.layers.quantization.utils.quant_utils import (
get_and_maybe_dequant_weights,
)
@@ -462,7 +465,7 @@ class MultiHeadLatentAttention(nn.Module, AttentionLayerBase):
replace_parameter(self, "W_UK_T", W_UK.permute(1, 2, 0), prefer_copy=True)
quant_method = (
self.quant_config.get_quant_method(self, prefix=self.layer_name)
resolve_quant_method(self.quant_config, self, prefix=self.layer_name)
if self.quant_config
else None
)