[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
@@ -24,6 +24,7 @@ from vllm.model_executor.layers.quantization.base_config import (
QuantizationConfig,
QuantizeMethodBase,
method_has_implemented_embedding,
resolve_quant_method,
)
from vllm.model_executor.layers.utils import dispatch_unquantized_gemm
from vllm.model_executor.parameter import BasevLLMParameter
@@ -291,7 +292,7 @@ class VocabParallelEmbedding(PluggableLayer):
# Avoid overriding a preselected model-specific method with generic
# config-based dispatch.
if quant_method is None and quant_config is not None:
quant_method = quant_config.get_quant_method(self, prefix=prefix)
quant_method = resolve_quant_method(quant_config, self, prefix=prefix)
if quant_method is None:
quant_method = UnquantizedEmbeddingMethod()