[Quantization] add humming moe backend to all dense/moe oracles (#41652)
Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com> Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
@@ -764,7 +764,7 @@ def _convert_sublayer_to_humming(
|
||||
Returns:
|
||||
Tuple of (converted_weight_schema, converted_input_schema)
|
||||
"""
|
||||
from humming.schema import HummingWeightSchema
|
||||
from vllm.utils.humming import HummingWeightSchema
|
||||
|
||||
if isinstance(weight_schema, HummingWeightSchema):
|
||||
# Already in Humming format
|
||||
@@ -814,7 +814,7 @@ def _prepare_and_transform_sublayer(
|
||||
|
||||
This calls Humming's prepare_layer_meta and transform_humming_layer.
|
||||
"""
|
||||
from humming.layer import HummingMethod
|
||||
from vllm.utils.humming import HummingMethod
|
||||
|
||||
HummingMethod.prepare_layer_meta(
|
||||
layer=layer,
|
||||
@@ -866,7 +866,7 @@ def _process_single_sublayer(
|
||||
Returns:
|
||||
Tuple of (final_weight_schema, final_input_schema)
|
||||
"""
|
||||
from humming.schema import HummingWeightSchema
|
||||
from vllm.utils.humming import HummingWeightSchema
|
||||
|
||||
# Step 1: Convert from checkpoint format to humming format if needed
|
||||
current_weight_schema, current_input_schema = _convert_sublayer_to_humming(
|
||||
@@ -958,12 +958,10 @@ def convert_to_humming_moe_kernel_format(
|
||||
"Must provide either weight_schema/input_schema or quant_config"
|
||||
)
|
||||
|
||||
from humming.layer import HummingInputSchema
|
||||
from humming.schema import BaseWeightSchema
|
||||
|
||||
from vllm.model_executor.layers.quantization.utils.humming_utils import (
|
||||
humming_is_layer_skipped,
|
||||
)
|
||||
from vllm.utils.humming import BaseWeightSchema, HummingInputSchema
|
||||
|
||||
if weight_schema is None:
|
||||
weight_schema = BaseWeightSchema.from_config(quant_config)
|
||||
|
||||
Reference in New Issue
Block a user