[Quant][Kernel] Remove GPTQ Group/Dynamic Activation Ordering (#54809)
Signed-off-by: Roderick-Wu <roderickwu2003@gmail.com> Signed-off-by: Roderick Wu <roderickwu2003@gmail.com> Signed-off-by: mgoin <mgoin64@gmail.com> Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
@@ -160,6 +160,17 @@ class TestMaybeRemapKvScaleName:
|
||||
assert result is None
|
||||
|
||||
|
||||
def test_checkpoint_weight_mapper_discards_serialized_g_idx():
|
||||
from vllm.model_executor.layers.quantization.base_config import (
|
||||
QuantizationConfig,
|
||||
)
|
||||
|
||||
mapper = QuantizationConfig.get_checkpoint_weight_mapper()
|
||||
|
||||
assert mapper._map_name("model.layers.0.g_idx") is None
|
||||
assert mapper._map_name("model.layers.0.qweight") == "model.layers.0.qweight"
|
||||
|
||||
|
||||
class TestKvCacheScaleMapper:
|
||||
"""The `WeightsMapper` returned by `get_cache_scale_mapper` replaces the
|
||||
per-model `maybe_remap_kv_scale_name` calls. It must remap the same set of
|
||||
|
||||
Reference in New Issue
Block a user