[Bugfix][Quant] Read safetensors headers of checkpoints with other file names (#60411)
Signed-off-by: Stefan Koncarevic <Stefan.Koncarevic@amd.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com>
This commit is contained in:
co-authored by
Andreas Karatzas
parent
ee406adf0c
commit
814301b268
@@ -23,7 +23,10 @@ from vllm.model_executor.layers.quantization.auto_gptq import (
|
||||
AutoGPTQLinearMethod,
|
||||
AutoGPTQMoEMethod,
|
||||
)
|
||||
from vllm.model_executor.layers.quantization.utils.gptq_utils import override_config
|
||||
from vllm.model_executor.layers.quantization.utils.gptq_utils import (
|
||||
is_layer_gptq_quantized,
|
||||
override_config,
|
||||
)
|
||||
from vllm.platforms import current_platform
|
||||
|
||||
PROMPT = "On the surface of Mars, we found"
|
||||
@@ -71,6 +74,19 @@ def test_auto_gptq_config_get_name():
|
||||
assert AutoGPTQConfig.get_name() == "auto_gptq"
|
||||
|
||||
|
||||
def test_auto_gptq_quantizes_every_layer_when_the_module_list_is_unknown():
|
||||
"""An empty module list means the safetensors metadata could not be read,
|
||||
not that the checkpoint holds no quantized layer."""
|
||||
assert is_layer_gptq_quantized(
|
||||
prefix="model.layers.0.mlp.down_proj",
|
||||
quantized_layers=[],
|
||||
)
|
||||
assert not is_layer_gptq_quantized(
|
||||
prefix="model.layers.0.mlp.down_proj",
|
||||
quantized_layers=["self_attn.q_proj"],
|
||||
)
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"dynamic",
|
||||
[
|
||||
|
||||
Reference in New Issue
Block a user