[Hardware][ROCM] using current_platform.is_rocm (#9642)
Signed-off-by: wangshuai09 <391746016@qq.com>
This commit is contained in:
@@ -49,9 +49,9 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
|
||||
from vllm.model_executor.model_loader.weight_utils import (
|
||||
default_weight_loader, kv_cache_scales_loader, maybe_remap_kv_scale_name)
|
||||
from vllm.model_executor.sampling_metadata import SamplingMetadata
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import IntermediateTensors
|
||||
from vllm.transformers_utils.configs.exaone import ExaoneConfig
|
||||
from vllm.utils import is_hip
|
||||
|
||||
from .interfaces import SupportsLoRA, SupportsPP
|
||||
from .utils import (PPMissingLayer, is_pp_missing_parameter,
|
||||
@@ -595,7 +595,7 @@ class ExaoneForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
|
||||
if not isinstance(self.transformer.h[layer_idx], nn.Identity):
|
||||
layer_self_attn = self.transformer.h[layer_idx].attn
|
||||
|
||||
if is_hip():
|
||||
if current_platform.is_rocm():
|
||||
# The scaling factor convention we are assuming is
|
||||
# quantized_value * scaling_factor ~= true_value
|
||||
# which is consistent with the practice of setting
|
||||
|
||||
@@ -49,8 +49,8 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
|
||||
from vllm.model_executor.model_loader.weight_utils import (
|
||||
default_weight_loader, kv_cache_scales_loader, maybe_remap_kv_scale_name)
|
||||
from vllm.model_executor.sampling_metadata import SamplingMetadata
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import IntermediateTensors
|
||||
from vllm.utils import is_hip
|
||||
|
||||
from .interfaces import SupportsLoRA, SupportsPP
|
||||
from .utils import PPMissingLayer, is_pp_missing_parameter, make_layers
|
||||
@@ -534,7 +534,7 @@ class GraniteForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
|
||||
if not isinstance(self.model.layers[layer_idx], nn.Identity):
|
||||
layer_self_attn = self.model.layers[layer_idx].self_attn
|
||||
|
||||
if is_hip():
|
||||
if current_platform.is_rocm():
|
||||
# The scaling factor convention we are assuming is
|
||||
# quantized_value * scaling_factor ~= true_value
|
||||
# which is consistent with the practice of setting
|
||||
|
||||
@@ -50,8 +50,8 @@ from vllm.model_executor.model_loader.weight_utils import (
|
||||
default_weight_loader, kv_cache_scales_loader, maybe_remap_kv_scale_name)
|
||||
from vllm.model_executor.pooling_metadata import PoolingMetadata
|
||||
from vllm.model_executor.sampling_metadata import SamplingMetadata
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import IntermediateTensors, PoolerOutput
|
||||
from vllm.utils import is_hip
|
||||
|
||||
from .interfaces import SupportsLoRA, SupportsPP
|
||||
from .utils import (AutoWeightsLoader, PPMissingLayer, is_pp_missing_parameter,
|
||||
@@ -423,7 +423,7 @@ class LlamaModel(nn.Module):
|
||||
if not isinstance(self.layers[layer_idx], nn.Identity):
|
||||
layer_self_attn = self.layers[layer_idx].self_attn
|
||||
|
||||
if is_hip():
|
||||
if current_platform.is_rocm():
|
||||
# The scaling factor convention we are assuming is
|
||||
# quantized_value * scaling_factor ~= true_value
|
||||
# which is consistent with the practice of setting
|
||||
|
||||
@@ -12,7 +12,7 @@ import cloudpickle
|
||||
import torch.nn as nn
|
||||
|
||||
from vllm.logger import init_logger
|
||||
from vllm.utils import is_hip
|
||||
from vllm.platforms import current_platform
|
||||
|
||||
from .interfaces import (has_inner_state, is_attention_free,
|
||||
supports_multimodal, supports_pp)
|
||||
@@ -247,7 +247,7 @@ def _try_load_model_cls(
|
||||
model_arch: str,
|
||||
model: _BaseRegisteredModel,
|
||||
) -> Optional[Type[nn.Module]]:
|
||||
if is_hip():
|
||||
if current_platform.is_rocm():
|
||||
if model_arch in _ROCM_UNSUPPORTED_MODELS:
|
||||
raise ValueError(f"Model architecture '{model_arch}' is not "
|
||||
"supported by ROCm for now.")
|
||||
|
||||
@@ -49,8 +49,8 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
|
||||
from vllm.model_executor.model_loader.weight_utils import (
|
||||
default_weight_loader, kv_cache_scales_loader, maybe_remap_kv_scale_name)
|
||||
from vllm.model_executor.sampling_metadata import SamplingMetadata
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import IntermediateTensors
|
||||
from vllm.utils import is_hip
|
||||
|
||||
from .interfaces import SupportsLoRA, SupportsPP
|
||||
from .utils import (PPMissingLayer, is_pp_missing_parameter,
|
||||
@@ -558,7 +558,7 @@ class SolarForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
|
||||
if not isinstance(self.model.layers[layer_idx], nn.Identity):
|
||||
layer_self_attn = self.model.layers[layer_idx].self_attn
|
||||
|
||||
if is_hip():
|
||||
if current_platform.is_rocm():
|
||||
# The scaling factor convention we are assuming is
|
||||
# quantized_value * scaling_factor ~= true_value
|
||||
# which is consistent with the practice of setting
|
||||
|
||||
Reference in New Issue
Block a user