apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[Feature] Add an unquantized option for MoE and Dense quant type (#4813)

Browse Source
This commit is contained in:
Sunny-bot1
2025-11-19 16:24:03 +08:00
committed by GitHub
parent 9ff418db73
commit 43f0c7557e
3 changed files with 48 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/model_executor/layers/linear.py
+7 -5
View File
@@ -143,11 +143,13 @@ class LinearBase(nn.Layer):
self.with_bias = with_bias
self.add_bias = add_bias
self.prefix = prefix
self.is_quantized = fd_config.model_config.is_quantized
self.is_quantized = fd_config.model_config.is_quantized and not (
fd_config.quant_config.name() == "mix_quant" and fd_config.quant_config.dense_quant_type is None
)
# key
if weight_key:
self.weight_key = f"{prefix}.{weight_key}"
elif fd_config.model_config.is_quantized and not skip_quant:
elif self.is_quantized and not skip_quant:
self.weight_key = f"{prefix}.quant_weight"
self.weight_scale_key = f"{prefix}.weight_scale"
self.act_scale_key = f"{prefix}.activation_scale"
@@ -170,7 +172,7 @@ class LinearBase(nn.Layer):
self.output_size,
]
if fd_config.quant_config and not skip_quant:
if fd_config.quant_config and not skip_quant and fd_config.quant_config.get_quant_method(self):
self.quant_method = fd_config.quant_config.get_quant_method(self)
else:
self.quant_method: Optional[QuantMethodBase] = UnquantizedLinearMethod()
@@ -232,7 +234,7 @@ class LinearBase(nn.Layer):
# weight
self.state_dict = state_dict
assert self.weight_key is not None, "weight_key should not be None."
if self.fd_config.model_config.is_quantized:
if self.is_quantized:
self.load_prequant_weight(state_dict)
else:
self.load_weight(state_dict)
@@ -784,7 +786,7 @@ class QKVParallelLinear(ColumnParallelLinear):
assert self.weight_key is not None, "weight_key should not be None."
# qkv fused in disk
if self.fd_config.model_config.is_quantized:
if self.is_quantized:
self.load_prequant_weight(state_dict)
else:
self.load_weight(state_dict)