apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-09-29 13:52:26 +08:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

Simplify the Config code (#2770)

Browse Source 
* simplify the code

* fix vl

* delete config

* fix

* perfect code

* fix ci

* fix xpu

* fix xpu

* fix server

* resolve conflict

* fix mtp

* resolve conflict

* fix xpu

* fix xpu

* fix vl

* fix log

* fix qwen moe

* fix qwen moe

* fix qwen moe
This commit is contained in:
YuanRisheng
2025-07-14 19:50:05 +08:00
committed by GitHub
parent 2e81792d64
commit 4c7b8bc458
34 changed files with 551 additions and 911 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
fastdeploy/model_executor/layers/linear.py
View File

@@ -266,7 +266,7 @@ class ColumnParallelLinear(LinearBase):
with_bias=with_bias,
add_bias=add_bias,
skip_quant=skip_quant)
self.nranks = fd_config.parallel_config.tensor_parallel_degree
self.nranks = fd_config.parallel_config.tensor_parallel_size
self.input_size = input_size
self.output_size = divide(
output_size,
@@ -348,7 +348,7 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
"""
self.activation = activation
self.hidden_size = fd_config.model_config.hidden_size
self.nranks = fd_config.parallel_config.tensor_parallel_degree
self.nranks = fd_config.parallel_config.tensor_parallel_size
super().__init__(fd_config=fd_config,
prefix=prefix,
@@ -410,7 +410,7 @@ class QKVParallelLinear(ColumnParallelLinear):
self.kv_num_heads = fd_config.model_config.num_key_value_heads
self.hidden_size = fd_config.model_config.hidden_size
self.head_dim = fd_config.model_config.head_dim
self.nranks = fd_config.parallel_config.tensor_parallel_degree
self.nranks = fd_config.parallel_config.tensor_parallel_size
self.num_heads_per_rank = divide(self.num_heads, self.nranks)
if self.kv_num_heads < self.nranks and self.nranks % self.kv_num_heads == 0:
self.kv_num_heads_per_rank = 1
@@ -545,7 +545,7 @@ class RowParallelLinear(LinearBase):
skip_quant=skip_quant)
self.fd_config = fd_config
self.skip_quant = False
self.nranks = fd_config.parallel_config.tensor_parallel_degree
self.nranks = fd_config.parallel_config.tensor_parallel_size
self.hidden_size = fd_config.model_config.hidden_size
self.head_dim = fd_config.model_config.head_dim
self.num_heads = fd_config.model_config.num_attention_heads // self.nranks
@@ -638,7 +638,7 @@ class KVBatchLinear(LinearBase):
with_bias (bool): Whether to include bias or not. Defaults to False.
skip_quant (bool): Whether to skip quantization. Defaults to False.
"""
self.nranks = fd_config.parallel_config.tensor_parallel_degree
self.nranks = fd_config.parallel_config.tensor_parallel_size
self.kv_lora_rank = kv_lora_rank
self.num_attention_heads = num_attention_heads
self.qk_nope_head_dim = qk_nope_head_dim