xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-13 20:25:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Fix] unwantted bias in InternLM Model (#740)

Browse Source
This commit is contained in:
WRH 2023-08-12 02:40:37 +08:00 committed by GitHub
parent 66c54aa9c3
commit 462ae5220a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/model_executor/models/internlm.py
View File

@ -7,15 +7,15 @@ from transformers import LlamaConfig
from vllm.model_executor.input_metadata import InputMetadata from vllm.model_executor.input_metadata import InputMetadata
from vllm.model_executor.layers.activation import SiluAndMul from vllm.model_executor.layers.activation import SiluAndMul
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.attention import PagedAttentionWithRoPE from vllm.model_executor.layers.attention import PagedAttentionWithRoPE
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.sampler import Sampler from vllm.model_executor.layers.sampler import Sampler
from vllm.model_executor.weight_utils import (hf_model_weights_iterator,
load_tensor_parallel_weights)
from vllm.model_executor.parallel_utils.parallel_state import ( from vllm.model_executor.parallel_utils.parallel_state import (
get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size) get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size)
from vllm.model_executor.parallel_utils.tensor_parallel import ( from vllm.model_executor.parallel_utils.tensor_parallel import (
VocabParallelEmbedding, ColumnParallelLinear, RowParallelLinear) ColumnParallelLinear, RowParallelLinear, VocabParallelEmbedding)
from vllm.model_executor.weight_utils import (hf_model_weights_iterator,
load_tensor_parallel_weights)
from vllm.sequence import SequenceOutputs from vllm.sequence import SequenceOutputs
KVCache = Tuple[torch.Tensor, torch.Tensor] KVCache = Tuple[torch.Tensor, torch.Tensor]
@ -32,12 +32,12 @@ class InternLMMLP(nn.Module):
super().__init__() super().__init__()
self.gate_up_proj = ColumnParallelLinear(hidden_size, self.gate_up_proj = ColumnParallelLinear(hidden_size,
2 * intermediate_size, 2 * intermediate_size,
bias=True, bias=False,
gather_output=False, gather_output=False,
perform_initialization=False) perform_initialization=False)
self.down_proj = RowParallelLinear(intermediate_size, self.down_proj = RowParallelLinear(intermediate_size,
hidden_size, hidden_size,
bias=True, bias=False,
input_is_parallel=True, input_is_parallel=True,
perform_initialization=False) perform_initialization=False)
if hidden_act != "silu": if hidden_act != "silu":