biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] Support Qwen2 embeddings and use tags to select model tests (#10184)

Browse Source
This commit is contained in:
Cyrus Leung
2024-11-15 12:23:09 +08:00
committed by GitHub
parent 2885ba0e24
commit b40cf6402e
19 changed files with 252 additions and 178 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
vllm/model_executor/models/qwen2_rm.py
View File

@@ -16,7 +16,7 @@ from vllm.model_executor.layers.pooler import Pooler, PoolingType
from vllm.model_executor.pooling_metadata import PoolingMetadata
from vllm.sequence import IntermediateTensors, PoolerOutput
from .interfaces import SupportsPP
from .interfaces import SupportsLoRA, SupportsPP
from .qwen2 import Qwen2Model
from .utils import AutoWeightsLoader, maybe_prefix
@@ -32,7 +32,7 @@ class ReLU(nn.Module):
return self.activation(input)
class Qwen2ForRewardModel(nn.Module, SupportsPP):
class Qwen2ForRewardModel(nn.Module, SupportsLoRA, SupportsPP):
packed_modules_mapping = {
"qkv_proj": [
"q_proj",
@@ -58,21 +58,9 @@ class Qwen2ForRewardModel(nn.Module, SupportsPP):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
config = vllm_config.model_config.hf_config
cache_config = vllm_config.cache_config
quant_config = vllm_config.quant_config
lora_config = vllm_config.lora_config
pooler_config = vllm_config.model_config.pooler_config
# TODO (@robertgshaw2): see if this can be moved out
if (cache_config.sliding_window is not None
and hasattr(config, "max_window_layers")):
raise ValueError("Sliding window for some but all layers is not "
"supported. This model uses sliding window "
"but `max_window_layers` = {} is less than "
"`num_hidden_layers` = {}. Please open an issue "
"to discuss this feature.".format(
config.max_window_layers,
config.num_hidden_layers,
))
self.config = config
self.lora_config = lora_config