[5/N] pass the whole config to model (#9983)

Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao
2024-11-09 14:17:28 +08:00
committed by GitHub
parent 49d2a41a86
commit 1a95f10ee7
75 changed files with 583 additions and 654 deletions
+6 -6
View File
@@ -25,7 +25,7 @@ from transformers import BartConfig
from transformers.utils import logging
from vllm.attention import Attention, AttentionMetadata, AttentionType
from vllm.config import CacheConfig, LoRAConfig
from vllm.config import CacheConfig, LoRAConfig, VllmConfig
from vllm.distributed import get_tensor_model_parallel_world_size
from vllm.model_executor.layers.activation import get_act_fn
from vllm.model_executor.layers.linear import (ColumnParallelLinear,
@@ -810,13 +810,13 @@ class BartModel(nn.Module):
class BartForConditionalGeneration(nn.Module):
base_model_prefix = "model"
def __init__(self,
config: BartConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
lora_config: Optional[LoRAConfig] = None):
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
# currently all existing BART models have `tie_word_embeddings` enabled
assert config.tie_word_embeddings
self.config = config