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

[6/N] pass whole config to inner model (#10205)

Browse Source 
Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao
2024-11-10 22:41:46 -08:00
committed by GitHub
parent f0f2e5638e
commit f89d18ff74
69 changed files with 681 additions and 963 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
vllm/model_executor/models/minicpm.py
View File

@@ -29,7 +29,7 @@ from transformers import PretrainedConfig
from vllm.attention import Attention, AttentionMetadata
from vllm.compilation.decorators import support_torch_compile
from vllm.config import CacheConfig, LoRAConfig, VllmConfig
from vllm.config import CacheConfig, VllmConfig
from vllm.distributed import (get_pp_group, get_tensor_model_parallel_rank,
get_tensor_model_parallel_world_size,
tensor_model_parallel_all_reduce)
@@ -53,7 +53,8 @@ from vllm.sequence import IntermediateTensors
from .interfaces import SupportsLoRA, SupportsPP
from .utils import (is_pp_missing_parameter,
make_empty_intermediate_tensors_factory, make_layers)
make_empty_intermediate_tensors_factory, make_layers,
maybe_prefix)
class MiniCPMMoE(nn.Module):
@@ -351,15 +352,14 @@ class MiniCPMDecoderLayer(nn.Module):
@support_torch_compile
class MiniCPMModel(nn.Module):
def __init__(
self,
config: PretrainedConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
lora_config: Optional[LoRAConfig] = None,
prefix: str = "",
) -> 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
self.config = config
self.cache_config = cache_config
self.quant_config = quant_config
@@ -461,24 +461,22 @@ class MiniCPMForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
}
embedding_padding_modules = ["lm_head"]
def __init__(
self,
vllm_config: VllmConfig,
prefix: str = "",
) -> 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
self.prefix = prefix
self.vllm_config = vllm_config
self.config = config
self.lora_config = lora_config
self.cache_config = cache_config
self.quant_config = quant_config
self.num_experts = getattr(self.config, "num_experts", 0)
self._init_model()
self._init_model(vllm_config=vllm_config, prefix=prefix)
unpadded_vocab_size = config.vocab_size
if lora_config:
unpadded_vocab_size += lora_config.lora_extra_vocab_size
@@ -502,11 +500,9 @@ class MiniCPMForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
self.make_empty_intermediate_tensors = (
self.model.make_empty_intermediate_tensors)
def _init_model(self):
self.model = MiniCPMModel(config=self.config,
cache_config=self.cache_config,
quant_config=self.quant_config,
lora_config=self.lora_config)
def _init_model(self, *, vllm_config: VllmConfig, prefix: str = ""):
self.model = MiniCPMModel(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
def forward(
self,