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

27
vllm/model_executor/models/persimmon.py
View File

@@ -45,7 +45,8 @@ from vllm.sequence import IntermediateTensors
from .interfaces import 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 PersimmonMLP(nn.Module):
@@ -212,12 +213,13 @@ class PersimmonDecoderLayer(nn.Module):
@support_torch_compile
class PersimmonModel(nn.Module):
def __init__(self,
config: PersimmonConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = ""):
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
self.vocab_size = config.vocab_size
self.embed_tokens = VocabParallelEmbedding(config.vocab_size,
@@ -265,20 +267,13 @@ class PersimmonModel(nn.Module):
class PersimmonForCausalLM(nn.Module, SupportsPP):
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
self.config = config
self.vocab_size = config.vocab_size
self.model = PersimmonModel(config,
cache_config=cache_config,
quant_config=quant_config)
self.model = PersimmonModel(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
self.lm_head = ParallelLMHead(config.vocab_size,
config.hidden_size,
bias=False)