[Model] Add base class for LoRA-supported models (#5018)

This commit is contained in:
Cyrus Leung
2024-06-27 16:03:04 +08:00
committed by GitHub
parent d12af207d2
commit 96354d6a29
20 changed files with 270 additions and 75 deletions
+9 -2
View File
@@ -45,6 +45,8 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import SamplerOutput
from .interfaces import SupportsLoRA
class XverseMLP(nn.Module):
@@ -266,7 +268,9 @@ class XverseModel(nn.Module):
return hidden_states
class XverseForCausalLM(nn.Module):
class XverseForCausalLM(nn.Module, SupportsLoRA):
supports_lora = True
packed_modules_mapping = {
"qkv_proj": [
"q_proj",
@@ -299,10 +303,13 @@ class XverseForCausalLM(nn.Module):
config: PretrainedConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
lora_config=None,
lora_config: Optional[LoRAConfig] = None,
) -> None:
super().__init__()
self.config = config
self.lora_config = lora_config
self.quant_config = quant_config
self.model = XverseModel(config, cache_config, quant_config)
self.lm_head = ParallelLMHead(config.vocab_size, config.hidden_size)