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

[torch.compile] support all attention backends (#10558)

Browse Source 
Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao
2024-11-22 14:04:42 -08:00
committed by GitHub
parent db100c5cde
commit eebad39f26
77 changed files with 876 additions and 648 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -77,7 +77,8 @@ class StablelmAttention(nn.Module):
def __init__(self,
config: PretrainedConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None) -> None:
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "") -> None:
super().__init__()
self.config = config
self.hidden_size = config.hidden_size
@@ -131,7 +132,8 @@ class StablelmAttention(nn.Module):
self.scaling,
num_kv_heads=self.num_key_value_heads,
cache_config=cache_config,
quant_config=quant_config)
quant_config=quant_config,
prefix=f"{prefix}.attn")
def forward(
self,
@@ -155,9 +157,13 @@ class StablelmDecoderLayer(nn.Module):
config: PretrainedConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
) -> None:
super().__init__()
self.self_attn = StablelmAttention(config, cache_config, quant_config)
self.self_attn = StablelmAttention(config,
cache_config,
quant_config,
prefix=f"{prefix}.self_attn")
self.mlp = StablelmMLP(config, quant_config)
norm_eps = getattr(config, "norm_eps",
getattr(config, "layer_norm_eps", 1e-05))
@@ -207,8 +213,8 @@ class StableLMEpochModel(nn.Module):
)
self.start_layer, self.end_layer, self.layers = make_layers(
config.num_hidden_layers,
lambda prefix: StablelmDecoderLayer(config, cache_config,
quant_config),
lambda prefix: StablelmDecoderLayer(
config, cache_config, quant_config, prefix=prefix),
prefix=f"{prefix}.layers",
)
norm_eps = getattr(config, "norm_eps",