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

Update rope_scaling to rope_parameters in preparation for Transformers v5 (#28542)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-11-19 18:06:36 +01:00
committed by GitHub
parent d44e9df7d4
commit a8b70304d6
104 changed files with 542 additions and 910 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/model_executor/models/baichuan.py
View File

@@ -136,7 +136,7 @@ class BaiChuanAttention(nn.Module):
hidden_size: int,
num_heads: int,
position_embedding: str,
rope_theta: float = 10000,
rope_parameters: dict,
max_position_embeddings: int = 8192,
cache_config: CacheConfig | None = None,
quant_config: QuantizationConfig | None = None,
@@ -150,7 +150,6 @@ class BaiChuanAttention(nn.Module):
self.num_heads = self.total_num_heads // tensor_model_parallel_world_size
self.head_dim = hidden_size // self.total_num_heads
self.position_embedding = position_embedding
self.rope_theta = rope_theta
self.max_position_embeddings = max_position_embeddings
# pylint: disable=invalid-name
@@ -192,7 +191,7 @@ class BaiChuanAttention(nn.Module):
self.head_dim,
rotary_dim=self.head_dim,
max_position=self.max_position_embeddings,
base=self.rope_theta,
rope_parameters=rope_parameters,
)
self.scaling = self.head_dim**-0.5
self.attn = Attention(
@@ -229,13 +228,12 @@ class BaiChuanDecoderLayer(nn.Module):
):
super().__init__()
self.hidden_size = config.hidden_size
rope_theta = getattr(config, "rope_theta", 10000)
max_position_embeddings = getattr(config, "max_position_embeddings", 8192)
self.self_attn = BaiChuanAttention(
hidden_size=self.hidden_size,
num_heads=config.num_attention_heads,
position_embedding=position_embedding,
rope_theta=rope_theta,
rope_parameters=config.rope_parameters,
max_position_embeddings=max_position_embeddings,
cache_config=cache_config,
quant_config=quant_config,