xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-06 01:25:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Pixtral-Large] Pixtral actually has no bias in vision-lang adapter (#10449)

Browse Source
This commit is contained in:
Patrick von Platen 2024-11-19 18:33:06 +01:00 committed by GitHub
parent f028dff33d
commit 11fd7ea639
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/models/pixtral.py
View File

@ -331,6 +331,7 @@ class VisionEncoderArgs:
num_attention_heads: int num_attention_heads: int
rope_theta: float # for rope-2D rope_theta: float # for rope-2D
image_token_id: int image_token_id: int
adapter_bias: bool = True
def _reshape_for_broadcast(freqs_cis: torch.Tensor, def _reshape_for_broadcast(freqs_cis: torch.Tensor,
@ -595,10 +596,10 @@ class VisionLanguageAdapter(nn.Module):
self.w_in = nn.Linear( self.w_in = nn.Linear(
args.hidden_size, args.hidden_size,
dim, dim,
bias=True, bias=args.adapter_bias,
) )
self.gelu = nn.GELU() self.gelu = nn.GELU()
self.w_out = nn.Linear(dim, dim, bias=True) self.w_out = nn.Linear(dim, dim, bias=args.adapter_bias)
def forward(self, x: torch.Tensor) -> torch.Tensor: def forward(self, x: torch.Tensor) -> torch.Tensor:
return self.w_out(self.gelu(self.w_in(x))) return self.w_out(self.gelu(self.w_in(x)))