xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 02:15:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Retrieve sliding_window from text config in Gemma3 MM (#25085)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor 2025-09-18 07:29:05 +01:00 committed by GitHub
parent 349e0e3462
commit f4cd80f944
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 2 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/models/gemma3_mm.py
View File

@ -688,7 +688,8 @@ class Gemma3ForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP,
global_attn_mask = torch.where(img_mask == 2, 0, global_attn_mask) global_attn_mask = torch.where(img_mask == 2, 0, global_attn_mask)
global_attn_masks.append(global_attn_mask) global_attn_masks.append(global_attn_mask)
if (sliding_window := self.config.sliding_window) is not None: sliding_window = self.config.text_config.sliding_window
if sliding_window is not None:
# Create a local causal mask with sliding window (1024). # Create a local causal mask with sliding window (1024).
local_attn_mask = torch.ones_like(global_attn_mask) local_attn_mask = torch.ones_like(global_attn_mask)
local_attn_mask = torch.tril(local_attn_mask, local_attn_mask = torch.tril(local_attn_mask,

3
vllm/model_executor/models/gemma3n_mm.py
View File

@ -461,9 +461,6 @@ class Gemma3nForConditionalGeneration(nn.Module, SupportsMultiModal,
self.multimodal_config = multimodal_config self.multimodal_config = multimodal_config
self.vocab_size = config.text_config.vocab_size self.vocab_size = config.text_config.vocab_size
self.sliding_window = getattr(config.text_config,
"interleaved_sliding_window", None)
self.vision_tower = AutoModel.from_config(config=config.vision_config) self.vision_tower = AutoModel.from_config(config=config.vision_config)
self.audio_tower = AutoModel.from_config(config=config.audio_config) self.audio_tower = AutoModel.from_config(config=config.audio_config)
self.embed_vision = Gemma3nMultimodalEmbedder(config.vision_config, self.embed_vision = Gemma3nMultimodalEmbedder(config.vision_config,